Fed: Specialization in Banking

NO.
967
Specialization in
MAY 2021
Banking
Kristian Blickle | Cecilia Parlatore | Anthony Saunders

Specialization in Banking
Kristian Blickle, Cecilia Parlatore, and Anthony Saunders
Federal Reserve Bank of New York Staff Reports, no. 967
May 2021
JEL classification: D4, G20, G21
Abstract
Using highly detailed data on the loan portfolios of large U.S. banks, we document that these banks
"specialize" by concentrating their lending disproportionately into one industry. This specialization
improves a bank’s industry-specific knowledge and allows it to offer generous loan terms to borrowers,
especially to firms with access to alternate sources of funding and during periods of greater nonbank
lending. Superior industry-specific knowledge is further reflected in better loan and, ultimately, bank
performance. Banks concentrate more on their primary industry in times of instability and relatively
lower Tier 1 capital. Finally, specialization counteracts a well-documented trend in reduced lending by
large banks to opaque small and medium-sized enterprises.
Key words: bank specialization, bank concentration, asymmetric information, loan performance, bank
performance
_________________
Blickle: Federal Reserve Bank of New York (email: kristian.blickle@ny.frb.org). Parlatore: New York
University and NBER (email: cparlato@stern.nyu.edu). Saunders: New York University (email:
asaunder@stern.nyu.edu). The authors thank Tobias Berg, Quirin Fleckenstein, Sebastian Hillenbrand,
Stephan Luck, and Philipp Schnabl for helpful comments and suggestions.
This paper presents preliminary findings and is being distributed to economists and other interested
readers solely to stimulate discussion and elicit comments. The views expressed in this paper are those of
the author(s) and do not necessarily reflect the position of the Federal Reserve Bank of New York or the
Federal Reserve System. Any errors or omissions are the responsibility of the author(s).
To view the authors’ disclosure statements, visit

https://www.newyorkfed.org/research/staff_reports/sr967.html.
1 Introduction
Banks are traditionally tasked with selecting high quality borrowers and monitoring their
adherence to loan covenants. However, borrower selection and loan monitoring require the
costly acquisition of information. Economies of scale and deeper experience may be built up
through the specialization of lending to certain industries. After all, repeated interactions with
individual borrowers has been shown to improve a bank’s knowledge of a borrower.1 In a
similar vein, repeated lending in a specific industry can enable banks to better evaluate the
business models or collateral of borrowers in that industry.2
In this paper, we use detailed data on the loan portfolios of stress tested banks in the United
States to show that even large banks specialize in lending to specific industries. The average
bank in our sample directs around 20% of its total C&I lending into a single "top" industry –
far more than into any other industry. This specialization is correlated with more favourable
loan terms for borrowers; as measured by loan size, rate spread, and maturity. This holds
especially in times when competition, for instance from non-bank entities, is higher and in
lending to firms with access to alternate sources of funding. Specialization by banks may in
fact be a partly necessary answer to increased loan competition.
We argue that these generous loan terms are in part made possible through the superior
information acquisition, expertise, or monitoring ability of specialized banks. We perform four
broad tests to highlight the existence of this advantage. Firstly, and perhaps most importantly,
we find better ex-post loan performance of loans issued by bank’s specialized in the borrower’s
industry. This holds even when controlling for the ex-ante bank-internal loan risk rating that
is unobservable to the market. Secondly, our results on the superior performance of loans
originated by specialized banks are more pronounced in non-syndicated lending. A specialized
bank is likely most incentivized to leverage its superior information in instances where the
free-rider problem of other market participants is lowest. In a similar vein, we thirdly find that
1 For a discussion of relationship lending see: Bernanke (1983), James (1987), Petersen and Rajan (1995), Berger
and Udell (1995), or Degryse and Ongena (2005).

