WORKING PAPER SERIES
Bank Specialization and the Design of Loan Contracts
Marco Giometti
Universidad Carlos III de Madrid
Stefano Pietrosanti
Financial Stability Directorate, Bank of Italy
November 2022
FDIC CFR WP 2022-14
fdic.gov/cfr
The Center for Financial Research (CFR) Working Paper Series allows CFR staff and their coauthors to circulate
preliminary research findings to stimulate discussion and critical comment. Views and opinions expressed in CFR Working
Papers reflect those of the authors and do not necessarily reflect those of the FDIC or the United States. Comments and
suggestions are welcome and should be directed to the authors. References should cite this research as a “FDIC CFR
Working Paper” and should note that findings and conclusions in working papers may be preliminary and subject to
revision.
Bank Specialization and the Design of Loan Contracts
Marco Giometti
†
and Stefano Pietrosanti
‡
†
Universidad Carlos III de Madrid
‡
Financial Stability Directorate, Bank of Italy
First version: December 2019
This version: November 2022
Abstract. We study bank specialization in lending in the U.S. corporate loan market. We
document that banks specialize in lending to specific industries. Specialization is persistent
over time and common across industries. Using detailed information on syndicated loans,
we show that the typical loan contract between a bank specialized in an industry and a
firm in the same industry has less restrictive financial covenants and no higher spreads.
These results are not explained by relationship lending, high industry market shares, or
geographical proximity, and are robust to using default shocks on lenders’ loan portfolios as
a source of variation in banks’ self-assessment of screening abilities. Overall, our evidence
suggests that banks specialize in lending because of information advantages in monitoring
specific industries. Furthermore, the laxer contract terms offered by specialized banks could
provide an explanation for recent evidence that firms cannot easily substitute credit granted
by specialized banks.
Keywords: Bank Specialization, Security design, Covenants, Monitoring, Screening.
JEL Classification: L15, L22, G21, G30, G32.
We are grateful to David Musto, Guillermo Ordóñez, and Michael Roberts for their advice and feedback. We also thank Edoardo Acabbi,
Mitchell Berlin, Emilia Bonaccorsi di Patti, Elena Carletti, Alessandro Dovis, Marc Flandreau, Erik Gilje, Itay Goldstein, Richard Herring, Lorena
Keller, Tong Liu, David Martinez-Miera, Christian Opp, Francesco Palazzo, Dominik Supera, and Petra Todd for their feedback and comments.
We thank participants in the Philadelphia Fed and Wharton brown-bag seminars, the reading group on Financial Intermediation at Bocconi
University, the 20th FDIC Annual Bank Research Conference, the VPDE 14th PhD Workshop in Economics, the 2022 SGF Conference, the 2022
FMCG Conference, the 2022 FMA European Conference, the Summer Workshop on Money, Banking, Payments, and Finance, the 8th Bank of
Italy Annual Banking Research Network Workshop, and the MadBar Workshop on Corporate Finance and Banking for their comments. We
thank Juan Gorostiaga, Marco Pelosi, Olga Briukhova, Lorenzo Schönleber, and Carlos Ramirez for the insightful discussions. All errors are our
own. The opinions expressed do not necessarily reflect those of the Bank of Italy, the Eurosystem, and their staff.
† Marco Giometti: [email protected]
‡ Stefano Pietrosanti: [email protected]
1. Introduction
Diversification of risk plays a central role in many theories of financial intermediation (e.g.
Boyd & Prescott, 1986; Diamond, 1984). However, empirical evidence shows that banks often
concentrate their lending across multiple dimensions, including geography, scale, and industry.
1
There has been extensive work showing how portfolio concentration can have important
implications for banks’ economic performance and risk, as well as for their borrowers via the
transmission of shocks through the banking sector.
2
What is less well understood are the implications of bank specialization for security design.
In particular, there is no or little evidence on the role of specialization in lending for loan
contract terms, such as covenants or loan spreads.
3
We believe it is important to fill this gap
for three reasons. First, contracts reflect the preferences of and the information available to
the contracting parties and, as such, they provide insight into the objectives of those parties.
They might inform a better understanding of the lending advantages associated with bank
specialization and, ultimately, of how the structure of credit markets interacts with financial
contracting. Second, clauses allocating control rights within the lending relationship – covenants
– represent a key channel for the transmission of financial shocks to the real economy.
4
Third,
contract terms make differences in funding sources explicit, thereby helping identify the degree
to which bank loans are substitutable across different lenders.
The goal of this paper is to address the question of how specialization in bank lending affects
the design of loan contracts, in the context of the $2 trillion, corporate syndicated loan market.
5
1.
Berger and DeYoung (2001); Carey, Post, and Sharpe (1998); Hughes, Lang, Mester, and Moon (1996);
Paravisini, Rappoport, and Schnabl (2022).
2.
On the relation between bank portfolio concentration and related performances, see Acharya, Hasan, and
Saunders (2006); Beck, De Jonghe, and Mulier (2022); Boeve, Duellmann, and Pfingsten (2010); Hayden, Porath,
and Westernhagen (2007); Tabak, Fazio, and Cajueiro (2011). On the real effects of bank specialization, see
De Jonghe, Dewachter, Mulier, Ongena, and Schepens (2020); Gopal (2021); Paravisini et al. (2022); Schwert
(2018).
3.
Two exceptions are Daniels and Ramirez (2008), who document that banks specialize in lending towards large
firms and non-banks towards small firms, with banks demanding a lower loan spread, and Blickle, Parlatore, and
Saunders (2021), who show that specialized banks offer lower loan spreads, longer maturity, and greater amounts
using confidential FR Y-14 data from 2012 to 2020.
4.
Specifically, covenant violations imply a shift in firm control rights from the borrower to the lender, which
matters for firm investment (Chava & Roberts, 2008) and the transmission of bank-level shocks to firms’ credit
availability (Chodorow-Reich & Falato, 2022).
5. U.S. Syndicated Lending Topples Records in 2017, Reuters, December 2017.
1
First, we document that the average bank’s loan portfolio has a higher industry concentration
than the market; bank specialization is common across industries and persistent over time.
Then, we show that loan contracts display less restrictive covenants when the borrower belongs
to an industry in which the bank is specialized, with no higher spreads or fees. To interpret
this finding, we build on theoretical works such as Gârleanu and Zwiebel (2009), which argues
that covenant strictness reflects the degree of information frictions between borrowers and
lenders. In this sense, we suggest that the evidence we bring supports an explanation of bank
specialization based on information advantages in screening and monitoring specific types of
projects.
In order to perform our analysis, we obtain data on the syndicated loans from LPC DealScan,
and we merge it with Compustat. The resulting dataset is a loan-level panel with bank, firm,
and loan characteristics, from 1996 to 2016.
6
We use this data to estimate the degree of
diversification of bank loan portfolios. We then analyze the extent to which banks specialize
their lending towards different sectors adapting the approach in Paravisini et al. (2022) to our
setting. A bank is defined as specialized in a sector if it has an abnormally large portfolio share
of loans in a sector, relative to other banks. Intuitively, this measure captures the extent to
which corporate lending on banks’ balance sheets deviates from a value-weighted portfolio. In
doing so, the measure accounts for heterogeneity in the size of sectors in the economy and in
the size of bank sectoral lending relative to the bank’s overall corporate lending.
We find clear evidence of bank specialization. First, we show that the average bank displays
more concentration in lending than what would be implied by the overall distribution of credit
in the market. Second, we document that certain banks specialize in lending by holding a
disproportionately large share of loans in certain sectors. In particular, each sector consistently
displays at least one specialized bank. Furthermore, specialization is persistent: a bank that is
specialized in a given year has a 25% probability of being specialized 10 years after.
We then explore the implications of bank specialization for the design of loan contracts.
In particular, we focus on the allocation of control rights and cash flow rights between the
lender and the borrower. To proxy for the degree of ex-ante control rights allocated to the
6.
We choose this sample period because coverage of the syndicated loan market sharply improves in DealScan
after 1995 (Chava & Roberts, 2008).
2
lender, we employ the measure of covenant strictness developed by Demerjian and Owens
(2016).
7
Intuitively, this measure captures the ex-ante probability of violating at least one of
the covenants embedded in the contract. For cash flow rights, we use the All-In Drawn Spread
(AISD).
8
Looking at both is important as these contract terms are jointly determined, and there
might be a trade-off between them (Bradley & Roberts, 2015).
We find that the average loan contract between a bank specialized in a sector and a borrower
from that sector includes covenants that are 24 percentage points less restrictive and an all-in-
drawn spread that is 30 basis points lower—compared to a loan contract granted by the same
bank, in the same year, to a firm in another sector. The observed effects are economically and
statistically significant. For covenant strictness it amounts to 60% of the empirical standard
deviation; for the AISD, it amounts to 25% of the empirical standard deviation.
Comparing loans made by the same bank in the same year rules out the argument that our
finding is driven by unobserved, time-varying lender heterogeneity. However, the observed
variation in contracts might be simply driven by specialized banks matching systematically with
different firms. We take several steps to mitigate this concern. First, we control for observable
proxies of borrower risk, such as expected default probability, size, leverage, liquidity, ability to
provide collateral, profitability, and age. Second, we restrict our analysis to firms that borrow
from more than one bank over the duration of our sample, employing a within-firm approach.
Third, we restrict our comparison to loans made to firms that have the same credit rating. The
main finding does not change: the average loan contract between a bank specialized in a sector
and a borrower from that sector includes a covenant structure that is less restrictive, and it does
not display higher spreads.
We then ask whether our findings can improve our understanding of the lending advantages
associated with bank specialization. Theory suggests that the degree of allocation of ex-ante
control rights to the lender should be directly proportional to the level of asymmetric information
that exists between a borrower and a lender over potential future transfers from debt to equity
(Gârleanu & Zwiebel, 2009). In this view, the strictness of the covenant structure embedded in
a loan contract captures the information distance between a borrower and a lender. Therefore,
7. This measure is similar to the one developed by Murfin (2012).
8. The AISD is a fee paid over the base rate (usually LIBOR) for each dollar of credit drawn.
3
a plausible interpretation of our results implies the existence of an industry-specific information
advantage for banks specializing their lending towards a specific industry. The fact that a less
restrictive covenant structure is not compensated by a higher spread provides further support
to this interpretation.
We rule out a number of alternative explanations for our findings. First, we show that
specialization in lending toward an industry does not simply reflect a pattern of relationship
lending with borrowers in that industry. While it is indeed true that the longer the relationship
with a given borrower the lower the cost of credit – consistent with the empirical results of
Bharath, Dahiya, Saunders, and Srinivasan (2011) and Schenone (2010) – this appears to be
uncorrelated with bank specialization. Moreover, we do not find any linear effect between
relationship lending and covenant strictness.
9
Second, our results are not driven by geographical
specialization, which confirms the notion of an industry-specific information advantage. This
is consistent with the recent evidence provided by Di and Pattison (2022) and Duquerroy,
Mazet-Sonilhac, Mésonnier, and Paravisini (2022) for small business lending.
Third, specialized banks might have a high market share in an industry. Recent work by
Giannetti and Saidi (2019) suggests that lenders with high market shares in an industry have a
high propensity to internalize the spillovers of their credit decision. This might involve writing
less strict contracts to avoid triggering potentially costly defaults or renegotiations and could
represent a different economic mechanism that would explain our results.
10
We show that this
is not the case: Larger industry market shares do not have an effect on covenant strictness
and are associated with higher spreads, possibly implying a higher bargaining power in the
contracting process. These findings also alleviate the possible concern that specialized banks
might offer favorable contract terms to capture a higher industry market share and not as a
result of monitoring advantages.
Finally, to further validate our interpretation, we use defaults on lenders’ loan portfolios as a
plausible source of exogenous variation in lenders’ perception of their own screening ability
(Murfin, 2012). We look at the extent to which defaults of firms in each bank’s loan portfolio
9.
Prilmeier (2017) finds a non-linear, quadratic relationship between the intensity of the credit relationship and
covenant strictness.
10.
There is a large literature documenting the negative consequences of debt covenant violations on investment,
employment, and other firm-level outcomes. See Chava and Roberts (2008) and Chodorow-Reich (2014).
