Full text of Working Papers (Federal Reserve Bank of Chicago) : Banking Market Structure, Financial Dependence and Growth : International Evidence From Industry Data, Working Paper 1999-08

View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

The Journal of Finance, Forthcoming

Banking Market Structure,
Financial Dependence and Growth:
International Evidence from Industry Data
Nicola Cetorelli and Michele Gambera∗

ABSTRACT
This paper explores the empirical relevance of banking market structure on
growth. There is substantial evidence of a positive relationship between the level of
development of the banking sector of an economy and its long-run output growth.
Little is known, however, about the role played by the market structure of the
banking sector on the dynamics of capital accumulation. This paper provides
evidence that bank concentration promotes the growth of those industrial sectors
that are more in need of external ﬁnance by facilitating credit access to younger
ﬁrms. However, we also ﬁnd evidence of a general depressing eﬀect on growth
associated with a concentrated banking industry, which impacts all sectors and all
ﬁrms indiscriminately.
∗

Cetorelli is at the Federal Reserve Bank of Chicago and Gambera is at Morningstar, Inc. The
paper was written mainly while both authors were at the Federal Reserve Bank of Chicago. We thank
the editor, René Stulz, and two anonymous referees for inputs that substantially improved the overall
quality of the paper. We have also beneﬁted from the comments of seminar participants at the Federal
Reserve Bank of Chicago, Purdue University, Duke University, Loyola University, Ente Einaudi of Rome,
Banco Central de Chile, the 1999 Midwest Macro and 1999 Society of Economic Dynamics conferences,
the 1999 System Committee on Financial Structure and Regulation, the Wharton School-University of
Frankfurt conference on “Bank Competition: Good or Bad?”, the 2000 meetings of the Association of
Financial Economists, and the NBER-Universities research conference on the Macroeconomic Eﬀects
of Corporate Finance. In particular, we thank Judy Chevalier, Gary Gorton, Norman Loayza, Leonard
Nakamura, Marco Pagano, and Oren Sussman for their insightful remarks. We also thank Rob Bliss,
Doug Evanoﬀ, George Kaufman, Jim Kolari, Robert Marquez, Nara Milanich, Raghu Rajan, Klaus
Schmidt-Hebbel, Sherrill Shaﬀer, Shouyong Shi, and especially David Marshall and Luigi Zingales
for fruitful conversations. The views expressed in this paper are those of the authors and do not
necessarily reﬂect oﬃcial positions of the Federal Reserve Bank of Chicago, the Federal Reserve System,
or Morningstar, Inc.

The importance of ﬁnancial development for economic growth has been extensively analyzed in recent years. The amount of credit that the banking sector makes available for
productive uses is one of the most signiﬁcant measures of ﬁnancial development. Such
an indicator of size of the banking sector has been shown to have a signiﬁcant, positive
eﬀect on growth. In this paper we study whether for a given size, the market structure of
the banking sector has empirical relevance for economic growth. If it is agreed that the
size of the banking sector is important to capital accumulation, does it matter whether
the underlying industry structure is unconcentrated, thus approximating perfectly competitive conditions, or whether instead market power is concentrated among few banking
institutions?
We ﬁnd that concentration in the banking sector determines a general deadweight
loss which depresses growth. However, we also ﬁnd evidence that bank concentration
promotes the growth of those industries that are more in need of external ﬁnance by
facilitating credit access to ﬁrms, especially younger ones.
There are theoretical reasons, as well as anecdotal evidence, suggesting that the market structure of the banking sector has a non-trivial impact on the process of capital
accumulation. Conventional wisdom suggests that any departure from perfect competition in the credit market introduces ineﬃciencies that would harm ﬁrms’ access to credit,
thus hindering growth. Pagano (1993), for example, shows this eﬀect in a simple endogenous growth model. On the other hand, some recent contributions have pointed out that
banks with monopoly power have a greater incentive to establish lending relationships
with their client ﬁrms, thus facilitating their access to credit lines. Mayer (1988), Mayer
(1990) and Petersen and Rajan (1995) highlight this potential incompatibility between
bank competition and the establishment of close lending relationships.
There is some historical evidence on the positive role of concentrated credit markets
for economic development. Gerschenkron (1965), for example, mentions the importance

of institutions such as the Credit Mobilier for the industrialization of France, or that
of the Great Banks for Germany’s development. Cohen (1967) explains the similar role
played by Banca Commerciale Italiana and Credito Italiano for Italy, two banks whose
combined assets accounted for about 60 percent of the total market. Likewise, Sylla
(1969) argues that monopoly-enhancing regulation in the ﬁnancial sector at the time of
the Civil War contributed to industrialization in the United States. By the same token,
Mayer (1990) mentions how Japan’s post-war development has been boosted by their
main-bank system.
While the arguments on both sides of this theoretical debate are compelling, no
broad-scope, cross-country empirical study has been conducted to test either stance. In
this paper we choose to take an agnostic position on the issue in order to explore the
consistency of each theory with the available data.1
This paper contributes to the recent line of empirical research on ﬁnancial intermediation and growth. Following the original contributions by Goldsmith (1969), Gurley
and Shaw (1967), McKinnon (1973), and Shaw (1973), economists in recent years have
returned to this problem. Among the newer contributions, King and Levine (1993)
present the ﬁrst broad, cross-country analysis of the importance of various indicators of
ﬁnancial development. They ﬁnd that countries initially endowed with a more sizeable
credit sector experienced faster growth in the following thirty years. Also using crosscountry regression analysis, Levine and Zervos (1998) make an important reﬁnement by
showing the joint, independent relevance for growth of both banks and capital markets.
1

Petersen and Rajan (1995) present some indirect empirical evidence analyzing credit availability
for a cross-section of U.S. small businesses located in markets characterized by diﬀerent degrees of bank
concentration. They ﬁnd that ﬁrms are less credit constrained in more concentrated banking markets,
and younger ﬁrms are charged lower loan rates. Shaﬀer (1998), on the other hand, ﬁnds evidence
from cross-sectional U.S. data that household income grows faster in markets with a higher number
of banks. In two very recent contributions, Bonaccorsi di Patti and Dell’Ariccia (2000), using crossindustry, cross-provinces Italian data, ﬁnd that ﬁrms in more informationally opaque sectors grow more
in more concentrated banking markets, while Black and Strahan (2000), with cross-state U.S. data ﬁnd
a negative relationship between banking concentration and the number of new ﬁrms.