2 Specialization in industries has been discussed by Paravisini et al. (2020), specialization in small bank business
models has been discussed in Blickle (2018), and specialization in in bank collateral has been discussed in Gopal
(2019).
1
banks are less likely to divest themselves fully of exposure to firms operating in industries in
which they are more specialized.3 This follows from the fact that the adverse selection discount
due to the asymmetric information might be large for loans a specialized bank can judge better
than the market. Finally, despite an aggregate decrease in bank lending to SMEs in recent
years, specialized banks are more willing to lend to smaller – and thereby more opaque – firms
in their industry of specialization. This may reflect that small borrower opacity is reduced
through bank specialization.
It is, of course, possible that specialization, as we capture it, may simply be the by-product
of a bank attempting to capture an industry by crowding out other banks so as to eventually
extract monopoly rents. We find that an increase in specialization allows banks to better capture
a large share of an industry in the following period. However, such an attempt to capture an
industry would likely lead to worse, as opposed to better, loan performance. Moreover, we
look at industry capture separately from specialization. We show that while specialization
is correlated with industry capture, our results on specialization hold despite of – and not
because of – a bank holding a dominant share of an industry.
A natural question that arises in the context of specialization is whether concentrated banks,
offering generous loan terms, suffer worse performance and greater aggregate risk. On the one
hand, a diversified portfolio reduces a banks’ exposure to local shocks maximizing risk-sharing
and minimizing the risk of runs 4 . On the other hand, however, diversification increases the
correlation across banks’ portfolios and the probability of systemic crises 5 . During the period
of our sample, which covers 2011q3-2020q2, we find that specialized banks earn more stable,
though slightly lower, returns and charge-off fewer loans on aggregate. Loans originated
by banks specialized in an industry even perform better in the first half of 2020, which saw
significant economic disruptions due to the spread of COVID-19. Finally, specialized banks
show on average higher levels of Tier1-capital. Banks gravitate toward increased specialization
in times of lower Tier-1 capital ratios and expand into other industries, with which they are
3 Blickleet al. (2020) and Bord and Santos (2015) document a propensity of lead arrangers to fully sell out of
loans they originate.
4 See Diamond and Dybvig (1983) and Allen and Gale (1998).
5 For a discussion see: Haldane (2009), Haldane and May (2011), Yellen (2013), Goldstein et al. (2020).
2
much less familiar, in times of relatively high Tier-1 capital; reflecting the perceived safety and
stability of specialization. This is consistent with specialization being largely driven by banks
having better information, which allows them to reduce their risks.
The analyses in this paper are based on the FR Y-14 Q archive, which tracks all C&I loans
over 1 million USD in size for all stress tested US banks. Our data is the closest thing to a
credit registry that exists in the United States and encompasses >75% of corporate lending.
Unfortunately, we observe only originated loans and not loan applications. As such, our
regressions reflect ex-post equilibrium outcomes. We are careful about interpreting our results
as causal. Nevertheless, we can account for a host of loan and bank characteristics in all
regressions. We can also account for firm-bank interactions (i.e. past relationships). This
ensures the effects we are measuring are not driven by specific firm-related knowledge as
opposed to a wider industry-level specialization.
A valid concern with the interpretation of our results may be that unobserved bank or firm
characteristics drive not only the decision of a borrower to engage with a certain bank, but
also the observable loan terms. Most crucially, our analysis assumes a supply-driven effect
and a relatively constant loan demand. Given the detail of our data, we are able to account
for firm, time, and bank fixed effects. Our effect of interest is therefore identified within a
borrower with multiple loans, in a given period, and within a given bank. We are even able to
include specifications with a Khwaja and Mian (2008)-style control, ultimately holding constant
firm-year effects. In such specifications the effect is identified in firms obtaining two loans
from two differently specialized banks in the same period. We include a number of other
specifications to attempt to assuage concerns that the effects are demand- rather than supply
driven.
Where possible, we attempt to address further concerns about potential loan-level differ-
ences. For instance, we are able to account for loan purpose, loan type, and even risk ratings.
These risk ratings are standardized across all banks in the sample as part of the stress testing
process and therefore highly comparable across the loans in our sample. Even in the face of
detailed controls, endogeneity concerns about borrower-bank selection remain. We address
3
these openly throughout the paper and attempt a variety of robustness analyses to highlight
the stability of our results.
Our work contributes to a growing literature on the importance of bank business model
specialization. Given the size of their balance sheets, banks may acquire significant amounts of
knowledge about an industry through specialization. The natural incentives to specialize and
possibly capture an industry must be weighed against possible risks, This paper analyzes, loan,
bank, and industry performance to attempt a better quantification of such risks.
2 Literature Review
This paper is related to the growing empirical literature on bank specialization. Acharya et al.
(2006) find that bank diversification is not associated with superior returns or safer portfolios.
Saidi and Streitz (2020) show that a bank’s concentration affects the non-financial sector to
which it lends. They show that concentrated lenders charge lower cost of debt for firms
competing with substitute products. Tabak et al. (2011) document better bank performance and
lower risk in more diversified banks in Brazil. Paravisini et al. (2020) develop an approach to
identify bank specialization in lending and show that Peruvian banks specialize across export
markets. This specialization, they show, has real economic and business-level effects for their
borrowers.
Our paper also contributes to the wider literature on banking and bank business models
in general. Broadly speaking, there are two main theories of banking.6 On one hand, banks
provide liquidity and maturity transformation to their depositors by issuing demandable
deposits and investing in longer term loans. Under this view, risk averse banks will choose to
diversify their loan portfolio to maximize risk-sharing among their depositors and minimize
the risk of bank-runs (see Diamond and Dybvig (1983) and Allen and Gale (1998)). On the other
hand, banks may have an informational advantage in lending, by screening and monitoring
loans (Diamond (1984)). In this context, increasing returns to information acquisition push
banks towards holding specialized loan portfolios. Understanding how specialization impacts
6 See Battacharya and Thakor (1993) for a survey of the theories of financial intermediation.
4
bank riskiness is therefore key to understanding the interplay between risk related capital
regulation and such specialization.
By monitoring loans and learning about a specific sector, banks can significantly increase
their informational advantage relative to less informed investors or banks. This monopoly on
information can translate into market power in the loan market and lead to a hold-up problem,
even in a competitive banking industry (see: Broecker (1990), Sharpe (1990) and Rajan (1992),
Riordan (1993), and von Thadden 1998). Alternatively, specialization may be a byproduct of
banks’ search for monopoly power and lead to anti-competitive behavior solely in search for
rents. Which of these two forces is the main driver of bank specialization again has important
implication on the regulation front.
3 Methodology
In this paper, we define a bank’s industry specialization as the share of all loans "l" that bank
"b" has invested in industry/sector "s" at time "t". Formally, a bank’s specialization in industry
∑l ∈s LoanAmtb,t,l
s is given by ∑s ∑l ∈s LoanAmtb,t,l
, where the sum in the numerator is over all l made to borrowers
in industry s at time t and the denominator represents all loans made by banks b at time t. This
definition implies that the greater the share of a bank’s portfolio that is invested in a single
industry, the more specialized the bank is in that industry. The benefits of this definition of
specialization are twofold. Firstly, it provides intuitive coefficients. Secondly, it does not rely
on ranking industries within a bank’s portfolio or across banks. A bank may be specialized in
more than one industry or in no industry if it is highly diversified. Similarly, multiple banks
could be specialized in the same industry, as we are not ranking banks relative to one another.
We showcase the robustness of our results by using alternative definitions of specialization
in the Appendix. We provide results measuring specialization as the share of loans – by count–
that are invested in a single industry, the share of logged loans in an industry, and binary
variables denoting a bank’s "most preferred" industry. These measures are less likely to be
driven by massive individual loans to a single firm, while still capturing a bank’s increased
5
exposure to a particular industry. All our main results remain valid using these alternative
definitions.
Our main hypothesis is that, driven by experience and incentives, a bank becomes more
knowledgeable about an industry as the share of funds it has invested in it grows. This
knowledge can translate to better screening and monitoring abilities. We argue that this is the
case by looking at how the loan terms offered by a bank, bank performance, and a bank’s SME
lending correlate with the bank’s specialization in that industry.
3.1 Specialization and Loan Characteristics
Our first set of results relate the degree of a bank’s specialization to loan characteristics. Loan
terms offered by specialized banks may differ from those offered by non-specialized lenders
for three reasons. Firstly, if a bank becomes more knowledgeable about an industry, it may
be better able to accurately judge the value of a project or business. This follows from a long
literature on the value of experience and relationship lending. A more accurate borrower
assessment may enable a bank to offer better loan terms to high quality borrowers at the cost
of driving sub-par borrowers to look for funds among less informed lenders. Secondly, banks
with superior industry knowledge may also be better at monitoring and guiding a borrower,
decreasing the risk of the loan, independently of its initial risk. This superior ability to ensure
borrower performance may also lead to bank specialized banks offering better loan conditions
than non-specialized ones. Finally, a bank may choose to specialize to capture an industry and
pursue monopolistic rents. Specialization, as measured by the share of lending in an industry,
may be a reflection of the bank’s attempts to crowd out other lenders. In this case, the banks
may offer beneficial loan terms to snare the largest number of borrowers, independently of the
knowledge the bank has about the industry. We address this last point separately below by
specifically analyzing and accounting for a bank’s position in an industry.
In our baseline regressions, we relate the characteristics of loan "l" that firm "i" receives
from bank "b" at time "t" on the degree to which "b" is specialized in the industry/sector "s" of
the borrower.
6
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xl,b
(1)
+ θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
The primary loan characteristics used as left-hand-side (LHS) variables are log loan amount,
the interest rate, remaining maturity, and whether a loan is secured or unsecured. We readily
acknowledge that all the loan characteristics are simultaneously determined. We therefore
use those loan characteristics that are not employed as the dependent variable as additional
independent regressors. That is to say that in regressions focusing on the correlation between
loan size and specialization, for example, we control for the loan’s interest rate, its time to
maturity, and whether it is secured. We observe only equilibrium outcomes and, as such, we
can measure only correlations as opposed to causal relationships. This is discussed in greater
detail below.
X is a vector of these loan characteristics as well as bank controls, such as bank size or tier 1
capital. It also includes relationship variables that capture the number of times the bank and
firm have interacted in our data. Previous relationships may, by themselves, build knowledge
about a borrower’s quality that is unrelated to industry specialization. We wish to disentangle
the direct effect of information a bank has on a single borrower from the experience a bank may
have in an entire industry. The regression further includes, bank, firm, year*quarter, industry,
loan-rating, and loan-purpose fixed effects. In our primary analyses, we use only loans at first
origination. This excludes any renegotiated loans, but allows for a cleaner discussion of the
effect in question.
Given the high number of controls and fixed effects, the regressions are extremely saturated.
These regressions highlights the differences in loan terms obtained by a given firm from two or
more banks that are differently specialized in its industry or from the same bank at different
points in time, independent of loan rating, purpose, or time effects.
Ultimately, we cannot observe an individual firm’s loan demand. To the extent that it is
either time-invariant or driven by aggregate trends, fluctuations in a firm’s loan demand would
be captured in our fixed effects. However, there may be firm-specific drivers of a firm’s demand
7
for loans. To control for these features, we includes a firm-year fixed effect in one of our
specifications. While adding firm-year fixed effects accounts for a share of a firm’s demand, it
also implies that our regression coefficients are identified only among firms obtaining multiple
loans from different banks in the same period. It is worth noting that this specific subset of
firms may not be fully representative.
We run a few additional specifications to further address the issue of supply- vs. demand-
driven results, as well as concerns that may arise by interacting the variable of interest,
specialization, with key bank, time, or loan characteristics. We begin by interacting special-
ization with a measure that tracks whether significant amounts of non-bank funding was
available. Competition by non-banks may drive specialization and loan terms. Information on
non-bank funding is taken from Fleckenstein et al. (2020). In a similar vein, we interact whether
a borrower had access to alternate sources of funding. This is most easily, if noisily, gauged by
whether a firm was publicly traded. Both of these interactions terms help distinguish between
demand and supply drivers of the effects we find.
Loan terms can also be driven by a bank’s availability of loanable funds. We explore how
capital requirements and bank specialization interact to determine loan characteristics by inter-
acting bank specialization with the bank’s Tier 1 capital. Finally, we interact specialization with
a dummy denoting whether the bank is the syndication agent in the loan. Bank specialization
should matter more for agents than for participants in a syndicated loan. After all, the agent is
tasked with loan selection and monitoring while the syndication participant has fewer direct
obligations (see Blickle et al. (2020) for a discussion).
Industry Capture
Banks with a high degree of specialization have a significant amount of capital invested in
a single industry and may capture a significant share of said industry. Many have documented
an increase in bank concentration over the past years (consider: Fernholz and Koch (2016) or
Laeven et al. (2016)). Therefore, specialization can be unrelated to a bank’s knowledge about
an industry and be driven by a bank’s rent-seeking behavior. Through decreased competition,
banks that capture an industry may be able to extract high rents from captive companies.
8
Moreover, capture may exacerbate asymmetric information problems and prevent the entry of
competitive new entrants (see Cetorelli and Strahan (2006) and Bikker and Haaf (2002)).
On the other hand, capture may be a by product of knowledge-driven bank specialization.
Banks with more knowledge about a certain industry, who are better suited to screening and
monitoring borrowers in that industry, may organically capture an industry by offering better
terms to borrowers and alleviating information asymmetries. Moreover, increasing returns to
knowledge and experience may lead a bank to increase its presence in a particular industry to
justify its initial investment in acquiring said expertise.
To test our hypothesis and gauge the effects of knowledge-driven specialization, divorced
from any rent seeking behavior by banks, we control for a bank’s industry capture. To do
so, we include a bank’s industry capture as additional variable in the baseline regressions
discussed above. The regressions take the following form.
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Captureb,s,t + β 3 Xl,b

(2)
We define capture as the share of the Herfindahl-Hirschman index (HHI) of industry "s"
that is accounted for by bank "b" at time t. HHI is a relatively common measure of competition.
As such, the degree to which a single bank affects the competitiveness of an industry is a
good measure of the degree to which it has captured that industry. An industry with only
one bank will be perfectly captured by that bank. Similar to our measure of specialization,
discussed above, the measure is continuous and bounded between 0 and 1, making it easy to
interpret. There naturally exists a high degree of correlation between industry capture and
specialization. However, given that these industries are defined at the two-digit NAICS code,
they are extremely large and therefore, even specialization of large banks is possible without
the automatic capture of an industry.
9
3.2 Specialization and Performance
Loan Performance A natural question that deserves to be tested is how specialization impacts
loan performance and, ultimately, the performance of specialized banks. If specialization is
associated with a greater ability to ex-ante select or ex-post monitor high quality loans, then
one would expect specialized banks to outperform their non-specialized counterparts. On the
other hand, the lower interest rates charged by specialized banks may negatively impact their
profitability. We perform a series of additional tests. Firstly, we regress loan performance on
the industry specialization of the bank that granted it, in the vein of the regressions discussed
above:
Per f ormancel,i,b,s,T = β 0 + β l Specializationb,s,t + β 2 Xl,b

(3)
Here "Performance" are a variety of measures that capture whether the loan was ever
declared non-performing or, in alternate measures, was downgraded during its term and the
rating at final observation. Non-performance occurs if a loan ever becomes 90-days past due
or if maturity becomes negative, as this implies the loan has not been resolved satisfactorily.
Ratings are standardized and a downgrade implies that the loan has become more risky in the
eyes of the lender, based on homogeneous metrics that are highly similar across lenders and
identical within lenders. We make use of only one observation per loan when it is first observed
in the data. We thereby allow the dependent variable to be extremely forward looking. If
specialized banks are better at discerning high quality loans or otherwise better at monitoring
the loan, one would expect the loan to be less likely to become risky or to default.
Bank Performance In a similar vein, we would expect banks with a high degree of special-
ization to have a safer more stable loan portfolio overall. Better performing loans should
translate to more stable returns and fewer write-downs. We test this hypothesis by collapsing
our specialization data at the bank-time level to identify the average specialization of a given
bank. We then combine this data with information from Y9-C data, which is maintained
10
by the Federal Reserve System and tracks key balance sheet information of individual bank
holding companies. This combined data allows us to relate aggregate bank performance to
specialization. The regression takes the following form:
Yb,t = β 0 + β l Specializationb,t + β 2 Totalassetsb,t + β 3 Tier1CapitalRatiob,t + θt + eb,t (4)
The dependent variable Y reflects various measures of bank performance. We are interested
in profitability, measured as net income relative to assets as well as the standard deviation of
income. A high degree of stability can be of interest to banks. If specialization is correlated
with managing and monitoring loans better, we further expect fewer loan defaults on average.
Industry Performance Finally, it is possible that bank specialization and or concentration

impacts the industry in which such banks operate. After all, Bander and Lewis (1986) document
that higher interest rates lead to more aggressive product market strategies. Asker and
Ljungqvist (2010) point out that large banks, which treat borrowers as individual plants of the
same whole, may pass information across competitors. Saidi and Streitz (2020) document a
relationship between bank concentration and markups while Cetorelli and Gambera (2001)
point out that bank concentration and product market competition could be correlated.
We relate measures of industry performance to the average degree of specialization in
the industry. Hereby, we attempt to hold constant as many factors as possible to extract the
component of industry performance that is affected by bank specialization.
3.3 Extension: Specialization and SME Lending
It has been well documented that large banks are somewhat less willing to lend to SMEs (see
for instance discussions in Berger et al. (1998), Strahan and Weston (1998), Peek and Rosengren
(1998), Berger et al. (2005), or Berger and Udell (2008)). This may in part be a consequence of
the fixed cost of each loan contract, which becomes unattractive in the case of large banks able
to lend to large borrowers. It may also be related, however, to the opacity of small firms and
11
the physical distance between loan officers and SME borrowers that prevents the buildup of
soft information.
If specialization is driven by the bank’s incentives to acquire industry-specific knowledge,
we would expect specialized banks to be more willing to lend to small firms. If a bank has
an advantage in assessing firms in an industry, it should be able to assess small firms in that
industry better than competitors and ultimately be more willing to engage in lending. We test
this proposition with a regression that relates the propensity of a bank to lend to small firms to
its specialization in the respective industry. The regression takes the same form as those above:
SMELendingl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Captureb,s,t + β 3 Xl,b