4
affects the covenant strictness of subsequent loans arranged by a particular bank. We show that
a bank is more sensitive to the default of a firm whenever such a firm belongs to a sector in
which the bank is specialized, as it is expected under an interpretation of specialization patterns
as stemming from information advantage.
With this paper, we contribute to the growing literature on different forms of specialization in
the credit market. A first strand of papers, such as Carey et al. (1998) and Daniels and Ramirez
(2008), studies how different kinds of intermediaries specialize in lending to different types of
firms and the consequences of such choices. Black, Krainer, and Nichols (2020) show that in the
commercial real estate mortgage market banks systematically fund riskier collateral compared
to arm’s length investors. Liberti, Sturgess, and Sutherland (2017) and Gopal (2021) document
how different lenders focus on demanding specific types of collateral to back the loans they
provide, and how this matters for lending decisions in new markets and in the presence of
lender constraints. Acharya et al. (2006), Beck et al. (2022), and Tabak et al. (2011) focus
instead on the overall risk effect of banks and financial intermediaries’ portfolio specialization,
finding null or negative effects.
A further strand, closer to our paper, focuses on banks only. Blickle et al. (2021), Di and
Pattison (2022), Duquerroy et al. (2022), De Jonghe et al. (2020), Jiang and Li (2022), and
Paravisini et al. (2022) show that even within a single class of intermediaries, i.e. banks, there
is specialization in lending towards specific firms, with positive effects on credit supply. Among
these, Jiang and Li (2022) and Paravisini et al. (2022) focus on the heterogeneity in credit
supply responses to shocks by specialized and non-specialized banks. Both papers document
the presence of a comparative advantage in lending, respectively towards specific industries
and specific export markets. Overall, they suggest a degree of non-substitutability between
credit provided by specialized banks and non-specialized banks.
To the best of our knowledge, we are the first to look at the implications of lender spe-
cialization for financial contracting and security design. Similarly to Blickle et al. (2021),
we show that industry specialization in lending drives loan spreads. Uniquely, we document
that specialization is a key determinant of loan covenants too, which represent a key channel
determining the impact of firm-level and aggregate shocks on investment, as shown by Chava
and Roberts (2008) and Chodorow-Reich and Falato (2022) respectively. Furthermore, by
5
leveraging detailed contract information and focusing on covenant strictness, we present a
possible explanation for why credit obtained from specialized lenders is difficult to substitute:
specialized lenders offer overall more generous terms that non-specialized lenders might not
be willing to offer. Finally, we provide an alternative test to the one proposed by Paravisini
et al. (2022) to identify the presence of information advantages in lending associated with
specialization, based on a measure of information distance between borrowers and lenders.
Our findings regarding the importance of bank specialization for contract features also
contribute more generally to the study of financial contracting and its determinants. Several
works highlight the role of borrower or lender characteristics for the determination of loan
covenants (e.g. Abuzov, Herpfer, & Steri, 2020; Berlin & Mester, 1992; Billett, King, & Mauer,
2007; Bradley & Roberts, 2015; Demerjian, Owens, & Sokolowski, 2018; Demiroglu & James,
2010; Murfin, 2012), or pricing (e.g. Cai, Eidam, Saunders, & Steffen, 2018; Ivashina, 2009).
Closer to our paper, a smaller set of studies stresses the importance of jointly taking into
account borrowers’ and lenders’ characteristics when looking at the determinants of contract
features. Prilmeier (2017) shows that relationship lending affects the design of financial
covenants. Gorostiaga (2022) shows that banks with higher market shares in an industry offer
loans with stricter covenants to firms in that industry, supporting an explanation based on banks’
incentives to limit competition among similar customer firms.
11
Hubbard, Kuttner, and Palia
(2002) and Santos and Winton (2019) document that the interaction of bank capital and firm
profitability matters for the determination of loan spreads. Bao (2019) finds that peer effects in
loan portfolios affect the cost of credit. With respect to these studies, we provide an additional
joint dimension—lender’s industry specialization and borrower’s industry—that is relevant for
the determination of price and non-price terms.
The paper proceeds as follows. In Section 2 we describe the sample construction, discuss
how we measure specialization, and provide evidence on bank specialization in the syndicated
loan market. In Section 3 we present our empirical strategy, our findings, and discuss several
alternative explanations. In Section 4 we provide concluding remarks.
11.
Note that we find the opposite effect for banks’ portfolio shares. In Gorostiaga (2022) total credit to the
industry is the denominator, in our work, total credit from the bank. The two results are not at all in contrast.
6
2. Data and Measurement
To characterize specialization and to study its implications, we construct a sample of syn-
dicated loans matched with bank and firm characteristics. Below we describe the sample
construction, introduce and discuss the way we measure bank specialization, present the other
economic variables we employ in our analysis, and summarize the sample characteristics.
2.1. Sample Construction
Our two main sources of data for this paper are LPC DealScan and Compustat. LPC DealScan
contains detailed information on syndicated loans, including loan amounts, covenants, pricing,
and maturity. Compustat provides balance-sheet information for both banks and firms. We
merge the loan data with borrower characteristics thanks to the linking table provided by Chava
and Roberts (2008), which matches firms in Compustat to borrowers in DealScan from 1987
to 2017.
12
We also merge firm characteristics in Compustat with the industry classification
developed by Hoberg and Phillips (2010, 2016), which is available for most public companies
present in Compustat starting from 1987.
We obtain information on banks by matching lenders in DealScan with bank characteristics,
thanks to the linking table provided by Schwert (2018), which identifies the Bank Holding
Company (BHC) of all DealScan lenders with at least 50 loans, or $10 billion loan volume in
the matched DealScan-Compustat sample. We define a bank to be the BHC, not the individual
DealScan lender. As most loans in DealScan are syndicated, the same loans will be associated
with one or more lead arrangers, and several other participant banks. Consistently with other
studies, we focus only on the lead arranger(s), and we attribute the whole loan amount to the
lead arranger(s) of the syndicate.
13
This choice stems from observing that a lead arranger is
the bank in charge of the active management of the loan, even if it does not retain the entirety
of its amount on their balance sheets (Ivashina, 2009).
14
We identify a lead arranger following
12.
The linking table is constantly being updated; as of April 2022 this is the most recent and comprehensive
version.
13. See, for example, Chakraborty, Goldstein, and MacKinlay (2018); Prilmeier (2017); Schwert (2018).
14. If there are multiple lead arrangers, we split the loan amount equally among them.
7
the procedure outlined in Chakraborty et al. (2018).
15
We restrict the sample to loans originated between 1996 to 2016 since the coverage of the
syndicated lending activity and contract terms in Dealscan is sparse before 1996 (Chava &
Roberts, 2008). We further restrict the sample to loans that have borrowers headquartered in
the US. We also drop from our sample all loans to financial corporations (Compustat SIC codes
from 6000 to 6999).
16
All variables, except the measures of covenant strictness and of expected
default probability that are naturally bounded between 0 and 100, are winsorized at the 1st
and 99th percentile.
The unit of observation in DealScan is a loan facility. However, information on loan covenants
is available only at the package, or deal, level. Since in our analysis the main dependent variable
is covenant strictness, we conduct our analysis at the package level, aggregating facility-level
information by weighting the facility characteristics – such as the spreads and maturity – by the
respective facility amounts. Therefore the observation level in the dataset is the package-bank-
firm triplet at a quarterly frequency. Following Murfin (2012), we also report the contracting
date of a package as 90 days prior to the DealScan reported start date, to account for the time
lag between the effective moment in which banks and firms commit to loan contract terms and
the legal start date reported by DealScan.
2.2. Two Measures of Bank Specialization
We are interested in understanding whether banks specialize in lending to specific sectors of
the economy. To address this issue, we employ two approaches. The first consists in comparing
15.
Specifically, DealScan has two fields that can be used to determine the lead arranger, a text variable that defines
the lender role in the syndicate and a yes/no lead arranger credit variable, both are employed to define which
bank has a lead role. Chakraborty et al. (2018), who in turn follow Bharath, Dahiya, Saunders, and Srinivasan
(2007); Bharath et al. (2011), defines as lead arranger, within each syndicate, the bank that “scores” highest in the
following ten-part raking: “1) lender is denoted as “Admin Agent”, 2) lender is denoted as “Lead bank”, 3) lender is
denoted as “Lead arranger”, 4) lender is denoted as “Mandated lead arranger”, 5) lender is denoted as “Mandated
arranger”, 6) lender is denoted as either “Arranger” or “Agent” and has a “yes” for the lead arranger credit, 7)
lender is denoted as either “Arrange” or “Agent” and has a “no” for the lead arranger credit, 8) lender has a “yes”
for the lead arranger credit but has a role other than those previously listed (“Participant” and “Secondary investor”
are also excluded), 9) lender has a “no” for the lead arranger credit but has a role other than those previously
listed (“Participant” and “Secondary investor” are also excluded), and 10) lender is denoted as a “Participant” or
“Secondary investor”." (Chakraborty et al. 2018, Online Appendix, p.1)
16.
However, to compute the measures of specialization we retain every loan from 1987 to 2016 for which we can
identify a borrower in Compustat and for which the TFIC classification is available, regardless of headquarter or
SIC codes. Computing the measures of specialization from 1996 does not affect our results.
8
how concentrated the commercial lending portfolio of an average bank is relative to the whole
syndicated-loan market portfolio. Intuitively, if banks are more concentrated than the market, it
means at the very least that they prefer to focus their lending towards some, but not all, sectors
of the economy – implying a certain degree of specialization. The second involves identifying
those banks that are abnormally exposed to a given industrial sector with respect to the other
banks active in that sector, that is, specialized in that industry.
The goal of the first approach is to present a stylized fact about the syndicated loan market
and understand whether banks, on average, display more or less portfolio concentration than
what would be implied by the overall distribution of credit to the US economy. The goal of the
second approach is instead to classify each bank as specialized – or not – in lending to a given
industry. In the rest of our empirical analysis we will use this second approach to classify banks
as specialized or not.
2.2.1. Methodology
In the first approach, we employ the Herfindahl-Hirschman Index (HHI), commonly used to
measure the degree of market concentration. Specifically, we use it to characterize the level of
concentration of the market portfolio and of the average bank, with respect to the different
industries in the economy.
17
The HHI of the commercial lending portfolio of a given bank is
defined as follows:
HH I
b,t
=
I
X
i=1
L
2
i,b,t
(1)
in which
L
i,b,t
denotes the portfolio share of loans from bank
b
, towards industry
i
, at time
t
.
HH I
b,t
reaches its maximum – which is equal to 1 – in presence of a perfectly concentrated
portfolio, i.e.
L
i,b,t
= 1 for only one industry
i
, and 0 for all the others, and its minimum – equal
to 1/I in presence of a perfectly diversified portfolio, i.e. L
i,b,t
= 1/I ∀i ∈ I.
18
We can then compute the HHI for the average bank by simply taking a weighted average of
17.
There are various ways to characterize portfolio concentration/diversification. See Avila, Flores, Lopez-Gallo,
and Marquez (2013) for a comparison of the various approaches employed in banking and finance.
18.
We define portfolio shares as decimal values between 0 and 1, i.e. they are not in percentage terms. Therefore
the HHI is bounded between 0 and 1. In other applications in which percentage terms are used the HHI varies
between 0 and 10000.
9
the HHI of all banks, in which the weights are represented by a bank’s share of total credit:
HH I
t
=
B
X
b=1
L
b,t
L
t
I
X
i=1
L
2
i,b,t
(2)
in which
L
b,t
=
P
I
i=1
L
i,b,t
is the total amount of credit issued by bank
b
still outstanding at
time t and L
t
=
P
B
b=1
L
b,t
is the total amount of credit outstanding at time t.
Similarly, we can define the HHI for the market portfolio. If we think of all the credit
exposures of all the banks, summed together at a given time, as the “market” portfolio for the
syndicated loan market at that time, we can define the HHI for the “market” portfolio as follows:
HH I
M,t
=
I
X
i=1
L
2
i,t
(3)
in which L
i,t
=
P
B
b=1
L
i,b,t
denotes the share of credit – from all banks – towards industry i, in
the whole syndicated loan market.
In the second approach, we adapt the methodology developed by Paravisini, Rappoport, and
Schnabl (2017) to capture bank specialization at the industry level. According to Paravisini
et al. (2017), a bank is specialized in lending towards a given industry if its portfolio share of
loans outstanding in that industry is abnormally large, relative to other banks. More formally,
specialization is a dummy variable, defined as follows:
Spec
i,b,t
=
1 if L
i,b,t
≥ L
∗
it
0 otherwise
(4)
where L
∗
it
= L
75-th pctile
i,b,t
+ 1.5 ∗
L
75-th pctile
i,b,t
− L
25-th pctile
i,b,t
in which
L
i,b,t
is, as above, the share of credit issued bank
b
to industry
i
outstanding at time
t
, and
L
∗
it
is an extreme value defined as the sum of two items: i) the 75
th
percentile of the
distribution of bank portfolio shares in industry
i
at time
t
, and ii) 1
.
5 times the inter-quartile
range of the same distribution. In other words, according to this approach, a bank is specialized
in an industry if it is a right-tail outlier in the distribution of portfolio shares of lending by all
10
banks towards that industry.
To help understand this approach and highlight its advantages over measures that capture
the overall concentration of the loan portfolio, such as the HHI, Figure 1 presents some simple
examples involving two banks and an economy with only two sectors. In panel (a) neither
bank is specialized as each bank’s balance sheet is split in half between the two sectors, and the
pattern is equal across banks. Panel (c) is similar to the first case. Although one bank is larger
and the other smaller, and they are both mostly exposed to sector A, the pattern of exposure is
the same. Thus, large exposures to sector A might simply reflect a different demand of credit
from sector A with respect to sector B in that particular economy, and we cannot detect evidence
that one particular bank is specialized.
In panel (b), instead, we have an example of specialization. In this case, Bank 1 is specialized
in sector A, and Bank 2 in sector B. Each bank may lend to both sectors – and they do – but each
of them is abnormally exposed to one sector, indicating a bank-level pattern that is coherent
with comparative advantage in lending towards that sector. This does not depend simply on
the amount of credit that goes from each bank to each sector. In fact, in panel (d), Bank 1 is
specialized in sector A, and bank 2 is specialized in sector B. Bank 1 provides overall more
credit to sector B than Bank 2, but its portfolio share is really small compared to Bank 2, which
only lends to sector B.
2.2.2. Specialization in Lending in the US Syndicated Loan Market
To compute these measures of specialization, we need granular information on banks’
commercial lending portfolios. For this purpose, we rely on DealScan, which allows us to obtain
data on bank-firm credit relationships. The focus is on syndicated lending, which represents a
sizable portion of the corporate loan market in the US. Since DealScan only provides information
on loan originations, we create a panel akin to a credit registry by aggregating DealScan loan-
level data at the bank-firm relationship level over time, similarly to Chakraborty et al. (2018),
Gomez, Landier, Sraer, and Thesmar (2021), and Lin and Paravisini (2012).
We assume each loan is outstanding until the original end date, or, if the information
is available on DealScan, until the amended end date.
19
In this way we obtain a dynamic
19.
To track loan amendments, we exploit the information present in the "facilityamendment" table present in the
11
representation of the commercial lending portfolio for each bank in our sample, which we then
use to compute time-varying portfolio shares in each industry by aggregating loan amounts for
each bank-firm relationship at each given point in time.
Since a bank portfolio share towards a given industry is a proxy to capture comparative
advantage in lending towards specific types of projects in the economy, we use the Text-based
Fixed Industry Classifications (TFIC) developed by Hoberg and Phillips (2010, 2016), which
better measures similarities across firms with respect to a standard SIC or NAICS classification,
and is updated annually. Specifically, TFIC uses textual data to track the products (types of
projects) that characterize each firm’s core business activity. Then, it classifies firms as belonging
to a specific cluster (industry) based on the similarity of the firm’s core activity. This classification
follows the evolution of the firm’s core business over time, and thus it is closer in spirit to what
we aim to measure than a static NAICS or SIC industry definition.
We employ the 25-industry version of their classification, as this ensures a good balance
between the number of firms present in DealScan in each industry and sufficient precision in
the characterization of the different sets of projects in the economy. We apply the methodology
described in the previous subsection and compute the two measures of specialization for all the
banks in the sample of syndicated loans granted to firms that have a TFIC classification, from
1987 to 2016.
First, we look at the measure of loan portfolio diversification. In Figure 2, we plot the HHI
of the commercial lending portfolio for the average bank computed for each quarter as in
Equation (2), and the same measure computed for the market portfolio as in Equation (3).
Given that a larger value of this measure implies larger concentration of exposure, a comparison
of the two reveals that the loan portfolio of the average bank is more concentrated than the
market. Comparing the average HHI of the market portfolio (
∼
0
.
07) and that of the average
bank (
∼
0
.
105) over time, we see that the average bank is significantly more concentrated
than the market. This implies that not every bank is lending to every industry in the same way,
legacy version of Dealscan in WRDS. One potential caveat is that renegotiated/amended loans could appear as
new loans in DealScan; if loan renegotiations are not identically and independently distributed across bank-firm
pairs, this could imply an imperfect measurement of a bank’s lending activity. To partially address this issue, we
perform our analysis dropping from our sample all the loans that have a description such as "This loan amends
and restates..." in the various "comment" fields available in Dealscan. All the results of the paper are robust to not
dropping these loans.
12
providing suggestive evidence of specialization in lending.
Second, we look at specialization by industry. Specifically, we are interested in understanding
whether we can observe abnormally large loan portfolio shares towards certain industries,
similarly to what Paravisini et al. (2017) do for countries of destination for Peruvian exporting
firms. Figure 3 shows, at four different moments in time, the box-and-whisker plots of the
distribution of
L
i,b,t
−
¯
L
it
, that is of bank portfolio shares towards each industry
i
demeaned
by the average share of lending in that industry. We can see that across time almost every
industry display at least one or more right-tail outliers; that is, one or more specialized lenders.
Moreover, specialization is persistent. In Figure 4 we plot the autocorrelation of
Spec
i,b,t
defined
in Equation (4), and we can see that a bank specialized in lending towards an industry in a
given year is 25% more likely to be specialized in lending towards the same industry 10 years
later, with respect to a bank that was not specialized.
Overall, the evidence presented in this section points to bank specialization in lending as a
defining feature of the US syndicated loan market.
2.3. Measurement of Economic Variables
2.3.1. Dependent Variables: Loan Covenant Strictness and Loan Spreads
Our goal is to understand whether specialization is associated with information advantages
in lending towards specific sectors of the economy. We, therefore, need an empirical proxy to
capture the notion of information advantage when a bank is lending to firms in a given industry.
To achieve this, we build upon the theoretical work by Gârleanu and Zwiebel (2009), and
consider the covenant structure embedded in a loan contract as capturing the information
“distance” between a bank and a firm. The more restrictive the contract – in terms of what
the firm can or cannot do in order not to trigger a technical default by violating a covenant –
the less information a bank has about a borrower, according to the theory. However, when a
contract includes more than one covenant it is not obvious how to assess the overall strictness
of the covenant package. Therefore, we are going to rely on the measure of covenant strictness
developed and made available by Demerjian and Owens (2016).
20
20.
The measure developed by Demerjian and Owens (2016) can be downloaded on Edward L. Owens’ personal
13
Covenant strictness is defined as the ex-ante probability of violating at least one financial
covenant during the lifetime of the loan, ranging from 0 to 100. This measure is characterized
by four properties, all valid on an “all else equal” basis. First, it increases in the number of
covenants; second, for a fixed number of covenants, it decreases in the initial slack of a covenant,
defined as the distance between the level of the covenant threshold and the starting level of
the corresponding financial ratio; third, it increases in the volatility of the ratios targeted
by covenants; fourth, it decreases in the correlation between covenants—intuitively, since a
technical default is triggered even if a single covenant is violated, contracting on independent
financial ratios increases the probability of violating at least one.
In order to draw conclusions it is also important to track the cost of credit, since there is a
trade-off between covenants and the cost of credit—stricter contracts might be associated with
lower costs and vice versa (Bradley & Roberts, 2015; Matvos, 2013; Reisel, 2014). Therefore
we also collect information on loan pricing available on DealScan. In particular, we focus on the
All-in Drawn Spread (AISD), which is the sum of two terms: i) the spread over LIBOR, which a
borrower needs to pay for every dollar of credit drawn down, and ii) the facility fee, which a
borrower needs to pay annually.
2.3.2. Bank, Firm, and Relationship Level Variables
We obtain bank- and firm-level variables from Compustat, and information on loan quan-
tities and characteristics from Dealscan. Using this merged dataset, we construct proxies for
relationship lending and banks’ industry market share.
We create different proxies to capture the strength of a bank-firm credit relationship. Specifi-
cally, we define four measures, following Bharath et al. (2007, 2011), Schenone (2010), and
Prilmeier (2017).
Previous Rel.
f ,b,t
captures the presence of an existing credit relationship
between firm
f
and bank
b
at the extensive margin. It is a dummy variable that takes value 1 if
bank
b
granted a loan to firm
f
in the 3 years prior to a loan at time
t
.
Rel. Intensity (Amt)
f ,b,t
and
Rel. Intensity (Num)
f ,b,t
capture the strength of the credit relationship at the intensive
margin. They are defined, respectively, as the fraction of credit (loans) that firm
f
obtained
website https://sites.google.com/site/edowensphd/researchdata. We thank Demerjian and Owens (2016) for
making the measure available.
14
from bank
b
over the total amount of credit (number of loans) firm
f
took out over the 3 years
prior to a loan at time
t
. Finally, we compute the length of an outstanding bank-firm relationship.
Rel. Length
f ,b,t
is defined as the time elapsed between period
t
and the first interaction between
firm f and bank b in DealScan.
We also collect information on the geographic distance between the borrower and the lender,
to proxy for “arms-length” credit relationships. In particular, we construct a dummy variable,
Same State
f ,b,t
, which takes value 1 if bank
b
and firm
f
are in the same state at time
t
, and 0
otherwise.
21
Finally, we compute each bank’s industry
Market Share
b, f ,t
. This is the fraction of
credit that a bank
b
provides to the industry of firm
f
over the total credit that the industry
receives at period
t −
1. Taking into account a bank’s industry market share is important since
bank specialization in lending to an industry could be correlated with a high industry market
share. All other variables are defined in Table 1.
2.4. Sample Characteristics
Table 2 reports summary statistics for the samples we use in our empirical analysis. In particu-
lar, we distinguish two samples. The first one, “Matched Sample" is the full DealScan-Compustat
matched sample obtained from the sample selection procedure described in Section 2.1. The sec-
ond one, “Strictness Sample", is the subsample of loans for which both the All-In Drawn Spread
and covenant strictness measure developed by Demerjian and Owens (2016) are non-missing.
We conduct our main empirical analysis over this subsample.
The top panel of Table 2 reports information on the characteristics of loan-level variables
in our samples. The Strictness Sample includes 11
,
684 distinct loans. On average, a loan
agreement contains more than two financial covenants and displays a level of strictness such
that the borrower has 36% probability to violate at least one covenant as well as an All-In-Drawn
Spread of 188 basis points. The average loan package has a maturity of almost 4 years, amounts
to $567 million, and the average syndicate size (number of lenders) is 9. These statistics are
similar to the larger Matched sample, which displays on average a smaller number of covenants,
21.
We use the historical data on firm and bank locations collected from the SEC filings by Bai, Fairhurst, and
Serfling (2020) and Gao, Leung, and Qiu (2021), supplementing them with Compustat header information when
missing.
15
a larger average loan amount, and a slightly smaller number of syndicate members.
The mid panel of Table 2 reports information on the borrowers in our samples. The Strictness
Sample includes 11
,
231 firm-quarter observations for 3
,
634 firms. These are public firms,
large – on average $1 billion in total assets – and mature – on average 20 years since IPO.
55% do not have a long-term issuer credit rating, and for those that have a rating, the average
rating is BBB-/BB+.
22
Over our sample period (1996-2016), they enter, on average, into 9
syndicated loan agreements. Overall, there are no major differences between the Strictness and
the Matched Sample.
Finally, the bottom panel of Table 2 reports information on the lenders in our samples. The
Strictness Sample includes 2
,
093 bank-quarter observations for 95 banks. The average bank is
large, with $200 billion in total assets, a deposit-to-asset ratio of 60%, with book equity capital
amounting to 7%.