(5)
Here, SME lending is a series of dummy variables. In our basic specification, it can take the
value of 1 if the loan is smaller than a certain threshold. We make use of 3 mil. USD or 2 mil.
USD. Small loans are more likely to be intended for small firms. However, it is also possible
that a bank may split a multi-purpose loan into small components. For a subset of firms in the
Y14, we are able to ascertain their total assets at loan origination. Using total assets we can
determine whether a firm is small (<10 mil USD) and thereby whether the loan is intended for
a small firm. In the subset analysis, the dependent variable takes the value of 1 if the loan is
made to a firm with less than 10 mil. USD.
Our variable of interest is, as before, β 1 . It measures the increased likelihood that a bank
may lend to a small firm if it is highly specialized in its industry. We include a number
of additional loan and bank characterisics as controls. Importantly, we include "capture"
to disentangle the effects of specialization from the effects of a single bank having a near
monopoly on lending in an industry. X also includes measures of firm-bank relationships.
This ensures that our measure of interest is not actually identifying the impact of firm-specific
knowledge that the bank has developed through past interactions.
12
4 Data
Our primary data set comes from the FR Y-14. This data is maintained by the Federal Reserve
and used in supporting the stress testing of major financial institutions. As such, it includes
a variety of details for every bank that has ever been subject to the stress tests. In this paper,
we specifically use the sub-database "H.1", which contains detailed quarterly information
on the C&I loans of reporting banks. Reporting institutions must file all loans with a total
balance-sheet commitment of more than 1 million USD. In the sample period between 2012:Q2
and 2020:Q3 we thus observe 40 banks which report over 3.5 million loan observations. In our
fully cleaned sample, we focus on about 75.000 term-loans in the quarter in which they are first
reported to the Federal Reserve and for whom we can observe size, maturity, and interest rate.
We naturally remove observations with interest rates or maturities that are likely the result of
coding errors.
Unlike other commercially available databases, which cover a subset of the market or
specialize in syndicated lending, our data contains highly detailed information on over 75% of
all C&I lending in the United States (by USD volume) during the sample period. Moreover, it
includes both syndicated and non-syndicated loans. This, in particular, allows us to look at the
differential impacts of "specialization" on larger loans, which may include multiple syndicate
members compared with smaller loans, which are issued and held by a single bank.
Banks report a large set of characteristics for each loan, many of which are useful for
our analysis. Loan characteristics include the type of loan (credit line vs. term loan), total
committed amount, total drawn amount, interest rate, whether a loan is collateralized or
unsecured, loan maturity, a loan’s risk rating, as well as whether a loan has become non-
performing. Besides loan characteristics, the data contains additional information on borrower
characteristics. These include borrower name, location and, most importantly, industry. We use
a borrower’s 2 and 4 digit NAICS industry classifications to define specialization (see above).
[Table 1 about here]
The average bank has 8% of its C&I term loan assets invested in an industry, though
13
this measure is associated with significant variance. The specialization measure is higher in
credit lines, as fewer banks engage in credit line lending (not reported for brevity). In all the
analyses discussed below, we run separate analyses for credit lines and term loans. We do this
because the risks faced by banks for each type of lending, and consequently the benefits to
specialization, may be different in each type of lending. Term loans involve acute credit risk.
Credit lines involve both credit risk, for lines that are drawn, as well as liquidity risk, especially
for long-maturity credit lines.
The average log-size of a loan in our sample is 15 million, which is skewed toward a few
very large loans. Our data is reported in thousands of USD and logged. As can be seen
on Table 1 the average size of our logged loans is 8.6. The average interest rate for loans in
our sample period is 3.3%. Around 21% of loans make use of no form of collateral at first
observation. Non-performing loan is a dummy that takes the value of 1 if a loan is either
flagged as non-accruing by the reporting bank, more than 89 days in payment arrears, or has
negative maturity. A negative maturity implies the loan was not resolved satisfactorily. Despite
a liberal definition, only 3.3% of term loans ever become non-performing during their time in
the sample. Loan ratings are given by the bank in accordance with standardized principles laid
out by the Federal Reserve, so that the ratings are comparable. They range from 1 to 10 with
1 as the best rating. A downgrade occurs when, over the life of the loan, the rating rises, i.e.
the risk of the loan increases. The average number of past interactions between a bank and
a borrower is 4.7. This figure naturally rises towards the end of our sample and is driven by
larger institutions. The average number of times a bank and borrower interact while in the
sample is 9 times. The regressions below can make use of either variable without impacting
results. Finally, the average stress tested bank has over 130 bn. USD invested in C&I lending at
any given time during our sample period.
14
5 Results
In this section we document that even large banks specialize by concentrating into certain in-
dustries. We show that this specialization correlates with better loan terms, especially for larger
firms with access to alternate sources of funding and during periods of higher competition.
This seems to be facilitated by an information advantage gained through specialization, as
specialization is correlated with better loan performance and bank performance. We extend our
analyses along several dimensions. For instance, we show that banks revert to their preferred
industry in times of relatively lower Tier-1 capital.
5.1 Documenting Specialization
We begin our analysis by documenting specialization. Figure 1 shows the share of a bank’s
C&I lending it has invested in its "top" industry (i.e. the industry in which it has invested the
largest share of its assets), its second most preferred industry and all other industries. For the
purpose of these graphs, we define industry according to 2-digit NAICS codes; there are 20
industries based on 2 digit NAICS codes. From Figure 1, it is readily evident that, on average,
the share of a bank’s portfolio in its most favored industry is substantially larger than in any
other industry. Panel (b) of Figure 1 shows the distribution of the share of C&I lending a bank
has invested in a single industry in any given quarter. Some banks appear to have no real
specialization. However, it is also evident that some banks have invested a significant share
of their lending in a single industry. All the data is based on term-loan lending. Patterns
look very similar focusing on credit lines instead. Given data confidentiality, we cannot reveal
which banks have specialized in which industries. However, it is worth noting that different
banks have specialized in different industries. Moreover, there is a high degree of persistence
in specialization. A bank specialized in an industry when it is first observed in the data is 90%
likely to still be specialized in that industry at the end of the sample period.
[Figure 1 about here]
Figure 2 uses the averaged raw data in our sample to identify aggregate trends. In panel (a)
15
we plot the average concentration of a bank’s C&I lending in its "top" industry to its investment
in all other industries. We thereby compute a relative concentration measure that denotes the
share of lending in the top industry less the average share of lending in all other industries. As
can be seen, this share rises over time, implying an increased degree of specialization.
In panels (b) and (c) we repeat the exercise focusing on loan size and interest rate. Panel
(b) showcases the average loan size in a bank’s top industry relative to the average loan size
in its other industries. The loan size in specialized industries is larger, even absent any other
industry-level controls. While there appears to be a high degree of variance in relative loan
size, it too has been growing over the past few years. Panel (c) similarly shows that the average
rate paid by firms in a bank’s specialized industry is lower and has been falling relative to
the rates paid by firms in the industries in which banks are not specialized. Finally, panel
(d) compares the share of loans that turn non-performing in industries in which banks are
specialized with industries in which they are not specialized. Loans in a bank’s specialized
industry, absent any other controls, are consistently less likely to become non performing. the
relative difference in performance has been growing in the latter part of the sample period.
[Figures 2 and 3 about here]
Finally, in Figure 3, we document the relationship between bank specialization in a certain