3. Bank Specialization and Loan Contract Terms
In this section, we explore the effect of bank specialization in lending on loan covenant
strictness and the cost of credit. We first perform a simple univariate analysis, which highlights
potential non-randomness in the matching between banks and firms. Employing different mul-
tivariate specifications aimed at mitigating this concern, we then show that bank specialization
is associated with significantly lower covenant strictness and no higher spreads.
We interpret this evidence as support for explanations of bank specialization based on
lending advantages, and we suggest that part of this advantage is an information advantage,
which is sector-specific. Finally, using default on lenders’ loan portfolios as a possible source
of exogenous variation in banks’ perception of their own expertise in dealing with a certain
industrial sector, we show that specialized banks are more sensitive to defaults of firms in their
sector of specialization, further substantiating our interpretation.
22.
Rating is a categorical variable. We assign value 1 to AAA ratings, 2 to AA, and so on. The largest value is 9,
assigned to “D" or “SD" indicating default in the Capital IQ Long-Term Issuer Credit Rating.
16
3.1. Univariate Analysis
We begin by comparing the characteristics of loans arranged by a bank specialized in lending
towards the industry a given borrower belongs to, with all other loans. To make things clear,
a loan to a firm
f
starting at time
t
is considered to be arranged by a specialized bank
b
if
Spec
i,b,t−1
, defined in Equation (4), is equal to 1 and the firm
f
belongs to industry
i
. The top
panel of Table 3 reports the results of these basic univariate t-tests.
Loans arranged by specialized banks in their industries of specialization display several
different features compared to loans arranged to other industries and/or non-specialized banks,
even though they are similar in their amount. In particular, “specialized loans” display stricter
covenants, higher spreads, shorter maturities, a more concentrated syndicate, and a lower
fraction of revolving credit, compared to “non-specialized loans”. Even though this is suggestive
of a relationship between bank specialization and contract features, this evidence may simply
arise from the different characteristics of specialized banks and their borrowers. By performing
t-tests on bank and firm characteristics, we aim to understand whether this is the case.
The mid panel of Table 3 displays the results of the
t
-tests for firm characteristics.
23
The
estimates confirm that firms obtaining loans from banks specialized in the industry they belong
to are generally different from other firms. They are smaller, younger, less likely to have a
long-term issuer credit rating – even though if they do have a credit rating, it is on average
similar to other firms. This implies that a firm borrowing from banks specialized in its own
industry is less likely to have access to public debt/equity markets and thus subject to more
severe information frictions. These firms also appear to perform slightly better in terms of
liquidity, tangibility, and leverage.
The bottom panel of Table 3 shows the estimate of
t
-tests on bank characteristics.
24
To
be clear, a bank can appear both in the “specialized” and “non-specialized” sample at a given
moment in time. With this caveat in mind, what emerges is that banks specialized in lending
towards a given sector are different compared to other banks. Specifically, they are smaller,
have a larger reliance on deposits, appear to be better capitalized, and are more profitable, with
23.
We split all firm-quarter observations into those that are associated with a loan arranged to any sector any
bank is specialized in, and those that are not. We do the same for bank-quarter observations.
24.
Since the same bank issue more than one loan, the standard errors for the
t
statistics in Table 3 have been
adjusted for clustering at the bank level.
17
a similar ratio of non-performing assets.
Overall, the evidence in Table 3 suggests that bank specialization might play a role in
determining loan characteristics, but any conclusion based on simple univariate analysis would
be distorted by the pervasive selection in the matching between borrowers and lenders. In the
next Section, we analyze this in a multivariate regression framework with fixed effects.
3.2. Empirical Strategy: A Within-Bank Approach
To retrieve the effect of bank specialization on loan covenant strictness, ideally, we would
like to observe identical firms borrowing from two different banks, one specialized in lending
towards the firm’s industry and one not specialized. In particular, the firms should be randomly
assigned to the banks, and each bank should differ from each other only for its specialization
status. However, matching between banks and firms is rarely random and loan contract terms
are an outcome of this matching process. If, as Table 3 suggests, specialized banks are small
banks that in general tap a pool of borrowers that are smaller, more opaque, and riskier, any
observed variation in the loan covenant strictness might just be the direct consequence of the
systematically different characteristics of the firms and banks involved.
25
To mitigate these concerns, we proceed in the following way. We start from a within-bank
approach, akin to the one proposed at the firm level by Khwaja and Mian (2008). Underlying
our empirical strategy there is the idea of comparing two loans arranged by the same bank in
the same year-quarter, one issued to a borrower in an industry in which the bank is specialized
in lending to, and one issued to a borrower in another industry. This, however, does not fully
account for borrower selection problems. Even after absorbing all bank-specific, time-varying
characteristics, it may be the case that within each bank’s borrower pool, firms that fall within
the industries in which the bank is specialized are systematically different. To address this,
we first include firm balance sheet controls, which absorb variation due to observable and
time-varying firm characteristics. Furthermore, we add firm fixed effects, which account for all
firm-specific, observable, and unobservable characteristics that are fixed in time.
26
25.
This systematic difference can regard both observable and unobservable characteristics. It is in fact well known
in the literature that covenant strictness reflects borrower riskiness (Demiroglu & James, 2010), and ex-ante bank
confidence in the underwritten loans (Murfin, 2012).
26.
Ideally, we would rather to have a within bank-time and within firm-time specification. Unfortunately, as we
18
Formally, we employ the following specification:
Loan Contract Term
f ,b,t
= θ
b,t
+ Other Fixed Effects + β · Specialization
f ,b,t−1
+ γ
F
· Firm Controls
f ,t
+ γ
L
· Loan Controls
f ,b,t
+ ϵ
f ,b,t
(5)
in which
Loan Contract Term
f ,b,t
stands for loan covenant strictness or the AISD for a loan
originated in quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects; the
term
Other Fixed Effects
includes borrower fixed effects and separate intercepts for each S&P
long-term issuer credit rating, with the omitted dummy variable capturing unrated firms. The
main explanatory variable of interest is
Specialization
, a lagged 12-quarters rolling average of
the specialization dummy
Spec
i,b,t
(defined in Equation (4)).
Firm Controls
includes firm-level
proxies of time-varying risk. Specifically, it includes the expected default probability (EDF),
based on the Merton model of credit risk (Merton, 1974) and computed implementing the
“naive” approach proposed by Bharath and Shumway (2008), as well as log of total assets,
debt to asset ratio, current ratio, tangible net worth to asset ratio, interest coverage ratio, and
years since IPO. These controls account for repayment risk (especially for non-rated firms), size,
leverage, liquidity, the ability to provide collateral, firm profitability, and firm age.
27
Finally,
Loan Controls
includes also loan-level controls such as log of maturity, log of loan amount, log
of number of syndicate participants, the fraction of revolving credit over the total package
amount, and separate intercepts for different loan purposes.
We make the choice to average the specialization dummy over 12 quarters to put less weight
on banks that are only sporadically specialized in a sector. This might be simply the result of a
single large loan at a time of relatively low lending activity in that industry, or measurement
error due to the limitations of our dataset. We chose 12 quarters as this length ensures a
good balance between capturing persistence and avoiding that our measure simply mimics the
origination of new loans—the average maturity of a loan in DealScan is around 4 years.
28
work on a sample of very large loans, we do not see many firms doing multiple deals in the same year-quarter.
This makes the adoption of such a strategy infeasible.
27.
Similar controls are used in similar studies focusing on the determinants of loan covenant strictness, such as
Murfin (2012) or Prilmeier (2017).
28.
However, we stress that performing the analysis with a measure of specialization averaged over rolling
windows of different lengths does not change the main results of the paper—see the robustness checks presented
in Section 3.6. Finally, we note that the same choice of rolling window length, performed for similar reasons, can
also be found in Paravisini et al. (2017).
19
3.3. The Effect of Specialization on Loan Covenant Strictness and Pricing
We now introduce the baseline results of our analysis. Table 4 reports the regression estimates
of the specification in Equation (5) over the Strictness Sample, for two of the main different loan
contract characteristics: Covenant Strictness and the All-In Drawn Spread (AISD). Looking at
covenant strictness first, the estimate on the specialization variable is negative and statistically
significant, indicating that banks specializing in lending towards a given industry write less
strict contracts when entering loan agreements with firms in that industry. The estimates on
AISD are also negative across specifications.
In particular, a simple regression of covenant strictness on bank-time fixed effects shows that
a loan contract with a firm in the bank’s area of specialization displays less strict covenants
by 12.4 percentage points compared to firms in other industries (column 1). This estimate is
economically significant, as it amounts to 33% of the mean value and 30% of the standard
deviation of the distribution of covenant strictness in our sample. Note that this is not associated
with higher loan spreads: the point estimate on the specialization variable in relation to the
AISD is negative (column 4), even though not statistically significant.
When we account for borrower selection and borrower risk by including firm fixed effects
and firm controls, the results are even stronger. The point estimate of the coefficient on bank
specialization doubles for both covenant strictness and AISD. Banks specialized in an industry
provide credit to firms in that industry with covenants looser by 24 p.p. relative to firms in
other industries, with no higher cost of credit (columns 2 and 5). Alternatively, a 1 standard
deviation increase in bank specialization implies a decrease in covenant strictness by 4 p.p. These
results reduce concerns that the effect of specialization is entirely a byproduct of unobserved
heterogeneity in borrower types and riskiness.
However, banks specializing in lending towards certain industries might provide credit with
characteristics that are systematically different compared to non-specialized banks; e.g. suppose
that specialized banks only agree to provide credit in the form of term loans, whereas non-
specialized banks only in the form of revolving credit. To address this concern we include
loan controls to the baseline specification, and as shown in columns 3 and 6, results are
virtually unchanged, while the effect on AISD becomes slightly larger and marginally statistically
20
significant.
The negative, economically and statistically significant effect of the specialization variable
on covenant strictness suggests less information asymmetry, or “distance”, between a bank
specialized in lending towards a given industry and firms in that industry, in line with the
theoretical framework developed by Gârleanu and Zwiebel (2009). The negative estimates
on the same specialization variable when loan spreads are the dependent variables support
this interpretation, ruling out that lower strictness is compensated with a higher cost of credit,
which would weaken the notion of lending advantage. It appears that specialized banks not
only leave more leeway to their borrowers, but they appear not to see this as a risk for which
they must be properly compensated. This suggests less restrictive covenants actually reflect
better ex-ante knowledge of the projects/capacity to screen them, and this is consistent with
explanations of bank specialization based on the existence of lending advantages, specifically
an industry-specific information advantage.
3.4. Assessing Alternative Explanations
There might be alternative explanations for the results presented in Table 4. In particular,
the results presented so far are consistent with at least three other economic mechanisms: the
presence of borrower-specific knowledge (relationship lending), insurance incentives stemming
from a high industry market share, and local knowledge spillovers implied by geographical,
rather than industry, specialization.
Relationship Lending
First, we could argue that the industry-specific information advantage could originate from
an information advantage that is borrower-specific. This would be consistent with widespread
“relationship lending” (Berger & Udell, 1995; Petersen & Rajan, 1994, 1995). For example,
Bharath et al. (2011) and Prilmeier (2017) specifically show that relationship lending matters
for the determination of covenants and other contract terms in syndicated loan agreements.
To explore the role that borrower-specific information might have on the determination of loan
covenant strictness, we include in our specification the various empirical proxies we described in
21
Section 2, which are meant to capture different aspects of relationship lending. Table 5 reports
the results for these regressions. Across all specifications, for both covenant strictness and loan
spreads, the estimated coefficient on the specialization variable is virtually unchanged and still
statistically significant, validating the hypothesis that banks have an information advantage
that stems from an industry-specific expertise and not only from borrower-specific information.
In conclusion, we see that an explanation based only on relationship lending does not
seem appropriate to rationalize the observed relationship between bank specialization and the
existence of an information advantage relative to that industry.
High Industry Market Share
Second, banks that are specialized in lending towards a given industry might also provide
a relatively large share of credit to that industry, i.e. have a high industry market share. This
would point to at least two other potential explanations for our results. On the one hand, the
observed effect of bank specialization on contract terms might in fact simply reflect a bank’s
willingness to gain market share in an industry. Banks could offer favorable credit terms to
crowd out other lenders from a given industry, thereby increasing both their industry market
share and their industry portfolio share. If this is true, the observed effect on contract terms
should be driven by the bank’s industry market share and not by specialization.
On the other hand, banks with a high market share might have incentives to offer better
contract terms to borrowers for reasons unrelated to an information advantage. Specifically,
Giannetti and Saidi (2019) show that banks with a high market share in an industry are more
likely to internalize negative spillovers and possible systemic effects of tougher credit conditions
in that industry – as well as upstream and downstream the related supply chain – in periods
of distress. For analogous reasons, they might have incentives to write less strict contracts to
avoid triggering covenant violations that might potentially be costly not only for the specific
firm – in terms of investment, for example – but also for the entire industry the firm is part of.
To address both these issues, we include in our specifications the variable
Market Share
,
defined in Section 2, which is the share of credit outstanding that a bank has in one industry
relative to the total credit supplied to the industry by all banks. Table 6 reports the results
for these regressions. When looking at covenant strictness, the estimates on the specialization
22
variable are slightly larger and still highly significant. For loan spreads, we still observe a
negative effect on the specialization variable, confirming the results of the main analysis.
Turning to the effect of a high market share, we can see that the estimated coefficient for
the market share variable on covenant strictness is not significant (columns 1-3), whereas it is
positive and significant on AISD (columns 5 and 6). We have two possible explanations for this
relation. First, it may be a result of the fact that banks with high market share in an industry
have a larger pool of loans in that industry by construction. As a consequence, their marginal
borrower is of lower quality, has a higher information distance from its lender, and receives
higher loan spreads. Second, it may be the case that these banks have overall higher market
power, which increases their charter value, making them less willing to take risks (Keeley,
1990).
Geographical Proximity
Third, the literature points to the role of geographic distance as an important proxy for
the degree of asymmetric information between borrowers and lenders. Loan terms are more
favorable when borrowers are geographically closer to lenders (Agarwal & Hauswald, 2010;
Alessandrini, Presbitero, & Zazzaro, 2008; Degryse & Ongena, 2005), even in presence of large
corporations (Hollander & Verriest, 2016).
We are thus concerned that banks specialized in lending towards a given industry have
an abnormal exposure to that industry because they are lending to specific locations that
feature business concentration in that industry and that are geographically close to these banks’
headquarters. This geographical proximity between banks and firms in specific industries might
in turn explain our results. If this is the case, we would still interpret our result in light of
an information advantage of these banks. However, this advantage would not stem from an
industry-specific expertise, but from the acquisition of soft information based on geographical
proximity. To address this issue, we construct a dummy variable,
Same State
, which takes value
1 if the bank and the firm headquarters are located in the same state, and we include it in our
specifications.
Table 7 presents the results for these regressions. Consistent with the notion that geographical
proximity between borrowers and lenders reflects a lower level of asymmetric information,
23
the estimates on the same-state dummy are negative for both covenant strictness and loan
spreads. However, they are not significant. On the other hand, the estimated coefficients on the
specialization variable are essentially the same as the baseline specifications.
3.5. Specialization and Defaults on L ender Portfolios
To provide further evidence in support of our proposed interpretation, we employ defaults
on lender loan portfolios as a relatively exogenous source of variation to the lenders’ perception
of their own screening ability (Murfin, 2012). We examine whether defaults of firms in industry
i
that have outstanding loans with bank
b
differentially affect the contracting behavior of banks
that are specialized in lending to
i
and banks that are not. In particular, we focus on how
covenant strictness changes for loans underwritten by specialized and non-specialized banks
following the default shocks.
We compute the number of defaults each bank experiences in its loan portfolio by counting
instances in which borrowers with an outstanding loan with a given bank have a credit rating of
“D” or “SD” over a period of 90 days, following Murfin (2012). Suppose that banks specialized
in lending towards one sector have an information advantage in screening or monitoring
specific projects in that sector. We posit that, for a given number of borrowers in an industry
defaulting while having outstanding loans with a bank, banks specialized in lending towards
that industry would revise more the perception of their own ability of screening borrowers in
that industry, compared to banks not specialized in lending towards that industry. Indeed, a
default in a given industry should be relatively more informative for those banks who have an
information advantage for that given industry. If defaults occur in industries out of a bank’s
area of specialization, on the other hand, we should observe a smaller or null revision of a
bank’s own screening ability.
We empirically test this implication by employing a specification similar to the one in
Equation (5), with the inclusion of interaction terms between the specialization variable and
24
the number of defaults on lender portfolio, as follows:
Loan Term
f ,b,t
= θ
b,t
+ θ
f
+ ρ · Specialization
f ,b,t−1
× Defaults
b,t−1
(6)
+ β · Specialization
f ,b,t−1
+ γ
D
· Defaults
b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
The coefficient of interest is
ρ
, which measures the differential effect on loan contract terms
of a specialized bank in response to one more default with respect to a non-specialized bank. In
particular, given a loan agreement between a bank
b
and a firm
f
that starts at time
t
, we are
going to consider two different types of Defaults variable: Defaults (same), which denotes only
the defaults that have occurred in the same industry as
f
, and Defaults (other), which considers
the total number of defaults occurred in all other industries. Crucially, we expect a positive and
significant effect only for the interaction term of the specialization variable and Defaults (same),
but not for the interaction term with Defaults (other). This amount to saying the following: a
bank that is specialized in lending towards industry
i
, when lending to a borrower in industry
i
,
is going to be more responsive in making covenants stricter– relative to a bank that is lending
to the same industry – only when it experiences borrower defaults in industry
i
, and not when
the defaults occur in other industries.
Table 8 shows the results of regressions as in Equation (6), and the evidence is consistent
with our hypotheses. From this table, we can observe that the coefficient on the specialization
variable is very similar to the baseline regression, and still highly significant. Two patterns also
emerge. When specialized banks incur defaults in their loan portfolios, they increase covenant
strictness relatively more than non-specialized banks, but only when these defaults occur to
borrowers in the industry the bank specializes in. When defaults occur outside of the industry
of specialization of the bank, there is no differential response in terms of covenant strictness. In
fact, the coefficient on the interaction term between the specialization variable and the number
of defaults is positive and highly statistically significant only when the defaults occur in the
industry of specialization of the bank
In terms of the economic interpretation of the coefficients, a specialized bank in an industry
that suffers from one default in a quarter, and this default concerns a borrower in its industry
of specialization, will respond by increasing covenant strictness by approximately 30 p.p.
25
relative to a specialized bank that does not experience default, and by approximately 6 p.p
(
−
24
.
36 + 30
.
28
×
1) relative to a non-specialized bank that suffers from 1 default in the same
industry.
3.6. Robustness Checks
The results presented so far stand to a series of robustness checks. First, restricting the
analysis only to loans with a single lead arranger confirms the baseline results, as shown in
Table 9. Second, computing the specialization measures starting from 1996 instead of 1987
leaves the results virtually unchanged; the estimates are presented in Table 10. Third, repeating
the analysis focusing on the pre-2008 sample period also confirms the main results, and if
anything the estimates are even stronger, as displayed in Table 11. This alleviates concerns
that our results are driven by the post-financial crisis period, in which the share of leveraged,
covenant-light loans increased dramatically, and relatedly the coverage of covenants offered by
Dealscan appears to have decreased in quality (Bräuning, Ivashina, & Ozdagli, 2022).
Finally, averaging the specialization dummy defined in Equation (5) over different time
horizons does not change the main message of the paper. As can be seen in Table 12, the
effect of the specialization variable on covenant strictness is very similar in both economic
magnitude and statistical significance when averaging over 3, 4 or 5 years, in particular for
covenant strictness (columns 3, 4 and 5). The estimate of specialization on covenant strictness
is attenuated both economically and statistically when averaging over 4 or 8 quarters (1 or 2
years), but on the other hand, it is larger and statistically significant when looking at the AISD.
The lower and mostly non-statistically significant estimates that we obtain when averaging
the specialization dummy over a period of 1 and 2 years could actually represent an indirect
validation of our proposed mechanism. It takes time to build expertise that is industry-specific,
and therefore estimates on covenant strictness are larger and less noisy once the average of the
specialization dummy is taken over longer periods.
26
4. Conclusion
In this paper we provide evidence that banks specialize in lending toward specific industries
even in a credit market for large borrowers, such as the U.S. syndicated loan market. We
show that loan contracts between borrowers in an industry and banks specialized in lending
towards that industry display a less restrictive covenant structure and no higher spreads. This,
comparing two loans made by the same bank in the same year-quarter, one towards the industry
of specialization and one to any other industry. Our results cannot be fully explained by borrower
risk, relationship lending, a high industry market share, or geographical proximity.
We look at our results in light of financial contracting theory, and interpret the restrictiveness
of the covenant structure as the degree of information asymmetry between a borrower and a
lender (Gârleanu & Zwiebel, 2009). Thus, we conclude that specialized banks have a compara-
tive advantage in monitoring specific industries. This carries implications for the understanding
of competition and monopoly power in credit markets, and thus for the transmission mechanism
of monetary policy and potential heterogeneous effects of regulation (see Corbae & D’Erasmo,
2021). Moreover, documenting implications on the non-price conditions of credit, we propose
a possible mechanism that makes credit by specialized banks difficult to substitute (Paravisini
et al., 2022).
27
References
Abuzov, R., Herpfer, C., & Steri, R. (2020). Do Banks Compete on Non-Price Terms? Evidence
from Loan Covenants. Available at https://ssrn.com/abstract=3278993
Acharya, V. V., Hasan, I., & Saunders, A. (2006). Should Banks Be Diversified? Evidence from
Individual Bank Loan Portfolios. Journal of Business, 79(3), 1355–1412.
Agarwal, S., & Hauswald, R. (2010). Distance and Private Information in Lending. Review of
Financial Studies, 23(7), 2757–2788.
Alessandrini, P., Presbitero, A. F., & Zazzaro, A. (2008). Banks, Distances and Firms’ Financing
Constraints. Review of Finance, 13(2), 261–307.
Avila, F., Flores, E., Lopez-Gallo, F., & Marquez, J. (2013). Concentration Indicators: Assessing
the Gap Between Aggregate and Detailed Data. In Proceedings of the Sixth IFC Conference
on “Statistical issues and activities in a changing environment” (Vol. 36, pp. 542–559).
Bai, J. J., Fairhurst, D., & Serfling, M. (2020). Employment Protection, Investment, and Firm
Growth. Review of Financial Studies, 33(2), 644–688.
Bao, Y. (2019). Peer Information in Loan Pricing. Available at https://ssrn.com/abstract_id=
3288841
Beck, T., De Jonghe, O., & Mulier, K. (2022). Bank Sectoral Concentration and Risk: Evidence
from a Worldwide Sample of Banks. Journal of Money, Credit and Banking. Available at
https://onlinelibrary.wiley.com/doi/abs/10.1111/jmcb.12920
Berger, A. N., & DeYoung, R. (2001). The Effects of Geographic Expansion on Bank Efficiency.
Journal of Financial Services Research, 19(2-3), 163–184.
Berger, A. N., & Udell, G. F. (1995). Relationship Lending and Lines of Credit in Small Firm
Finance. Journal of Business, 68(3), 351–381.
Berlin, M., & Mester, L. J. (1992). Debt Covenants and Renegotiation. Journal of Financial
Intermediation, 2(2), 95–133.
Bharath, S. T., Dahiya, S., Saunders, A., & Srinivasan, A. (2007). So What Do I Get? The Bank’s
View of Lending Relationships. Journal of Financial Economics, 85(2), 368–419.
Bharath, S. T., Dahiya, S., Saunders, A., & Srinivasan, A. (2011). Lending Relationships and
Loan Contract Terms. Review of Financial Studies, 24(4), 1141–1203.
28
Bharath, S. T., & Shumway, T. (2008). Forecasting Default with the Merton Distance to Default
Model. Review of Financial Studies, 21(3), 1339–1369.
Billett, M. T., King, T.-H. D., & Mauer, D. C. (2007). Growth Opportunities and the Choice of
Leverage, Debt Maturity, and Covenants. Journal of Finance, 62(2), 697–730.
Black, L. K., Krainer, J. R., & Nichols, J. B. (2020). Safe Collateral, Arm’s-Length Credit:
Evidence from the Commercial Real Estate Market. Review of Financial Studies, 33(11),
5173–5211.
Blickle, K., Parlatore, C., & Saunders, A. (2021). Specialization in Banking. Federal Reserve
Bank of New York Staff Report No. 967.