industry (as a continuous variable that denotes the share of its total C&I lending) to various
loan characteristics and loan performance. Each of these figures represent bin-scatters that
absorb firm and time fixed effects. They thereby reflect the degree to which a firm, borrowing
from multiple banks, may experience differing loan conditions based on the specialization of
the bank. In panel (a) we show specialization and loan size and document a clear positive
relationship. In Panel (b) we show that, even within a given firm and time period, specialization
is negatively associated with interest rate paid. Finally, in panel (c) we show that specialization
corresponds with fewer loan defaults. In the remainder of the paper, we continue to explore
these observations to shed light to the forces behind the results and test the implications of our
hypotheses.
16
5.2 Specialization and Loan Terms
As detailed above, the first part of our analysis focuses on how specialization of a bank in a
given industry relates to standard loan characteristics. We measure specialization as the share
of C&I lending a bank has invested in an industry in a given period. We hold constant as
many loan characteristics as possible as well as bank, firm, time, loan purpose, and loan rating
fixed effects. The effect of specialization is therefore identified across comparable loans to the
same firm. This specification is very stringent. The number of observations highlight that the
large number of fixed effects cause our sample to drop by over 50%, given a large number
of singleton or perfectly estimated observations. The Appendix contains a few alternative
specifications that use alternative fixed effects to highlight that the results discussed here are
not merely identified within a narrow group of firm-loan observations. We discuss this further
in Section 6. Notwithstanding the stringent specification, Table 2 shows that specialization is
correlated with loan characteristics.
Firstly, specialization is positively correlated with loan size. The average bank invests 7.6%
of its C&I loan portfolio in any given industry and over 23% in its "top" industry. Moving from
an average bank to a specialized bank in a given industry would be associated with a loan-size
increase of 26%, all else held equal. This is a sizeable effect given the narrow estimation
parameters. Specialization is further associated with lower interest rate payments. If we again
move from an average to a highly specialized bank we would expect a reduction in interest
by a little over 0.6%. This would still represent 25% of the unconditional average interest rate
payments made by firms in our sample that excludes singleton observations. Though the effect
is insignificant, specialization is weakly associated with increased loan maturity. Finally, loans
made by specialized banks are less likely to be unsecured.
In the Appendix A.1 we explore the relationships between specialization and collateral
further. We separate out different forms of collateral used for a given loan, using the high
degree of detail in our data. Specialization in an industry is associated with a higher likelihood
17
that the loan is secured with accounts receivable, fixed assets, or other assets. All three are
arguably assets that are more easily priced and, if need arises, liquidated by banks with
specialized knowledge.7 Specialization is associated with a lower likelihood that the loan is
secured by marketable securities, which include cash, or a blanket lien and would require less
specialized knowledge.
Our results suggest that a specialized bank is able to acquire industry specific knowledge
which allows it to better evaluate potential borrowers. Better loan terms are reflective of
borrowers being able to extract some of the rents banks derive from specialization. This
is likely to be exacerbated by competition from non-banks or other sources of funding. In
Appendix Table A.3 we interact specialization with a dummy denoting whether a loan is
originated/renegotiated in a time when the availability of non-bank funding is high - i.e. above
the 75th percentile of available non-bank funding during the sample. In Appendix Table A.2
we use a dummy denoting whether the firm is publicly traded. Traded firms are typically
large, rated, and most likely to have access to alternative sources of funding. Taking a firm’s
loan demand as given, we expect the competitive pressures to affect loan terms, all else equal.
Specialized banks with industry specific knowledge may be best positioned to compete on
terms. We find that interest rate, maturity, and whether a loan is secured are all significantly
impacted by competition and specialization. In fact, it appears that the majority of the interest
rate effect is driven by times in which the availability of non-bank funding is high. The effect of
specialization, interacted with our dummies for competitive pressure, on loan size is positively
signed, implying lager loans in the face of greater competition, but statistically insignificant.
Given the saturation of the regressions, these tests are indicative evidence.
In Table 3 we interact the variable of interest, specialization, with a dummy that takes the
value of 1 if the bank is the syndicating agent for the loan. We expect the effects of specialization
to play a large role if a bank is the agent of a loan as opposed to a participant. We find that
7 Thespecialization banks can develop in valuing and dealing with certain collateral is explored in part in
Gopal (2019)
18
this is the case in our data. The interaction of specialization and syndication agent compounds
our baseline results; loans are larger, less costly, and have significantly longer maturity for
the borrower. Somewhat counter to our observations above, we find that the interaction of
bank-industry specialization and the syndication agent dummy attenuate the likelihood that a
loan will be secured with collateral somewhat. However, the aggregate effect remains negative.
In Table 4 we interact the variable of interest, specialization, with the Tier1 ratio of a bank.
A higher Tier 1 ratio implies that a bank is further from any minimum regulatory threshold.
Regulation may influence the type of loans into which a bank invests, as was discussed by Irani
and Meisenzahl (2017) or Irani et al. (2020). We find that being far from a regulatory threshold
attenuates the effects of specialization. A bank that is better capitalized and therefore more
likely to be able to weather possible bad investments – relative to its sample average – may be
more willing to invest in industries in which it has little knowledge and perceives as riskier.
Conversely, a bank that is closer to regulatory thresholds may wish to re-concentrate on its
primary industry by attracting borrowers. This holds most strongly for a loan’s maturity, with
significantly shorter loan maturities offered by well capitalized banks. Similarly, the generous
loan size and interest rate terms are attenuated by high Tier-1 capital of specialized banks.
Only the propensity of a loan being unsecured is influenced by tier-1 capital in the same way as
through specialization directly. We discuss the impacts of Tier-1 capital on bank specialization
in more detail in the following sections.
Specialization vs Capture Finally, bank specialization is correlated with industry capture.