Boeve, R., Duellmann, K., & Pfingsten, A. (2010). Do Specialization Benefits Outweigh
Concentration Risks in Credit Portfolios of German Banks? Discussion Paper Deutsche
Bundesbank, Series 2: Banking and Financial Studies, No 10/2010.
Boyd, J. H., & Prescott, E. C. (1986). Financial Intermediary-Coalitions. Journal of Economic
Theory, 38(2), 211–232.
Bradley, M., & Roberts, M. R. (2015). The Structure and Pricing of Corporate Debt Covenants.
Quarterly Journal of Finance, 05(02), 1550001.
Bräuning, F., Ivashina, V., & Ozdagli, A. K. (2022). High-Yield Debt Covenants and Their Real
Effects. NBER Working Paper No 29888.
Cai, J., Eidam, F., Saunders, A., & Steffen, S. (2018). Loan Syndication Structures and Price
Collusion. Available at https://ssrn.com/abstract=3250817
Carey, M., Post, M., & Sharpe, S. A. (1998). Does Corporate Lending by Banks and Finance
Companies Differ? Evidence on Specialization in Private Debt Contracting. Journal of
Finance, 53(3), 845–878.
Chakraborty, I., Goldstein, I., & MacKinlay, A. (2018). Housing Price Booms and Crowding-Out
Effects in Bank Lending. Review of Financial Studies, 31(7), 2806–2853.
Chava, S., & Roberts, M. R. (2008). How Does Financing Impact Investment? The Role of Debt
Covenants. Journal of Finance, 63(5), 2085–2121.
Chodorow-Reich, G. (2014). The Employment Effects of Credit Market Disruptions: Firm-
Level Evidence from the 2008–9 Financial Crisis. Quarterly Journal of Economics, 129(1),
1–59.
29
Chodorow-Reich, G., & Falato, A. (2022). The Loan Covenant Channel: How Bank Health
Transmits to the Real Economy. The Journal of Finance, 77(1), 85–128.
Corbae, D., & D’Erasmo, P. (2021). Capital Buffers in a Quantitative Model of Banking Industry
Dynamics. Econometrica, 89(6), 2975–3023.
Daniels, K., & Ramirez, G. G. (2008). Information, Credit Risk, Lender Specialization and Loan
Pricing: Evidence from the DIP Financing Market. Journal of Financial Services Research,
34(1), 35–59.
Degryse, H., & Ongena, S. (2005). Distance, Lending Relationships, and Competition. Journal
of Finance, 60(1), 231–266.
De Jonghe, O., Dewachter, H., Mulier, K., Ongena, S., & Schepens, G. (2020). Some Borrowers
Are More Equal than Others: Bank Funding Shocks and Credit Reallocation. Review of
Finance, 24(1), 1–43.
Demerjian, P. R., & Owens, E. L. (2016). Measuring the Probability of Financial Covenant
Violation in Private Debt Contracts. Journal of Accounting and Economics, 61(2-3), 433–
447.
Demerjian, P. R., Owens, E. L., & Sokolowski, M. (2018). Lender Capital Adequacy and
Financial Covenant Strictness. Available at https://ssrn.com/abstract=3240916
Demiroglu, C., & James, C. M. (2010). The Information Content of Bank Loan Covenants.
Review of Financial Studies, 23(10), 3700–3737.
Di, W., & Pattison, N. (2022). Industry Specialization and Small Business Lending. Available at
https://pattison-nate.github.io/files/DiPattison_march2022.pdf
Diamond, D. W. (1984). Financial Intermediation and Delegated Monitoring. Review of Economic
Studies, 51(3), 393–414.
Duquerroy, A., Mazet-Sonilhac, C., Mésonnier, J.-S., & Paravisini, D. (2022). Bank Local
Specialization. Banque de France Working Paper No. 865.
Gao, M., Leung, H., & Qiu, B. (2021). Organization Capital and Executive Performance
Incentives. Journal of Banking & Finance, 123, 106017.
Gârleanu, N., & Zwiebel, J. (2009). Design and Renegotiation of Debt Covenants. Review of
Financial Studies, 22(2), 749–781.
Giannetti, M., & Saidi, F. (2019). Shock Propagation and Banking Structure. Review of Financial
30
Studies, 32(7), 2499–2540.
Gomez, M., Landier, A., Sraer, D., & Thesmar, D. (2021). Banks’ Exposure to Interest Rate Risk
and the Transmission of Monetary Policy. Journal of Monetary Economics, 117, 543–570.
Gopal, M. (2021). How Collateral Affects Small Business Lending: The Role of Lender
Specialization. U.S. Census Bureau Center for Economic Studies Working Paper 21-22.
Gorostiaga, J. (2022). It’s Not You, It’s Them: Industry Spillovers and Loan Portfolio Optimiza-
tion.
Hayden, E., Porath, D., & Westernhagen, N. (2007). Does Diversification Improve the Perfor-
mance of German Banks? Evidence from Individual Bank Loan Portfolios. Journal of
Financial Services Research, 32(3), 123–140.
Hoberg, G., & Phillips, G. (2010). Product Market Synergies and Competition in Mergers and
Acquisitions: A Text-Based Analysis. Review of Financial Studies, 23(10), 3773–3811.
Hoberg, G., & Phillips, G. (2016). Text-Based Network Industries and Endogenous Product
Differentiation. Journal of Political Economy, 124(5), 1423–1465.
Hollander, S., & Verriest, A. (2016). Bridging the Gap: The Design of Bank Loan Contracts and
Distance. Journal of Financial Economics, 119(2), 399–419.
Hubbard, R. G., Kuttner, K. N., & Palia, D. N. (2002). Are There Bank Effects in Borrowers’
Costs of Funds? Evidence from a Matched Sample of Borrowers and Banks. Journal of
Business, 75(4), 559–581.
Hughes, J. P., Lang, W. W., Mester, L. J., & Moon, C.-G. (1996). Safety in Numbers? Geographic
Diversification and Bank Insolvency Risk. Federal Reserve Bank of Philadelphia Working
Paper No. 96-14.
Ivashina, V. (2009). Asymmetric Information Effects on Loan Spreads. Journal of Financial
Economics, 92(2), 300–319.
Jiang, S., & Li, J. Y. (2022). He Who Lends Knows. Journal of Banking & Finance, 106412.
Keeley, M. C. (1990). Deposit Insurance, Risk, and Market Power in Banking. American Economic
Review, 80(5), 1183–1200.
Khwaja, A., & Mian, A. (2008). Tracing the Impact of Bank Liquidity Shocks: Evidence from an
Emerging Market. American Economic Review, 98(4), 1413–1442.
Liberti, J. M., Sturgess, J., & Sutherland, A. (2017). Information Sharing and Lender
31
Specialization: Evidence from the U.S. Commercial Lending Market. Available at
https://ssrn.com/abstract_id=2965830
Lin, H., & Paravisini, D. (2012). The Effect of Financing Constraints on Risk. Review of Finance,
17(1), 229–259.
Matvos, G. (2013). Estimating the Benefits of Contractual Completeness. Review of Financial
Studies, 26(11), 2798–2844.
Merton, R. C. (1974). On the Pricing of Corporate Debt: The Risk Structure of Interest Rates.
Journal of Finance, 29(2), 449–470.
Murfin, J. (2012). The Supply-Side Determinants of Loan Contract Strictness. Journal of
Finance, 67(5), 1565–1601.
Paravisini, D., Rappoport, V., & Schnabl, P. (2017). Specialization in Bank Lending: Evidence
from Exporting Firms. CEP Discussion Paper No 1492.
Paravisini, D., Rappoport, V., & Schnabl, P. (2022). Specialization in Bank Lending: Evidence
from Exporting Firms. Journal of Finance (forthcoming).
Petersen, M. A., & Rajan, R. G. (1994). The Benefits of Lending Relationships: Evidence from
Small Business Data. Journal of Finance, 49(1), 3–37.
Petersen, M. A., & Rajan, R. G. (1995). The Effect of Credit Market Competition on Lending
Relationships. Quarterly Journal of Economics, 110(2), 407–443.
Prilmeier, R. (2017). Why Do Loans Contain Covenants? Evidence from Lending Relationships.
Journal of Financial Economics, 123(3), 558–579.
Reisel, N. (2014). On the Value of Restrictive Covenants: Empirical Investigation of Public
Bond Issues. Journal of Corporate Finance, 27, 251–268.
Santos, J. A. C., & Winton, A. (2019). Bank Capital, Borrower Power, and Loan Rates. Review
of Financial Studies, 32(11), 4501–4541.
Schenone, C. (2010). Lending Relationships and Information Rents: Do Banks Exploit Their
Information Advantages? Review of Financial Studies, 23(3), 1149–1199.
Schwert, M. (2018). Bank Capital and Lending Relationships. Journal of Finance, 73(2),
787–830.
Tabak, B. M., Fazio, D. M., & Cajueiro, D. O. (2011). The Effects of Loan Portfolio Concentration
on Brazilian Banks’ Return and Risk. Journal of Banking & Finance, 35(11), 3065–3076.
32
Appendix A. Tables
Table 1. Variable Definitions
Variable Name Definition Data Source Unit
Specialization Rolling 12-qtr sum of specialization dummy Dealscan float
as defined in Equation (4) (0-1)
Specialization (nY) Rolling 4n-qtr sum of specialization dummy Dealscan float
as defined in Equation (4) (0-1)
EDF See Bharath and Shumway (2008), pp. 1247-48 CRSP/ float (%)
Compustat
Assets atq Compustat USD Mil
Tangibility (atq - intanq - ltq)/atq Compustat float
Leverage (dlttq + dlcq)/atq Compustat float
Current Ratio actq/lctq Compustat float
Int. Cover. Ratio Rolling 4-qtr sum of oibdq/Rolling 4-qtr sum of xintq Compustat float
Years since IPO Current date minus first date in Compustat Compustat float
Rated Dummy variable equal to 1 if firm-quarter Capital IQ int
has a long-term issuer credit rating, 0 otherwise (0/1)
Rating Categorical variable equal to 1 for credit rating "AAA", Capital IQ int
to 2 for "AA", ... , to 9 for "D"/"SD" (indicating default)
N. Loans Number of packages per borrower over sample period Dealscan
Covenant Strictness Ex-ante prob of violating a financial covenant. Ed. Owens’ float (%)
See Demerjian and Owens (2016) Website
N. Covenants Number of financial covenants in package Dealscan int
All-In Drawn Spread Average of each facility’s allindrawn Dealscan basis
in package weighted by facilityamt points
Loan Amount Ln(dealamount) Dealscan USD
Maturity Ln(average of each facility’s maturity Dealscan months
in package weighted by facilityamt)
Lenders Ln(N. syndicate members) Dealscan int
Revolver Fraction Revolver credit amount in package / dealamount Dealscan float
(0-1)
Previous Rel. Described in Section 2.3.2 Dealscan int
(0/1)
33
Table 1. Variable Definitions
Rel. Intensity (Amt) Described in Section 2.3.2 Dealscan float
(0-1)
Rel. Intensity (Num) Described in Section 2.3.2 Dealscan float
(0-1)
Rel. Length Described in Section 2.3.2 Dealscan years
Market Share Described in Section 2.3.2 Dealscan float (%)
Same State Described in Section 2.3.2 Compustat/ int
SEC (0/1)
Defaults N. outstanding loans to firms with credit rating Dealscan/ int
changed to "D"/"SD" over prev. qtr Capital IQ
Defaults (Same) N. defaults in bank’s portfolio in same industry Dealscan/ int
as the borrower’s Capital IQ
Defaults (Other) N. defaults in bank’s portfolio in different industry Dealscan/ int
from the borrower’s Capital IQ
Bank Assets atq Compustat USD Mil
Deposits (dptcq/atq)×100 Compustat float (%)
Book Equity (ceqq/atq)×100 Compustat float (%)
Market Equity (prccq×cshoq)/(atq − ceqq + prccq×cshoq) Compustat float
Tier 1 Capital capr1q Compustat float (%)
Non-Performing Assets (npatq/atq)×100 Compustat float (%)
Profitability (niq/atq)×100 Compustat float (%)
34
Table 2. Descriptive Statistics
This table reports the descriptive statistics for the full matched Dealscan-Compustat sample obtained after applying
the selection criteria described in Section 2.1, and for the Strictness sample, which further restricts the sample to
observations with non-missing covenant strictness and all-in drawn spread. All variables are described in Table 1.