This is consistent with specialized banks being able to discern borrower quality to offer
attractive terms to high-quality borrowers and, therefore, attract a larger fraction of the loans.
At the same time, the larger a bank’s participation in a given industry the higher the incentives
the bank has to acquire knowledge in it. These increasing returns to scale to knowledge may
make some industries captive, leaving them with only a few oligopolistic banks. Captive
industries may see very little bank competition and an associated ability of the lead bank
19
to extract quasi-monopoly rents. In Appendix Table A.6 we show that specialization leads
capture in a temporal sense. A bank may specialize and subsequently capture an industry.
The correlation between lagged specialization on capture is large and significant while the
correlation between lagged capture on specialization is largely insignificant.
We include both our measure of specialization as well as a measure that denotes "industry
capture" by a single bank in a regression. We define capture as a single bank’s contribution
to industry HHI. An industry with one bank would be perfectly captured. An industry with
many small and one very large bank may also rate as highly on our scale. Our variable of
capture, as our variable of specialization, is bounded between 0 and 1.
In Table 5 we show that both capture and specialization affect loan characteristics. However,
these effects often run counter to one another. Capture is associated with larger loans. This may
in part be mechanical and related to our definition of capture and in part related to the fact
that banks with a captured market are less likely to lend to small borrowers (this is explored
further below). However, capture is also related to higher interest payments for loans. This is
in keeping with our hypothesis on monopoly rent extraction. Similarly, capture is associated
with shorter loan maturity while the effect of specialization becomes positive and significant.
Capture is insignificantly related to whether a loan is unsecured or not. Importantly, from
the perspective of this paper, the effects of specialization are not driven by capture and stand
independently.
It is worth highlighting that, throughout this section, we use observable loan characteristics
for loans that have been issued. We are unable to observe applications. In essence, we are
simply highlighting ex-post equilibrium outcomes for characteristics that are jointly determined.
Without an ability to observe the negotiation process, we cannot rule out some mechanics
of credit rationing. Even as equilibrium outcomes, the results discussed above are worth
highlighting. We explore firm-level access to credit in Section 5.4 below.
20
5.3 Specialization and Performance
Loan Performance A bank’s industry specialization has implications for loan performance.
One would expect this outcome if specialized banks are better able to select high-quality loans
ex-ante and better able to monitor loans – or otherwise steer company actions – ex-post due
to an informational advantage over other lenders. Table 6 shows that a loan is less likely to
become non-accruing, less likely to ever be downgraded due to risk, and more likely have a
better (lower) risk rating when last observed in our sample. Both loan-non-accrual as well as
loan downgrades are measured as binary variables that take the value of one if the event ever
occurs in any observation after the first. Loan risk is a standardized measure that, for stress
testing purposes, is harmonized across banks; it takes a value between 1 and 10 with 1 being
the lowest risk. Loan risk at last observation denotes the risk of the loan, after accounting for
interest rates.
Non-accrual (column (1)) occurs rarely in our data and the coefficient on specialization is
often imprecisely estimated in very saturated regressions. Even so, an increase in specialization
would explain the majority of the variation of the dependent variable. Rating downgrades
(column (2)) are more common and enforced by stress testing requirements of the Federal
Reserve. This is true even for private loans held fully by the bank. Importantly, these loan
ratings are standardized by stress testing examiners across banks. We find that an increase in
specialization from an average to a "top" industry would imply a more than 3% reduction of
the likelihood of a downgrade. This represents nearly 20% of the dependent variable mean.
Column (3) shows the loan’s rating at last observation in the data. As such, we can interpret
the coefficient to signify that even in the face of extensive loan controls, which include size
and rate, examiners find loans by bank’s specialized in an industry to be less risky. Given that
our regressions use the first observation, they are each forward-looking in time. The average
loan to a bank’s specialized industry carries a 0.3-points better rating than a loan to its non
specialized industry at last observation.
We use the same saturated specification in our analyses of loan performance as in our
21
regressions on loan characteristics, discussed above. This implies that any effect is estimated
within firm, across differently specialized lenders or within a lender with different specialization
across time. We include rating at first observation as well as the interest rate paid on the loan in
our regressions to account for ex-ante riskiness. As such, our results suggest that ex-post loan
monitoring by specialized and well informed banks may be a key driver of loan performance.
Appendix A.12 uses a specification without firm fixed effects to highlight both the stability of
our results and the slightly increased magnitude of our coefficients. The larger magnitude may
be a reflection of the combination of ex-ante loan selection and ex-post loan monitoring.
In Appendix Table A.5 we perform the same analyses as described in Table 6. However,
we additionally include an interaction term that denotes whether the loan is syndicated.
A syndicated loan is either originated by a different entity and/or includes participants
other than the specialized bank. These other participants may share the riskiness of a loan,
making specialization less necessary, while also being unable to appropriately value specialized
knowledge. The specialized bank would be less incentivized to use its abilities to monitor the
loan. We find, as one might expect, that the performance of loans originated by specialized
banks which remain wholly on the bank’s balance sheet is better than loans that are syndicated.
In Appendix Table A.4 we analyze the propensity of a bank to sell out of its exposure to
a borrower. We relate this propensity to its specialization in the borrower’s industry. It is
possible that a high degree of knowledge about an industry may create a large asymmetric
information discount in the relative price of loans. Market participants with no specialized
knowledge will be unable to distinguish one loan from another. Without an ability to signal
borrower quality, the bank may simply retain loans or exposure to borrowers that it considers
to be of high quality. We find evidence for this in Table A.4. Banks are less likely to reduce
exposure – which typically means selling a larger portion of syndicated loans – or sell out of a
relationship entirely the higher the specialization in an industry8 . Our results highlight the fact
that the results on loan performance are likely the consequence of superior bank information.
8 Inthese analyses we anchor a bank’s specialization to the loan’s origination date to avoid convolution later
stage specialization with reattained exposure to certain borrowers.
22
Bank Performance Better loan performance of specialized banks translates to slightly better
bank performance on aggregate. Table 7 shows various bank level outcomes taken from Y-9C
data. A bank that is on average more specialized experiences lower returns relative to assets.
This may in part be the result of lower loan rates on larger loans, which are discussed above.
However, specialized banks also experience a lower variance in the structure of their returns.
To borrow a concept from from asset pricing; specialized banks have slightly higher sharpe
ratios – the ratio of income to variation in income – than less specialized banks. However
this difference is statistically marginally significant. Importantly, specialized banks see lower
C&I loan charge offs. Chargeoffs are an extreme measure that occur when the loan becomes
irredeemably non-accruing. Lower charge-offs imply a better loan selection or better loan
management. Similarly, the standard deviation of charge offs is lower among specialized banks.
Finally, Table 7 shows that banks with higher specialization show higher levels of tier 1
capital, on average. This observation is explored further in Appendix Figure A.1. Banks with
high specialization have, on average, higher tier 1 capital ratios as well as lower charge-offs.
However, banks grow their specialization – to a certain extent– at the cost of tier 1 capital.
This implies that there may be costs associated with increasing specialization in an industry in
which a bank is not yet specialized. This would be commensurate with higher loses in loans
made by non-specialized banks (see tables above).
Industry Performance Some of the performance patterns discussed for specialized banks
are visible at the level of entire industries. We are able to separate industries by the average
specialization of banks in that respective industry. Industries served by a few highly specialized
banks will see high average specialization while industries served by many diversified banks
will see low average specialization. As can be seen in Table 8, this average specialization
correlates with a lower average share of non-performing loans in the industry (column (3)).
This corroborates the findings at the bank-level discussed above. Similarly, the share of loans
23
downgraded in any year is much lower if the average specialization of banks in an industry is
higher (column (4)).
Surprisingly, the average loan size in industries with high average specialization is lower
(column (1)). In a similar vein, the average rates paid in these industries is higher (column(2)).
While both estimates are insignificant, the observations still seem counter-intuitive relative to
the loan level results discussed above. However, it bears highlighting that the results above
are estimated within borrower; implying that firms obtain larger loans from specialized banks.
Specialized banks, on the other hand, may give more loans to smaller firms, which skews the
estimation at the industry level. We therefore explore this observation in greater detail.
5.4 Specialization and Small Firm Lending
It is a well documented phenomenon that small firms are increasingly less able to borrow from
banks in recent years. We are able to observe this phenomenon in our data. Figure 4 shows
that the share of small loans, as a proportion of stress tested bank portfolios, has been falling in
recent years. It is worth analyzing whether the trend holds for specialized banks as well. After
all, the reason that small firm lending may have decreased lies in part in the opacity of such
firms. If specialized banks are more able to evaluate firms in an industry, one would expect
more small firm lending among such banks.
[Figure 4 about here]
In Figure 4, panel (b), one can observe a positive correlation between the likelihood of a
small loan being granted and a bank’s industry specialization. More specialized banks are
more likely to grant a small loan. We explore this in greater detail in Table 9. Here we relate
the likelihood of a small loan being granted to borrower and loan characteristics. We define
a small loan as one that is smaller than either 3 mil. USD or 2 mil. USD. We use alternative
definitions to make sure our results are not driven by the selection of our sample. For a subset
24
of firms in our data, we are able to determine their total size by assets. As such, we are further
able to analyze whether a loan of any size is granted to a firm of less than 10 mil. USD in
assets. Our controls include the standard loan characteristics discussed above, as well as bank,
industry, year*quarter, and rating fixed effects.
As can be seen, a more specialized bank is more likely to grant a small loan – no matter
the definition – or lend to a small firm. The effect size is considerable when considering the
unconditional mean of the dependent variables. Going from an average to a bank’s specialized
industry, would lead to an almost 8%-pt increase in the likelihood of a small loan being granted.
This represents one 20% of the unconditional mean of the dependent variable in columns (1)
and (2).
Importantly, the analyses include a measure of industry capture in all columns after column
(1). The effect of capture runs counter to the effect of specialization. Banks with a monopoly
hold on an industry are less likely to lend to small borrowers. Conversely, banks with a high
degree of specialization are more likely to lend to such firms.
6 Robustness
As was alluded to above, our Online Appendix includes several alternative specifications
of our baseline regressions that relate specialization to loan performance. These alternative
specifications highlight the stability of our results to various alternative assumptions.
Appendix Table A.7 uses an alternative definition of specialization. The baseline methodol-
ogy uses the share of a bank’s portfolio that is invested in a single industry. This measure could
be driven by single large loans that might be one-time outliers in a bank’s lending behavior.
Specialization may develop not from volume but from frequency. The more frequently a
borrower and lender interact, the more knowledge is gained. As such, the specification uses
specialization defined by the number rather than the volume, of loans to a singe industry.
25
Our results are all confirmed using this specification. The scaled effects are of very similar
magnitude.
Similarly, we include A.8 which uses logged loans to define specialization. As such, we are
defining specialization by the volume of lending but are avoiding the impact of individual
outlier loans. Our baseline results are confirmed using this specification as well. Ultimately,
we observe a high degree of correlation across all of our measures of specialization.
In Appendix Table A.9 we further include firm*year fixed effects as controls in our regression.
This is akin to a khwajamian style regression. We are identifying the effect of bank specialization
within a borrower taking two loans – from differently specialized banks – within the same year.
Even within borrower-time observations, we find that our observations hold. In fact, some
of the effects are somewhat larger. Given the stringent nature of the specification, our effects
are naturally identified within a group of firms that are borrowing multiple times within a
short period. As such, this is likely to be a smaller sub-population of larger firms or firms with
strong growth trends.
Our baseline regressions made use of new loans only. As such, we inherently excluded
re-negotiations of existing loans. We do so to avoid double-counting the same loan – and
associated information – in our regressions. Conversely, it can be argued that re-negotiations
may similarly be influenced by a bank’s specialization. We therefore include Appendix Table
A.10, which includes renegotiated loans in our data. The majority of our baseline coefficients
are corroborated in our full data set. The effect magnitudes are somewhat larger in some
instances. Only loan interest rate is no longer significantly correlated with bank specialization,
though it remains negatively signed.
Finally, we include a set of Appendix Tables (A.11 and A.12) that use the same basic
regressions and samples as our baseline regressions above. However, we exclude firm fixed
effects. These specifications reflect the effect of specialization across different borrowers, as
opposed to within borrowers. The magnitude of our coefficients is larger than for our baseline
specification, which reflects the sizeable difference in loan characteristics for borrowers able to
borrow from banks specialized in their industry.
26
7 Conclusion
In this paper we show that even large stress tested banks specialize in certain industries.
Specialized banks invest a far larger share of their portfolios in a favorite industry compared to
all other industries. We further show that this specialization correlates with loan characteristics.
A firm is likely to receive a larger loan, a lower interest rate, a longer maturity, and be required
to post less cash-equivalent collateral for a loan from a specialized bank, all else equal. We
posit that this follows from the ability of specialized banks to discern borrower quality due
to a buildup of industry-specific knowledge. In an attempt to capture a large share of high
quality borrowers, banks offer superior loan terms.
Commensurate with theories that suggest that specialization may improve a lender’s ability
to monitor an existing loan or select a high-quality loan ex ante, we find that loans from
specialized banks are less likely to become non-accruing or even be downgraded prior to
maturity. This effect holds even when controlling for borrower fixed-effects and loan risk at
origination. Specialized lenders are also more likely to lend to SMEs, which have found bank
borrowing more difficult in recent years due to their opacity.
Finally, we document that specialization has industry-and bank wide implications. Special-
ization leads to more stable bank performance at the cost of slightly lower net interest earnings.
Industries with more specialized banks also experience fewer loan down-grades and write-offs
on average.
Our results speak to the role of banks as designated intermediaries. Their ability to
acquire superior knowledge through specialization has implications for both borrowers in
those industries in which banks specialize as well as bank stability overall.
27
Tables and Figures
(a) Share of Loan Portfolio in an Industry
(b) Distribution of Share in "Top" Industry
Figure 1: Industry Specialization. Panel a shows the average loan portfolio concentration, which is measured
as the share of loans to one two-digit industry at a given point in time, for banks in the sample. Data is split
into the average bank’s "top" industry, it’s secondary industry, and all other industries. A bank’s top industry
is defined as the two-digit NAICS code industry into which a bank has invested the largest share of its portfolio
during its time in the sample. Panel b depicts the distribution of a bank’s loan portfolio in it’s "top" industry.
28
(a) Relative Specialization (b) Relative Loan Size
(c) Relative Loan Rate (d) Relative Loan Performance
Figure 2: Specialization and Loan Terms over Time. This figure depicts concentration in - and average loan
terms to the industry in which a bank has "specialized", relative to all other industries. Panel a shows average
loan portfolio concentration in a bank’s "top" industry relative to all other industries. A bank’s top industry is the
one in which it has lent the largest share of its loan portfolio over time. Panel b compares the average loan size of
loans to the top industry with the average loan size of all other loans. Panel c compares the average loan rate of
loans in a bank’s top industry relative to its other industries. Finally, panel (d) analyzes loan performance and
compares the share of loans that are non-accruing/non-performing.
29
(a) Loan Amount (b) Interest Rate
(c) Performance
Figure 3: Specialization and Loan Characteristics This figure makes use of all loans in our data and depicts
specialization relative to loan characteristics, after absorbing firm and time fixed effects. Specialization is measured
as the concentration of the bank’s loan portfolio in a given industry. Panel (a) depicts loan size relative to a
bank specialization. Panel (b) depicts the interest rate charged relative to specialization and panel (c) depicts the
probability that a loan becomes non-performing.
30
(a) Aggregate Lending to SMEs by Banks in Sample (portfolio share %)
(b) Correlation between Specialization and SME Lending
Figure 4: SME lending This figure makes use of new term loans in our data. Panel (a) shows the share of small
loans (i.e. loans worth less than 2 mil. USD at origination), as a percentage of all lending by the average bank.
Panel (b) shows the relationship between a small loan and a bank’s specialization. Panel (b) accounts for bank*time
fixed effects.
31
Table 1: Data Sources
N Mean S.D. P25 P75

Share of bank’s assets in industry (Specialization) 74,859 0.076 0.063 0.033 0.098
Share of bank’s assets in 4-digit industry (Specialization) 74,859 0.032 0.058 0.003 0.026
Bank’s contribution to industry hhi (Capture) 74,859 0.120 0.163 0.009 0.167
Bank’s contribution to 4-digit industry hhi (Capture) 74,859 2.607 3.931 0.127 3.363
Interest rate 74,859 3.285 1.363 2.25 4.066
Log loan amount 74,859 8.560 1.245 7.555 9.499
Unsecured 74,859 0.212 0.409 0 0
Loan is ever non-performing 74,859 0.033 0.178 0 0
Loan rating is ever downgraded 74,859 0.194 0.395 0 0
Loan rating 74,835 5.262 1.104 5 6
Number of past interactions (relationship) 74,859 4.67 15.67 1 4
Number of bank-firm interactions (ever) 74,859 8.94 29.16 2 7
C and I lending by bank in quarter (bn. USD) 74,859 136.2 124.2 43.1 257.8
Note: This table shows summary statistics for key variables in the paper.
32
Table 2: Specialization and Loan Characteristics
(1) (2) (3) (4)

Log loan amount Interest Rate Maturity remaining Unsecured
Specialization 1.390*** -0.368*** 1.648 -0.972***
[0.201] [0.135] [1.510] [0.094]
Bank FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Industry FE Yes Yes Yes Yes
Year*Quarter FE Yes Yes Yes Yes
Loan Purpose and Rating FE Yes Yes Yes Yes
Controls Loan and bank Loan and bank Loan and bank Loan and bank
Mean of dependent variable 8.7 2.4 17 0.26
R2 0.75 0.87 0.71 0.77
N 32805 32805 31980 32805
Note: This table shows the coefficients of interest for equation:
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
It relates loan characteristics of loan l at time t made by bank b to firm i operating in industry/sector s. Y measures
a loan’s size, the interest rate paid, the maturity remaining or whether the loan is secured with collateral. All
loans are measured at origination only. X are loan characteristics such as amount, spread, maturity and whether
the loan is secured with collateral, unless the characteristic is also the dependent variable in question. It further
includes time-varying bank controls such as size or tier 1 capital. The variable of interest is bank "specialization"
in the given industry, which is measured as the share of the bank’s loan portfolio devoted to the industry. The
regressions include bank, firm, industry, year*quarter, loan risk and loan purpose fixed effects. Standard errors are
clustered at the firm-year level while *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
33
Table 3: Specialization and Syndicated Agents - Loan Characteristics
(1) (2) (3) (4)

Specialization 1.288*** -0.329** 1.074 -1.022***
[0.202] [0.133] [1.527] [0.096]
Specialization * Syndication Agent 0.750** -0.857*** 9.465*** 0.882***

[0.349] [0.313] [2.948] [0.195]
Syndication agent 0.272*** 0.130*** -0.451 -0.047***

[0.043] [0.036] [0.284] [0.016]
R2 0.75 0.87 0.71 0.78
N 32805 32805 31980 32805
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Specializationb,s,t ∗ SyndicationAgent+

β 3 SyndicationAgent + β 4 Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
a loan’s size, the interest rate paid, the maturity remaining, or whether the loan is secured with collateral. All
includes time-varying bank controls such as size or tier 1 capital. The variable of interest is β 1 which denotes
the impact of bank "specialization" in the given industry. Specialization is measured as the share of the bank’s
loan portfolio devoted to the industry. A second variable of interest is β 2 which measures the interaction of
specialization with a binary variable measuring whether the bank is syndication agent for the loan in question.
The regressions include bank, firm, industry, year*quarter, loan risk and loan purpose fixed effects. Standard
errors are clustered at the firm-year level while *, **, and *** indicate significance at the 10%, 5%, and 1% level,
respectively.
34
Table 4: Specialization and Tier 1 Capital
(1) (2) (3) (4)

Specialization 1.459*** -0.416*** 2.148 -0.933***
[0.206] [0.134] [1.532] [0.094]
Specialization * Tier1 -4.600** 2.925* -27.442* -2.274**

[1.927] [1.615] [15.518] [0.887]
Tier 1 Ratio 0.427 -0.597** 1.605 0.217

[0.294] [0.257] [2.748] [0.171]
Mean of dependent variable 8.7 2.4 17 .26
R2 0.75 0.87 0.7 0.77
N 32653 32653 31845 32653
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Specializationb,s,t ∗ Tier1Capital +

β 3 Tier1Capital + β 4 Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
loan portfolio devoted to the industry. A second variable of interest is β 2 which measures the interaction of
specialization with a variable measuring a bank’s Tier 1 capital relative to risk weighted assets. The regressions
include bank, firm, industry, year*quarter, loan risk and loan purpose fixed effects. Standard errors are clustered
at the firm-year level while *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
35
Table 5: Specialization and Capture - Loan Characteristics
(1) (2) (3) (4)

Specialization 1.215*** -0.492*** 2.684* -0.936***
[0.204] [0.141] [1.580] [0.098]
Industry capture 0.274*** 0.193*** -1.603** -0.056

[0.077] [0.066] [0.645] [0.045]
R2 0.75 0.87 0.71 0.77
N 32805 32805 31980 32805
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Capture+

β 3 Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
loan portfolio devoted to the industry. A second variable of interest is β 2 which measures the degree to which
an industry has been captured by the bank in question. Capture is measured as a single bank’s contribution to
the industry’s HHI. The regressions include bank, firm, industry, year*quarter, loan risk and loan purpose fixed
effects. Standard errors are clustered at the firm-year level while *, **, and *** indicate significance at the 10%, 5%,
and 1% level, respectively.
36
Table 6: Specialization and Loan Performance
(1) (2) (3)

Loan ever becomes Loan ever suffers
Loan rating
non-performing rating downgrade
Specialization -0.029 -0.149** -0.615***
[0.030] [0.074] [0.155]
Bank FE Yes Yes Yes
Firm FE Yes Yes Yes
Industry FE Yes Yes Yes
Year*Quarter FE Yes Yes Yes
Loan Purpose and Rating FE Yes Yes Purpose Only
Controls Loan and bank Loan and bank Loan and bank
Mean of dependent variable 0.027 0.17 4.3
R2 0.48 0.42 0.89
N 32805 32815 32805
It relates loan characteristics of loan l at time t made by bank b to firm i operating in industry/sector s. Y
measures whether a loan ever becomes non-performing before maturity or renegotiation, whether a loan is ever
downgraded before maturity, or the rating at origination. Loans with a lower rating are considered safer. All
the loan is secured with collateral. It further includes time-varying bank controls such as size or tier 1 capital. The
variable of interest is bank specialization in the given industry, which is measured as the share of the bank’s loan
portfolio devoted to the industry. The regressions include bank, firm, industry, year*quarter, loan risk and loan
purpose fixed effects. Standard errors are clustered at the firm-year level while *, **, and *** indicate significance
at the 10%, 5%, and 1% level, respectively.
37
Table 7: Specialization and Bank Performance
(1) (2) (3) (4) (5)

Net income SD. Interest Percentage of SD. Charge-offs
Tier 1 Ratio
to assets Income Charge-offs (CandI) (CandI)
Specialization -0.025*** -8.704*** -0.001*** -0.252*** 10.954***
[0.004] [2.048] [0.000] [0.052] [1.089]
Fixed Effects Time Time Time Time Time
Mean of dependent variable 0.015 2.5 0.09 0.067 13
R2 0.49 0.084 0.32 0.28 0.16
N 827 778 827 778 827
Yb,t = β 0 + β l Specializationb,t + β 2 Totalassetsb,t + β 3 Tier1Capitalb,t + θt + eb,t
It relates various characteristics of bank b’s performance at time t to its specialization. Specialization is measured
as it’s average loan concentration. Controls include bank size and Tier 1 capital as well as time fixed effects.
Robust standard errors; *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
38
Table 8: Effects on Industry
(1) (2) (3) (4) (5)

Share of Share of
Log avg. Avg. interest Mean
non-performing loans
loan size rate rating
loans downgraded
Mean specialization -0.306 0.215 -0.317*** -0.272*** -1.008**
[0.469] [1.011] [0.053] [0.101] [0.434]
Industry and Time FE Yes Yes Yes Yes Yes
Mean of dependent variable 9.7 3.4 .013 .046 5.3
R2 0.97 0.92 0.44 0.78 0.97
N 828 828 828 828 828
Yi,t = β 0 + β l MeanSpecializationi,t + γi + θt + ei,t
It relates various characteristics of average loans in industry i at time t to the mean specialization of banks in that
industry. Controls include industry and time fixed effects. Dependent variables of interest are the average loan
size, the average interest rate paid, the share of non-performing loans (at time t), the share of loans downgraded
(at time t) and the average loan rating. Robust standard errors; *, **, and *** indicate significance at the 10%, 5%,
39
Table 9: Specialization and Lending to Small Firms
(1) (2) (3) (4)

Loan < 3 mil Loan < 3 mil Loan < 2 mil Loan to
USD USD USD small firm
Specialization 0.473*** 0.575*** 0.471*** 0.296***
[0.041] [0.041] [0.038] [0.049]
Industry Capture -0.228*** -0.188*** 0.045