MATCHED SAMPLE STRICTNESS SAMPLE
Mean Std. Dev. Obs. Mean Std. Dev. Obs.
Loan Characteristics
Covenant Strictness 35.86 41.19 12, 124 35.85 41.17 11, 684
N. Covenants 2.46 1.13 14, 483 2.47 1.14 11, 684
All-In Drawn Spread 190.93 128.36 22, 724 188.22 118.43 11, 684
Loan Amount ($M) 620.64 1, 444.92 23, 164 567.59 1, 201.17 11, 684
Maturity (Months) 45.45 22.25 22, 266 46.02 19.99 11, 628
N. Lenders 7.98 8.72 23, 164 9.04 9.36 11, 684
Revolver Fraction 0.74 0.38 23, 164 0.79 0.34 11, 684
Previous Rel. 0.69 0.46 18, 167 0.70 0.46 9, 127
mean sd Obs mean sd Obs
Firm Characteristics
Ln(Assets) 7.02 1.92 21, 416 6.80 1.82 11, 231
EDF 0.05 0.17 19, 275 0.06 0.17 10, 209
Tangibility 0.20 0.30 13, 857 0.20 0.29 7, 200
Leverage 0.30 0.21 20, 656 0.31 0.20 10, 879
Current Ratio 1.91 1.25 20, 626 1.87 1.15 10, 885
Ln(1+Int. Cover. Ratio) 2.23 1.14 16, 432 2.20 1.05 8, 999
Years since IPO 20.18 16.79 20, 732 19.62 16.11 10, 932
Rated 0.46 0.50 21, 417 0.45 0.50 11, 231
Rating 4.50 1.18 9, 829 4.62 1.05 5, 064
N. Loans 9.03 6.89 21, 417 8.81 6.41 11, 231
mean sd Obs mean sd Obs
Bank Characteristics
Ln(Bank Assets) 12.39 1.58 2, 723 12.39 1.57 2, 093
Deposits 61.93 12.84 2, 069 61.47 12.66 1, 651
Book Equity 7.26 2.89 2, 673 7.35 2.75 2, 062
Market Equity 12.10 6.52 2, 503 12.58 6.52 1, 944
Tier 1 Capital 9.71 2.17 1, 935 9.50 2.13 1, 574
Non-Performing Assets 0.65 0.52 1, 754 0.63 0.49 1, 441
Profitability 0.25 0.17 2, 072 0.26 0.18 1, 654
35
Table 3. Univariate Evidence on Loan Contracts and Bank-Firm Selection
This table reports the results of univariate t-tests, meant to document systematic differences in loan, firm, and bank-
level observables between loans made by a specialized bank within its sector of specialization and out of its sector
of specialization. For each variable X listed in the table, H
0
is that E[X (Specialized) − X (Non-Specialized)] = 0.
Spec. Non Spec. Diff. t-Stat N (Spec.) N (Non Spec.)
Loan Characteristics
Covenant Strictness 40.23 35.06 5.17 3.30 733 11,484
N. Covenants 2.53 2.44 0.09 2.44 938 13, 616
All-In Drawn Spread 231.28 189.62 41.66 11.72 1, 416 21, 813
Loan Amount ($M) 666.63 668.65 −2.03 −0.05 1, 448 22, 237
Maturity (Months) 40.35 46.16 −5.80 −9.34 1, 391 21,394
N. Lenders 6.25 8.24 −1.99 −8.38 1, 448 22, 237
Revolver Fraction 0.69 0.74 −0.04 −4.11 1, 448 22, 237
Previus Rel. 0.67 0.68 −0.02 −1.15 1, 028 17, 668
Firm Characteristics
Ln(Assets) 6.37 7.24 −0.88 −16.65 1, 448 22, 236
EDF 0.07 0.05 0.02 3.24 1, 266 20, 107
Tangibility 0.21 0.19 0.03 2.56 833 14, 856
Leverage 0.28 0.31 −0.03 −5.14 1, 372 21, 505
Current Ratio 2.23 1.85 0.38 11.13 1, 403 21, 397
Ln(1+Int. Cover. Ratio) 2.19 2.21 −0.02 −0.64 888 17, 452
Years since IPO 16.88 21.09 −4.21 −8.92 1, 404 21, 512
Rated 0.36 0.49 −0.13 −9.61 1, 448 22, 237
Rating 4.58 4.49 0.09 1.75 524 10, 938
Bank Characteristics
Ln(Assets) 11.3 13.3 −1.96 −3.33 1, 375 21, 548
Deposits 68.1 55.3 12.7 3.13 1, 105 18, 391
Book Equity 8.10 7.65 0.45 1.01 1, 352 21, 357
Market Equity 14.9 12.1 2.83 2.44 1, 268 20, 743
Tier 1 Capital 10.0 9.36 0.69 0.92 1, 059 18, 005
Non-Performing Asset 0.63 0.62 0.012 0.15 997 17, 151
Profitability 0.29 0.25 0.039 2.29 1, 104 18, 351
36
Table 4. The Effect of Bank Specialization on Covenant Strictness and Loan Spread
This table reports the estimates of the coefficients from the following regression using our baseline sample, which
includes loans for which the loan covenant strictness and all-in drawn spread are available:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which
Loan Term
f ,b,t
is either the covenant strictness (columns 1 to 3) or the all-in drawn spread (columns 4
to 6) for a loan originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects,
and FEs include, depending on the specification, firm fixed effects, firm rating fixed effects, loan purpose fixed
effects.
X
f ,b,t
represents firm and loan controls. All the variables are defined in Table 1. In parentheses, t statistics
obtained from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the
1%, 5% and 10%, respectively.
COVENANT STRICTNESS ALL-IN DRAWN SPREAD
(1) (2) (3) (4) (5) (6)
Specialization −12.4
∗∗
−23.56
∗∗∗
−24.35
∗∗∗
−14.19 −28.45 −31.77
∗
(−2.51) (−3.16) (−3.38) (−1.25) (−1.45) (−1.78)
Ln(Assets) 3.498
∗∗
4.313
∗∗
−19.48
∗∗∗
−14.71
∗∗∗
(2.11) (2.46) (−4.79) (−3.37)
EDF 18.7
∗∗∗
19.7
∗∗∗
134.9
∗∗∗
127
∗∗∗
(3.72) (4.23) (5.32) (4.99)
Tangibility 24.69
∗∗∗
27.05
∗∗∗
−37
∗∗
−33.23
∗∗
(4.25) (4.64) (−2.26) (−2.04)
Leverage 32.16
∗∗∗
34.02
∗∗∗
30.91
∗∗
15.18
(4.41) (4.77) (2.17) (1.03)
Current Ratio −3.688
∗∗∗
−3.713
∗∗∗
3.055
∗
1.015
(−2.77) (−2.99) (1.70) (0.66)
Ln(1+Int. Cover. Ratio) −13.62
∗∗∗
−13.53
∗∗∗
−13.13
∗∗∗
−15.03
∗∗∗
(−9.32) (−9.47) (−5.86) (−8.57)
Years since IPO −29.36 −10.48 132.4
∗∗
138.5
∗∗
(−0.57) (−0.18) (2.37) (2.17)
Ln(Loan Maturity) .6531 11.44
∗∗∗
(0.63) (3.75)
Ln(Lenders) −.735 −6.604
∗∗∗
(−0.85) (−3.08)
Ln(Loan Amount) −1.098
∗
1.049
(−1.75) (0.35)
Revolver Fraction 1.818
∗
−52.63
∗∗∗
(1.93) (−6.95)
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes
Firm FE − Yes Yes − Yes Yes
Rating FE − Yes Yes − Yes Yes
Loan Purpose FE − − Yes − − Yes
Adj. R
2
.074 .565 .57 .278 .749 .779
Observations 9, 834 4, 653 4, 643 9, 834 4, 653 4, 643
37
Table 5. Bank Specialization and Loan Terms, Accounting for Bank-Firm Lending Relationships
This table reports the estimates of the coefficients from the following regression using our baseline sample, which
includes loans for which the loan covenant strictness and all-in drawn spread are available:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization
f ,b,t−1
+ β
R
· REL
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which Loan Term
f ,b,t
is either the covenant strictness (Panel A) or the all-in drawn spread (Panel B) for a loan
originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects, and FEs include,
depending on the specification, firm fixed effects, firm rating fixed effects, loan purpose fixed effects.
RE L
represent
the relationship lending variable(s), and
X
f ,b,t
represents firm and loan controls. All the variables are defined in
Table 1. In parentheses, t statistics obtained from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the 1%, 5% and 10%, respectively.
(1) (2) (3) (4) (5) (6) (7)
Panel A: COVENANT STRICTNESS
Specialization −24.46
∗∗∗
−24.39
∗∗∗
−24.31
∗∗∗
−24.29
∗∗∗
−24.47
∗∗∗
−24.41
∗∗∗
−24.37
∗∗∗
(−3.41) (−3.38) (−3.37) (−3.37) (−3.40) (−3.39) (−3.38)
Rel. Length .107 .104 .123 .126
(0.73) (0.70) (0.79) (0.81)
Previous Rel. .259 .088
(0.26) (0.089)
Rel. Intensity (Amt) −.306 −.523
(−0.30) (−0.48)
Rel. Intensity (Num) −.401 −.620
(−0.37) (−0.56)
Panel B: ALL-IN DRAWN SPREAD
Specialization −31.22
∗
−31.28
∗
−31.38
∗
−31.3
∗
−30.92
∗
−31
∗
−30.96
∗
(−1.76) (−1.75) (−1.75) (−1.74) (−1.73) (−1.74) (−1.73)
Rel. Length −.581
∗
−.496 −.509
∗
−.523
∗
(−1.98) (−1.63) (−1.69) (−1.70)
Previous Rel. −3.40
∗∗
−2.58
(−2.20) (−1.65)
Rel. Intensity (Amt) −3.12 −2.23
(−1.65) (−1.19)
Rel. Intensity (Num) −2.80 −1.89
(−1.43) (−0.95)
Firm Controls Yes Yes Yes Yes Yes Yes Yes
Loan Controls Yes Yes Yes Yes Yes Yes Yes
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes Yes
Rating FE Yes Yes Yes Yes Yes Yes Yes
Loan Purpose FE Yes Yes Yes Yes Yes Yes Yes
Observations 4, 643 4, 643 4, 643 4, 643 4, 643 4, 643 4, 643
38
Table 6. Bank Specialization and Loan Terms, Accounting for Bank Industry Market Share
This table reports the estimates of the coefficients from the following regression using our baseline sample, which
includes loans for which the loan covenant strictness and all-in drawn spread are available:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization
f ,b,t−1
+ β
M
· Mkt Share
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which
Loan Term
f ,b,t
is either the covenant strictness (columns 1 to 3) or the all-in drawn spread (columns 4
to 6) for a loan originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects,
and FEs include, depending on the specification, firm fixed effects, firm rating fixed effects, loan purpose fixed
effects.
X
f ,b,t
represents firm and loan controls. All the variables are defined in Table 1. In parentheses, t statistics
obtained from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the
1%, 5% and 10%, respectively.
COVENANT STRICTNESS ALL-IN DRAWN SPREAD
(1) (2) (3) (4) (5) (6)
Specialization −11.39
∗∗
−24.34
∗∗∗
−25.1
∗∗∗
−8.503 −30.11 −34.16
∗
(−2.20) (−3.30) (−3.52) (−0.78) (−1.52) (−1.90)
Market Share −8.375 8.45 8.337 −47.17 18.04
∗
26.6
∗∗∗
(−0.87) (1.19) (1.25) (−1.29) (1.73) (2.74)
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes
Firm FE − Yes Yes − Yes Yes
Rating FE − Yes Yes − Yes Yes
Firm Controls − Yes Yes − Yes Yes
Loan Purpose FE − − Yes − − Yes
Loan Controls − − Yes − − Yes
Adj. R
2
.074 .565 .57 .279 .749 .779
Observations 9, 834 4, 653 4, 643 9, 834 4,653 4, 643
39
Table 7. Bank Specialization and Loan Terms, Accounting for Bank-Firm Geographical Proximity
This table reports the estimates of the coefficients from the following regression using our baseline sample, which
includes loans for which the loan covenant strictness and all-in drawn spread are available:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization
f ,b,t−1
+ β
S
· Same State
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which
Loan Term
f ,b,t
is either the covenant strictness (columns 1 to 3) or the all-in drawn spread (columns 4
to 6) for a loan originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects,
and FEs include, depending on the specification, firm fixed effects, firm rating fixed effects, loan purpose fixed
effects.