[0.019] [0.017] [0.029]
Loan Rating FE Yes Yes Yes Yes
Mean of dependent variable 0.4 0.4 0.27 0.3
R2 0.1 0.11 0.076 0.26
N 74689 74689 74689 53497
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xl,b + θt + ξ b + σs + φloanriskrating + el,i,b,s,t
It relates loan characteristics of loan l at time t made by bank b to firm i operating in industry/sector s. Y are
three binary variables denoting whether the loan is smaller than 3 mil. USD smaller than 2 mil. USD or goes
to a small firm. All loans are measured at origination only. X are loan characteristics such as amount, spread,
maturity and whether the loan is secured with collateral, unless the characteristic is also the dependent variable
in question. It further includes time-varying bank controls such as size or tier 1 capital. The variable of interest is
bank "specialization" in the given industry, which is measured as the share of the bank’s loan portfolio devoted to
the industry. The regressions include bank, industry, year*quarter, and loan risk fixed effects. Standard errors are
40
References
Acharya, V., Hasan, I., Saunders, A., 2006. Should banks be diversified? evidence from individual bank
loan portfolios. The Journal of Business 79 (3), 1355–1412.
Allen, F., Gale, D., 1998. Optimal financial crises. Journal of Finance 53 (4), 1245–1284.
Asker, J., Ljungqvist, A., 2010. Competition and the structure of vertical relationships in capital markets.
Journal of Political Economy 118.
Bander, J., Lewis, T., 1986. Oligopoly and financial structure: The limited liability effect. American
Economic Review 76.
Battacharya, S., Thakor, A. V., 1993. Contemporary banking theory. Journal of Financial Intermediation
3 (1), 2–50.
Berger, A., Miller, N., Petersen, M., Rajan, R., Stein, J., 2005. Does function follow organizational form?
evidence from the lending practices of large and small banks. Journal of Financial Economics 76.
Berger, A., Saunders, A., Scalise, J., Udell, G., 1998. The effects of bank mergers and acquisitions on
small business lending. Journal of Financial Economics 50.
Berger, A., Udell, G., 2008. Universal banking and the future of small business lending. NYU Working
Paper No. FIN-95-009.
Berger, A. B., Udell, G., 1995. Relationship lending and lines of credit in small firm finance. Journal of
Business 68, 351–381.
Bernanke, B., 1983. Non monetary effects of the financial crisis in the propagation of the great depression.
The American Economic Review 73 (3).
Bikker, J., Haaf, K., 2002. Competition, concentration and their relationship: An empirical analysis of the
banking industry. Journal of Banking and Finance 26.
Blickle, K., 2018. Local banks, credit supply, and house prices. Federal Reserve Bank of New York Staff
Reports 874.
41
Blickle, K., Fleckenstein, Q., Hillenbrand, S., Saunders, A., 2020. The myth of the lead arranger’s share.
Federal Reserve Bank of New York Staff Report (922).
Bord, V. M., Santos, J. A., 2015. Does securitization of corporate loans lead to riskier lending? Journal of
Money, Credit and Banking 47 (2-3), 415–444.
Broecker, T., 1990. Credit-worthiness tests and interbank competition. Econometrica 58 (2), 429–452.
Cetorelli, N., Gambera, M., 2001. Banking market structure, financial dependence and growth: Interna-
tional evidence from industry data. Journal of Finance 56.
Cetorelli, N., Strahan, P., 2006. Finance as a barrier to entry: Bank competition and industry structure in
local u.s. markets. Journal of Finance 61.
Degryse, H., Ongena, S., 2005. Distance, lending relationships, and competition. Journal of Finance 60,
231–266.
Diamond, D. W., 1984. Financial intermediation and delegated monitoring. The Review of Economic
Studies 51 (3), 393–414.
Diamond, D. W., Dybvig, P. H., 1983. Bank runs, deposit insurance, and liquidity. Journal of Political
Economy 91 (3), 401–419.
Fernholz, R., Koch, C., 2016. Why are big banks getting bigger? Working Paper, Federal Reserve Bank of
Dallas.
Fleckenstein, Q., Gopal, M., Gutierrez Gallardo, G., Hillenbrand, S., 2020. Nonbank lending and credit
cyclicality. Working Paper.
Goldstein, I., Kopytov, A., Shen, L., Xiang, H., 2020. Bank heterogeneity and financial stability. working
paper.
Gopal, M., 2019. How collateral affects small business lending: The role of lender specialization. NYU
Working Paper.
Haldane, A. G., May, R. M., 2011. Systemic risk in banking ecosystems. Nature (469), 351–355.
42
Irani, R. M., Iyer, R., Meisenzahl, R. R., Peydro, J.-L., 2020. The rise of shadow banking: Evidence from
capital regulation. The Review of Financial Studies.
Irani, R. M., Meisenzahl, R. R., 2017. Loan sales and bank liquidity management: Evidence from a us
credit register. The Review of Financial Studies 30 (10), 3455–3501.
James, C., 1987. Some evidence on the uniqueness of bank loans. Journal of Financial Economics 19 (2),
217–235.
Khwaja, A., Mian, A., 2008. Tracing the impact of bank liquidity shocks: Evidence from an emerging
market. American Economic Review (98).
Laeven, L., Ratnovski, L., Tong, H., 2016. Bank size, capital, and systemic risk: Some international
evidence. Journal of Banking Finance 69.
Paravisini, D., Rappoport, V., Schnabl, P., 2020. Specialization in bank lending: Evidence from exporting
firms. working paper.
Peek, J., Rosengren, E., 1998. Bank consolidation and small business lending: It’s not just bank size that
matters. Journal of Banking and Finance 22.
Petersen, M. A., Rajan, R. G., 1995. The effect of credit market competition on lending relationships. The
Quarterly Journal of Economics 110 (2), 407–443.
Rajan, R., 1992. Insiders and outsiders: The choice between informed and arm’s-length debt. The Journal
of Finance 47 (4), 1367–1400.
Riordan, M. H., 1993. Chapter 11: Competition and bank performance: a theoretical perspective. C.
Mayer, X. Vives (Eds.), Capital markets and financial intermediation, Cambridge University Press.
Saidi, F., Streitz, D., 2020. Bank concentration and product market competition. The Review of Financial
Studies.
Sharpe, S., 1990. Asymmetric information, bank lending, and implicit contracts: A stylized model of
customer relationships. The Journal of Finance 45 (4), 1069–1087.
Strahan, P., Weston, J., 1998. Small business lending and the changing structure of the banking industry.
Journal of Banking and Finance 22.
43
Tabak, B., Fazio, D., Cajueiro, D., 2011. The effects of loan portfolio concentration on brazilian banks’
return and risk. Journal of Banking Finance 35 (11), 3065–3076.
44
Internet Appendix for “Specialization in
Banking”
A.45
(a) Tier 1 and Specialization - across banks
(b) Tier 1 and Specialization - within bank
Figure A.1: Tier 1 and Specialization. Panel a shows the relationship between the Tier 1 capital ratio
specialization across banks, the graph absorbs bank size and time fixed effects. Panel b shows the relationship
between the Tier 1 capital ratio specialization within banks. The graph absorbs bank size, time, and bank fixed
effects.
A.46
Table A.1: Specialization and Collateral
(1) (2) (3) (4) (5) (6)

Secured by Secured by Secured by Secured by Secured by
Blanket lien
real estate marketable securities AR fixed assets other
Specialization 0.036 -0.253*** 0.373*** 0.222*** -0.622*** 0.244**
[0.077] [0.079] [0.091] [0.077] [0.118] [0.106]
Bank FE Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes Yes
Industry FE Yes Yes Yes Yes Yes Yes
Year*Quarter FE Yes Yes Yes Yes Yes Yes
Loan Purpose and Rating FE Yes Yes Yes Yes Yes Yes
Controls Loan and bank Loan and bank Loan and bank Loan and bank Loan and bank Loan and bank
Mean of dep. var. 0.19 0.049 0.17 0.15 0.29 0.16
R2 0.86 0.65 0.71 0.77 0.74 0.75
N 23362 23362 23362 23362 23362 23362
Our data includes only those loan contracts that make use of collateral. The regressions relate the likelihood that
a certain type of collateral is pledged for loan l at time t made by bank b to firm i operating in industry/sector s.
Y is a binary variable denoting whether a certain type of collateral is used. All loans are measured at origination
only. X are loan characteristics such as amount, spread, maturity and whether the loan is secured with collateral,
unless the characteristic is also the dependent variable in question. It further includes time-varying bank controls
such as size or tier 1 capital. The variable of interest is bank "specialization" in the given industry, which is
measured as the share of a bank’s lending by the count of outstanding loans at any given time, regardless of loan
size. The regressions include bank, firm, industry, year*quarter, loan risk and loan purpose fixed effects. Standard
errors are clustered at the firm-year level while *, **, and *** indicate significance at the 10%, 5%, and 1% level,
respectively.
A.47
Table A.2: Specialization interacted with Traded Firm Dummy
(1) (2) (3) (4)

Log loan amount Interest rate Maturity remaining Unsecured
Specialization 0.595*** 0.035 6.457*** -0.089***
[0.065] [0.060] [0.936] [0.028]
Interaction: Specialization * Traded Firm 0.340 -0.222** -5.889*** -1.402***

[0.224] [0.108] [1.514] [0.117]
R2 0.85 0.9 0.73 0.81
N 997984 997984 928023 997984
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Specializationb,s,t ∗ TradedFirmi,t +

Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
the impact of bank "specialization" in the given industry interacted with a dummy denoting whether the firm
is traded. Specialization is measured as the share of the bank’s loan portfolio devoted to the industry. Traded
Firm takes the value of one for firm i in periods when it has a ticker code associated with any major index. The
regressions include bank, firm, industry, year*quarter, loan risk and loan purpose fixed effects. Standard errors are
A.48
Table A.3: Specialization interacted with Non-Bank Funding
(1) (2) (3) (4)

Specialization 0.669*** 0.001 4.818*** -0.417***
[0.072] [0.053] [0.784] [0.036]
Interaction: Specialization * High Non-Bank 0.013 -0.042*** 0.394* -0.010*

[0.016] [0.015] [0.233] [0.006]

R2 0.85 0.9 0.73 .8
N 997984 997984 928023 997984
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Specializationb,s,t ∗ HighNonBankFundingt−1 +

Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
the impact of bank "specialization" in the given industry interacted with a dummy denoting whether the period
has high non-bank funding. Specialization is measured as the share of the bank’s loan portfolio devoted to the
industry. High non-bank funding takes the value of one for times when the level of available non-bank funding
is above the 75th percentile of non-bank funding in the sample. The regressions include bank, firm, industry,
year*quarter, loan risk and loan purpose fixed effects. Standard errors are clustered at the firm-year level while *,
**, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
A.49
Table A.4: Specialization and Exposure to Firms
(1) (2) (3) (4) (5)

Exposure Reduced Maturity Reduced Relationship ended early Exposure Reduced Relationship ended early
Specialization -0.014** -0.035*** -0.004** -0.075*** -0.006
[0.006] [0.004] [0.002] [0.024] [0.006]
Fixed Effects Yes Yes Yes Yes Yes
Firm fixed effects Yes Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes Yes
Sample All loans All loans All loans Syndicated as agent Syndicated as agent
Mean of dep. var. 0.8 0.93 0.011 0.72 0.0073
R2 0.55 0.16 0.098 0.51 0.11
N 833,920 833,920 833,920 39,453 39,453
Yi,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xi,b + θt + γi + ei,b,s,t
We collapse data to the firm-bank-industry-time level. The regressions relate the likelihood that a bank decreases
its exposure to a borrower or sells out of its exposure entirely as a function of its specialization in the borrower’s
industry. Y is a binary variable denoting whether exposure is reduced or whether all contracts are terminated
early (>1 year prior to maturity). X are loan characteristics such as lagged amount and, spread. The variable
of interest is bank "specialization" in the given industry, which is measured as the share of a bank’s lending
by the count of outstanding loans at any given time, regardless of loan size, at the time loans to a borrower a
first originated. The regressions include firm and year*quarter fixed effects. Standard errors are clustered at the
firm-year level while *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
A.50
Table A.5: Specialization and Loan Performance – Syndicated vs Non-Syndicated Loans
(1) (2) (3)

Loan ever Loan ever
becomes suffers rating Loan rating
non-performing downgrade
Specialization -0.044 -0.055 -0.664***
[0.034] [0.091] [0.203]
Loan not syndicated * Specialization -0.021 -0.208* -0.019

[0.034] [0.112] [0.245]
Loan not syndicated -0.012* -0.096*** -0.053

[0.007] [0.025] [0.042]
Bank FE Yes Yes Yes
Firm FE Yes Yes Yes
Loan Purpose and Rating FE Yes Yes Purpose Only
Mean of dependent variable 0.027 0.18 0.18
R2 0.48 0.42 0.89
N 32805 32815 32805
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Specializationb,s,t ∗ Loannotsyndicated+ β 3 Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloa
downgraded before maturity, or the rating at origination. Loans with a lower rating are considered safer. Of
interest is the interaction between whether a loan is syndicated and specialization. All loans are measured at
origination only. X are loan characteristics such as amount, spread, maturity and whether the loan is secured with
collateral. It further includes time-varying bank controls such as size or tier 1 capital. The variable of interest is
bank specialization in the given industry, which is measured as the share of the bank’s loan portfolio devoted to
the industry. The regressions include bank, firm, industry, year*quarter, loan risk and loan purpose fixed effects.
Standard errors are clustered at the firm-year level while *, **, and *** indicate significance at the 10%, 5%, and 1%
level, respectively.
A.51
Table A.6: Specialization vs Capture
(1) (2) (3) (4)

Specialization Specialization Industry Capture Industry Capture
Industry Capture 0.088***
[0.008]
Laged Industry Capture -0.009 0.059***

[0.010] [0.008]
Laged[t−2] IndustryCapture -0.003 0.004

[0.010] [0.010]
Laged[t−3] IndustryCapture 0.006 0.009

[0.008] [0.008]
Specialization 0.454***
[0.043]
Laged Specialization -0.077 0.282***

[0.054] [0.043]
Laged[t−2] Specialization -0.039 -0.002

[0.053] [0.053]
Laged[t−3] Specialization -0.052 -0.016

[0.041] [0.041]
Fixed Effects Industry and Time Industry and Time Industry and Time Industry and Time
R2 0.37 0.37 0.35 0.34
N 18,157 18,157 18,157 18,157
Yb,s,t = β 0 + β l Specializationb,s,t + θt + σs + eb,s,t
It relates contemporaneous and lagged specialization to industry capture of bank b in industry/sector s at time t.
It also relates contemporaneous and lagged industry capture on specialization. It includes industry and time
fixed effects. Robust standard errors; *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively.
A.52
Table A.7: Specialization and Loan Characteristics - Specialization by Count
(1) (2) (3) (4)

Specialization 0.379* -0.340** 2.939* -0.387***
[0.196] [0.147] [1.676] [0.093]
R2 0.75 0.87 0.71 0.77
N 32805 32805 31980 32805
in the given industry, which is measured as the share of a bank’s lending by the count of outstanding loans at any
given time, regardless of loan size. The regressions include bank, firm, industry, year*quarter, loan risk and loan
A.53
Table A.8: Specialization and Loan Characteristics - Specialization by Log Lending
(1) (2) (3) (4)

Specialization 0.518*** -0.367** 2.829* -0.477***
[0.199] [0.149] [1.697] [0.096]
R2 0.75 0.87 0.71 0.77
N 32805 32805 31980 32805
in the given industry, which is measured as the share of a bank’s lending in one industry as measured by the
natural log of loan size. The regressions include bank, firm, industry, year*quarter, loan risk and loan purpose
fixed effects. Standard errors are clustered at the firm-year level while *, **, and *** indicate significance at the
10%, 5%, and 1% level, respectively.
A.54
Table A.9: Specialization and Loan Characteristics - including firm*time FE
(1) (2) (3) (4)

Specialization 1.431*** -0.417*** -2.584* -1.424***
[0.236] [0.146] [1.456] [0.129]
Firm*Year FE Yes Yes Yes Yes
R2 0.81 0.93 0.81 0.86
N 21513 21513 20995 21513
in the given industry, which is measured as the share of a bank’s lending in one industry as measured by the
natural log of loan size. The regressions include bank, firm*year*quarter, industry, loan risk and loan purpose
fixed effects. Standard errors are clustered at the firm-year level while *, **, and *** indicate significance at the
10%, 5%, and 1% level, respectively.
A.55
Table A.10: Specialization and Loan Characteristics - All Loans
(1) (2) (3) (4)

Specialization 0.674*** -0.017 4.986*** -0.421***
[0.072] [0.052] [0.781] [0.036]
R2 0.85 0.9 0.73 .8
N 997,984 997,984 928,023 997984
measures a loan’s size, the interest rate paid, the maturity remaining or whether the loan is secured with collateral.
All loans are at time t, the sample includes renegotiation. X are loan characteristics such as amount, spread,
bank "specialization" in the given industry, which is measured as the share of a bank’s lending in one industry as
measured by the natural log of loan size. The regressions include bank, firm*year*quarter, industry, loan risk
and loan purpose fixed effects. Standard errors are clustered at the firm-year level while *, **, and *** indicate
significance at the 10%, 5%, and 1% level, respectively.
A.56
Table A.11: Specialization and Loan Characteristics - No Firm FE
(1) (2) (3) (4)

Specialization 1.863*** -0.946*** 0.836 -0.099*
[0.114] [0.095] [0.924] [0.053]
Firm FE No No No No
R2 .23 .43 .23 .38
N 59025 59025 56707 59025
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xl,b + θt + ξ b + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
measures a loan’s size, the interest rate paid, the maturity remaining or whether the loan is secured with collateral.
All loans are at time t, the sample includes renegotiation. X are loan characteristics such as amount, spread,
bank "specialization" in the given industry, which is measured as the share of a bank’s lending in one industry as
measured by the natural log of loan size. The regressions include bank, industry, loan risk and loan purpose fixed
effects. Standard errors are clustered at the firm-year level while *, **, and *** indicate significance at the 10%, 5%,
A.57
Table A.12: Specialization and Loan Performance - No Firm FE
(1) (2) (3)

Loan ever becomes non-performing Loan ever suffers rating downgrade Loan rating
Specialization -0.018 -0.074** -1.165***
[0.012] [0.029] [0.133]
Bank FE Yes Yes Yes
Firm FE No No No
Loan Purpose and Rating FE Yes Yes Yes
Mean of dependent variable .033 .19 .19
R2 .15 .097 .23
N 59025 59040 59025
Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xl,b + θt + ξ b + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t
downgraded before maturity, or the rating at origination. Loans with a lower rating are considered safer. All
the loan is secured with collateral. It further includes time-varying bank controls such as size or tier 1 capital. The
variable of interest is bank specialization in the given industry, which is measured as the share of the bank’s loan
portfolio devoted to the industry. The regressions include bank, firm, industry, year*quarter, loan risk and loan
A.58

Fed: Specialization in Banking

Uploaded by

Copyright:

Available Formats

Fed: Specialization in Banking

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fed: Specialization in Banking

Uploaded by

Copyright:

Available Formats

NO.

Kristian Blickle | Cecilia Parlatore | Anthony Saunders

To view the authors’ disclosure statements, visit

and Udell (1995), or Degryse and Ongena (2005).

3.1 Specialization and Loan Characteristics

Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Captureb,s,t + β 3 Xl,b

Per f ormancel,i,b,s,T = β 0 + β l Specializationb,s,t + β 2 Xl,b

Yb,t = β 0 + β l Specializationb,t + β 2 Totalassetsb,t + β 3 Tier1CapitalRatiob,t + θt + eb,t (4)

Industry Performance Finally, it is possible that bank specialization and or concentration

3.3 Extension: Specialization and SME Lending

SMELendingl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Captureb,s,t + β 3 Xl,b

[Table 1 about here]

5.1 Documenting Specialization

[Figure 1 about here]

[Figures 2 and 3 about here]

Finally, in Figure 3, we document the relationship between bank specialization in a certain

[Table 2 about here]

[Table 3 about here]

[Table 4 about here]

Specialization vs Capture Finally, bank specialization is correlated with industry capture.

[Table 5 about here]

[Table 6 about here]

[Table 7 about here]

[Table 8 about here]

5.4 Specialization and Small Firm Lending

[Figure 4 about here]

[Table 9 about here]

(a) Share of Loan Portfolio in an Industry

(b) Distribution of Share in "Top" Industry

(c) Relative Loan Rate (d) Relative Loan Performance

(b) Correlation between Specialization and SME Lending

N Mean S.D. P25 P75

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

Note: This table shows the coefficients of interest for equation:

Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t

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

Specialization * Syndication Agent 0.750** -0.857*** 9.465*** 0.882***

Syndication agent 0.272*** 0.130*** -0.451 -0.047***

Note: This table shows the coefficients of interest for equation:

Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Specializationb,s,t ∗ SyndicationAgent+

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

Specialization * Tier1 -4.600** 2.925* -27.442* -2.274**

Tier 1 Ratio 0.427 -0.597** 1.605 0.217

Note: This table shows the coefficients of interest for equation:

Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Specializationb,s,t ∗ Tier1Capital +

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

Industry capture 0.274*** 0.193*** -1.603** -0.056

Note: This table shows the coefficients of interest for equation:

Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Capture+

(1) (2) (3)

Note: This table shows the coefficients of interest for equation:

Yl,i,b,s,t = β 0 + β l Specializationb,s,t + β 2 Xl,b + θt + ξ b + γi + σs + φloanriskrating + ωloanpurpose + el,i,b,s,t

(1) (2) (3) (4) (5)

Note: This table shows the coefficients of interest for equation:

Yb,t = β 0 + β l Specializationb,t + β 2 Totalassetsb,t + β 3 Tier1Capitalb,t + θt + eb,t

(1) (2) (3) (4) (5)

Note: This table shows the coefficients of interest for equation:

Yi,t = β 0 + β l MeanSpecializationi,t + γi + θt + ei,t

Specialization * Syndication Agent 0.750 -0.857* 9.465* 0.882*

Syndication agent 0.272* 0.130* -0.451 -0.047***

Industry capture 0.274* 0.193* -1.603** -0.056

Industry Capture -0.228* -0.188* 0.045

Interaction: Specialization * Traded Firm 0.340 -0.222 -5.889* -1.402***