X
f ,b,t
represents firm and loan controls. All the variables are defined in Table 1. In parentheses, t statistics
obtained from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the
1%, 5% and 10%, respectively.
COVENANT STRICTNESS ALL-IN DRAWN SPREAD
(1) (2) (3) (4) (5) (6)
Specialization −13.72
∗∗
−23.35
∗∗
−24.16
∗∗
−8.307 −37.37
∗
−38.8
∗
(−2.32) (−2.43) (−2.76) (−0.67) (−1.73) (−1.85)
Same State −.0431 −.6604 −.8592 −8.508 −7.705 −7.132
(−0.021) (−0.20) (−0.25) (−1.13) (−1.08) (−1.07)
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes
Firm FE − Yes Yes − Yes Yes
Rating FE − Yes Yes − Yes Yes
Firm Controls − Yes Yes − Yes Yes
Loan Purpose FE − − Yes − − Yes
Loan Controls − − Yes − − Yes
Adj. R
2
.072 .576 .58 .268 .755 .781
Observations 9, 072 4, 267 4, 259 9, 072 4, 267 4, 259
40
Table 8. Bank Specialization and Loan Terms After Defaults on Banks’ Loan Portfolio
This table reports the estimates of the coefficients from the following regression using our baseline sample, which
includes loans for which the loan covenant strictness and all-in drawn spread are available:
Loan Term
f ,b,t
=
θ
b,t
+
FEs
+
βSpecialization
f ,b,t−1
+
β
D
DEF
b,t−1
+
δSpecialization
f ,b,t−1
∗ DEF
b,t−1
+
γ· X
f ,b,t
+
ϵ
f ,b,t
in which
Loan Term
f ,b,t
is the covenant strictness for a loan originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects, and FEs include, depending on the specification, firm fixed effects,
firm rating fixed effects, loan purpose fixed effects.
X
f ,b,t
represents firm and loan controls. All the variables are
defined in Table 1. In parentheses, t statistics obtained from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the 1%, 5% and 10%, respectively.
COVENANT STRICTNESS
(1) (2) (3) (4) (5) (6)
Specialization −24.36
∗∗∗
−25.3
∗∗∗
−25.29
∗∗∗
−22.73
∗∗∗
−23.6
∗∗∗
−23.59
∗∗∗
(−3.38) (−3.25) (−3.24) (−2.98) (−2.79) (−2.79)
Specialization * Defaults (Same) 30.28
∗∗
30.35
∗∗
30.71
∗∗
30.76
∗∗
(2.28) (2.28) (2.34) (2.34)
Specialization * Defaults (Other) 2.073 2.075 1.837 1.839
(0.40) (0.40) (0.34) (0.34)
Defaults (Same) .8011 .8048 .7885 .7921
(1.02) (1.02) (1.01) (1.01)
Defaults (Other) −.824 −.8116
(−1.05) (−1.04)
Industry Portfolio Share −9.782 −9.547 −9.562
(−0.94) (−0.89) (−0.89)
Firm Controls Yes Yes Yes Yes Yes Yes
Loan Controls Yes Yes Yes Yes Yes Yes
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
Rating FE Yes Yes Yes Yes Yes Yes
Loan Purpose FE Yes Yes Yes Yes Yes Yes
Adj. R
2
.57 .57 .57 .57 .57 .57
Observations 4, 643 4, 643 4, 643 4, 643 4, 643 4, 643
41
Table 9. Robustness: Sample Restricted to Loans with Single Lead Arranger
This table reports the estimates of the coefficients from the following regression using the baseline sample, further
restricted to include only loans with a single-lead arranger:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which
Loan Term
f ,b,t
is either the covenant strictness (columns 1 to 3) or the all-in drawn spread (columns 4
to 6) for a loan originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects,
and FEs include, depending on the specification, firm fixed effects, firm rating fixed effects, loan purpose fixed
effects.
X
f ,b,t
represents firm and loan controls. All the variables are defined in Table 1. In parentheses, t statistics
obtained from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the
1%, 5% and 10%, respectively.
COVENANT STRICTNESS ALL-IN DRAWN SPREAD
(1) (2) (3) (4) (5) (6)
Specialization −11.67
∗∗
−17.32
∗∗
−18.59
∗∗
−15.53 −34.28 −37.66
∗
(−2.23) (−2.05) (−2.30) (−1.28) (−1.46) (−1.76)
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes
Firm FE − Yes Yes − Yes Yes
Rating FE − Yes Yes − Yes Yes
Firm Controls − Yes Yes − Yes Yes
Loan Purpose FE − − Yes − − Yes
Loan Controls − − Yes − − Yes
Adj. R
2
.071 .564 .569 .28 .751 .778
Observations 9, 370 4, 359 4, 348 9, 370 4,359 4, 348
42
Table 10. Robustness: Bank Specialization Measure Calculated Using Data from 1996
This table reports the estimates of the coefficients from the following regression using our baseline sample:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization (96)
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which
Loan Term
f ,b,t
is either the covenant strictness (columns 1 to 3) or the all-in drawn spread (columns 4
to 6) for a loan originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects,
and FEs include, depending on the specification, firm fixed effects, firm rating fixed effects, loan purpose fixed
effects.
X
f ,b,t
represents firm and loan controls. All the variables are defined in Table 1. In parentheses, t statistics
obtained from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the
1%, 5% and 10%, respectively.
COVENANT STRICTNESS ALL-IN DRAWN SPREAD
(1) (2) (3) (4) (5) (6)
Specialization (96) −11.9
∗∗
−23.56
∗∗∗
−24.33
∗∗∗
−8.043 −28.45 −31.76
∗
(−2.52) (−3.16) (−3.38) (−0.59) (−1.45) (−1.78)
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes
Firm FE − Yes Yes − Yes Yes
Rating FE − Yes Yes − Yes Yes
Firm Controls − Yes Yes − Yes Yes
Loan Purpose FE − − Yes − − Yes
Loan Controls − − Yes − − Yes
Adj. R
2
.081 .566 .571 .281 .75 .779
Observations 8, 990 4, 648 4, 638 8, 990 4, 648 4, 638
43
Table 11. Robustness: Sample Restricted to Loans Originated Before 2008
This table reports the estimates of the coefficients from the following regression using the baseline sample, further
restricted to include only loans originated before (excluding) 2008:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which
Loan Term
f ,b,t
is either the covenant strictness (Panel A) or the all-in drawn spread (Panel B) for a
loan originated in year-quarter
t
by bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects, and FEs
include, depending on the specification, firm fixed effects, firm rating fixed effects, loan purpose fixed effects.
X
f ,b,t
represents firm and loan controls. All the variables are defined in Table 1. In parentheses, t statistics obtained
from two-way clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the 1%, 5%
and 10%, respectively.
COVENANT STRICTNESS ALL-IN DRAWN SPREAD
(1) (2) (3) (4) (5) (6)
Specialization −14.6
∗∗
−30.15
∗∗∗
−32.83
∗∗∗
−17.29 −16.57 −17.77
(−2.54) (−2.86) (−2.87) (−1.33) (−0.48) (−0.59)
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes
Firm FE − Yes Yes − Yes Yes
Rating FE − Yes Yes − Yes Yes
Firm Controls − Yes Yes − Yes Yes
Loan Purpose FE − − Yes − − Yes
Loan Controls − − Yes − − Yes
Adj. R
2
.051 .536 .537 .234 .739 .773
Observations 6, 832 2, 089 2, 085 6, 832 2, 089 2, 085
44
Table 12. Robustness: Bank Specialization Measure Computed Over Different Time Windows
This table reports the estimates of the coefficients on the Specialization variable—averaged over different time windows—from the following regression using
our baseline sample:
Loan Term
f ,b,t
= θ
b,t
+ FEs + β · Specialization (nY)
f ,b,t−1
+ γ · X
f ,b,t
+ ϵ
f ,b,t
in which
Loan Term
f ,b,t
is either the covenant strictness (columns 1 to 3) or the all-in drawn spread (columns 4 to 6) for a loan originated in year-quarter
t
by
bank
b
to firm
f
.
θ
b,t
represents bank
×
year-quarter fixed effects, and FEs include, depending on the specification, firm fixed effects, firm rating fixed effects,
loan purpose fixed effects.
X
f ,b,t
represents firm and loan controls. All the variables are defined in Table 1. In parentheses, t statistics obtained from two-way
clustering at the bank and borrower level.
∗∗∗
,
∗∗
,
∗
indicate statistical significance at the 1%, 5% and 10%, respectively.
COVENANT STRICTNESS ALL-IN DRAWN SPREAD
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Specialization (1Y) −8.303 −33.57
∗∗
(−1.14) (−2.64)
Specialization (2Y) −15.93
∗
−37.91
∗∗
(−1.92) (−2.39)
Specialization (3Y) −24.35
∗∗∗
−31.77
∗
(−3.38) (−1.78)
Specialization (4Y) −23.04
∗∗∗
−29.41
(−2.76) (−1.42)
Specialization (5Y) −20.32
∗∗
−16.33
(−2.25) (−0.72)
Firm Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Loan Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Bank×YearQtr FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Rating FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Loan Purpose FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj. R
2
.569 .568 .57 .572 .57 .779 .779 .779 .777 .776
Observations 4, 767 4, 701 4, 643 4, 576 4, 455 4, 767 4, 701 4, 643 4, 576 4, 455
45
Appendix B. Figures
Figure 1. Simple Examples to Understand Bank Specialization at the Industry Level
This figure illustrates the notion of bank specialization in an industry according to the definition by Paravisini et al.
(2022) with simple examples from two-bank, two-sector lending markets. From the top left, we can see: (a) an
example of no specialized banks; (b) an example of specialized banks – Bank 1 in sector A and Bank 2 in sector B;
(c) a case in which no bank is specialized because both banks allocate the same portfolio shares to both sectors;
(d) a case in which both banks are specialized – Bank 1 in lending to sector A and Bank 2 in lending to sector B.
(a) Neither bank is specialized
(b) Both banks are specialized (1-A, 2-B)
(c) Neither bank is specialized
(d) Both banks are specialized (1-A, 2-B)
46
Figure 2. Comparison Between Portfolio Concentration of the Average Bank and the “Market”
This figure plots on the y-axis the HHI measure of loan portfolio concentration, and on the x-axis the year at which
it is recorded. HHI is computed for the Market (blue) and Average Bank (green) portfolios per each year-quarter.
A higher value of HHI implies that lending to sectors is more concentrated in the market/average bank’s portfolio.
The fact that the average bank is systematically characterized by a higher HHI compared to the market shows
graphically that the average lender in the syndicated loan market remained overall more concentrated than the
whole syndicated market over 1996-2016.
47
Figure 3. Specialization Is Common Across Industries and Time
This figure presents evidence of specialization in lending towards specific industries in four different moments
from our sample: 2000q2, 2005q2, 2010q2, 2015q2. Each subfigure reports the box-plot graph, for each of the
25 TFIC industries, of the distribution of banks’ demeaned loan portfolio shares in a given industry. Each dot
represents an outlier, and therefore, a banks specialized in that industry.
2000
-.5 0 .5 1
(Centered) Banks' Industry Portfolio Shares
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
TFIC Industry
2005
-.5 0 .5 1
(Centered) Banks' Industry Portfolio Shares
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
TFIC Industry
2010
-.5 0 .5 1
(Centered) Banks' Industry Portfolio Shares
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
TFIC Industry
2015
-.5 0 .5 1
(Centered) Banks' Industry Portfolio Shares
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
TFIC Industry
48
Figure 4. Specialization Is Persistent Over Time
This figure plots the n-year autocorrelation of the specialization dummy, averaged at the bank-year-sector level,
where n takes value from 1 to 10.
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
1 2 3 4 5 6 7 8 9 10
Autocorrelation
Years
49