Demirgüç-Kunt and Maksimovic (1998) use instead ﬁrm-level data and show, in a crosscountry study, that where the legal system is more developed ﬁrms have greater access
to external funds, which in turn allows them to grow faster. Meanwhile, Rajan and
Zingales (1998) render an innovative contribution to the ﬁeld by focusing on a crossindustry, cross-country analysis. First, they construct a measure of the dependence on
external ﬁnance of a wide range of industrial sectors, in which diﬀerences among sectors
depend mainly on technology-speciﬁc factors.2 Second, they test whether industries that
are more dependent on external ﬁnance grow faster in countries that are more ﬁnancially
developed. They ﬁnd that this is indeed the case, thus providing evidence conﬁrming
the overall importance of ﬁnancial development on growth.
Our paper tests the importance of banking market structure for growth. We use
an extension of the Rajan and Zingales data set, with both cross-industry and crosscountry characteristics. Similar to the approach taken by King and Levine or Levine
and Zervos, we begin by evaluating the total, average eﬀect of bank concentration on
industrial growth. That is, we test whether, by and large, industries grow more or less
if they are in countries with a more concentrated banking sector. Given the opposing
theoretical views described earlier, the answer to this question is not obvious. On the
one hand, if bank concentration simply results in lower credit availability, then growth
should be slower in countries with a more concentrated banking market. On the other
hand, if the market power associated with bank concentration generates positive eﬀects
by enhancing the formation of lending relationships, then growth should be faster in
countries with a concentrated banking sector. We ﬁnd that bank concentration has an
average depressive eﬀect on industry growth.
However, our empirical study goes beyond the analysis of this average eﬀect of bank
2

For example, an industrial sector at high R&D intensity is expected to rely more on external sources
of ﬁnance than other, more traditional, sectors (e.g. Computing or Chemical products as opposed to
Tobacco or Leather).

market structure. As remarked by Rajan and Zingales, industries diﬀer among each
other in terms of their relative dependence on external sources of ﬁnance. Again, given
the opposing theoretical views one might expect that ﬁrms in sectors especially dependent on external ﬁnance should suﬀer more, and therefore grow less than average, when
faced with a concentrated banking sector. On the other hand, if bank concentration enhances the formation of lending relationships, then one could expect that precisely those
ﬁrms in industries especially dependent on external ﬁnance should beneﬁt disproportionately more when faced with a concentrated banking sector. Exploiting industry-speciﬁc
information, we thus ask whether bank concentration has a heterogeneous impact across
industrial sectors.
The tests we carry out are actually even more precise: corporate ﬁnance theory
suggests that ﬁrms’ relative age may aﬀect their dependence on external ﬁnance. For
example, Rajan and Zingales show that, in median terms, U.S. ﬁrms raise a positive
amount of external ﬁnance only up to the tenth year of their life. Therefore, one would
expect to ﬁnd stronger evidence of either eﬀect of bank concentration by focusing the
analysis speciﬁcally on the external ﬁnancial needs of younger ﬁrms. Since the data
set provides separate information on the ﬁnancial needs of ﬁrms less than ten years
old and on the more mature ones, we are able to do that. Therefore, our empirical
test is: all else equal, do industries whose younger ﬁrms are especially dependent on
external ﬁnance grow more or less rapidly in countries where the banking sector is
highly concentrated? The results show robust evidence that industries in which young
ﬁrms are more dependent on external ﬁnance will in fact grow relatively faster in those
countries where the banking sector is more concentrated.
The two results are not in contradiction. On the contrary, taken together they allow
us to conﬁrm the basic predictions of both theories of banking market structure and
growth: a more concentrated banking industry imposes a deadweight loss in the credit

market as a whole, resulting in a reduction in the total quantity of loanable funds, exactly
as conventional wisdom would suggest. However, subjecting to more careful scrutiny
the complexity of the microeconomic relationship between ﬁrms’ ﬁnancial needs and
sources of ﬁnance, we also ﬁnd evidence that the eﬀect is heterogeneous across industrial
sectors, and that in fact, speciﬁc categories of ﬁrms and industries seem to beneﬁt from
a concentrated banking sector.
In the next section we describe in more detail the theory behind our empirical study.
Section II contains the illustration of the models used for hypothesis testing. In Section
III we describe the data set. The empirical results are presented in Section IV, while
Sections V and VI contain a large variety of robustness tests. In Section VII we present
various reﬁnements of the analysis, while Section VIII elaborates on the several policy
implications associated with the results of our investigation and presents concluding
remarks.
I. Theoretical Background
The negative eﬀect of banking market power is a direct application of standard results from market theory. Banks with monopoly power would determine, with respect
to perfect competition, an equilibrium with higher loan rates and a smaller quantity of
loanable funds. This would clearly reduce economic growth. Conversely, the positive
eﬀect derives from the greater incentive for monopolistic banks to establish lending relationships, which in turn promotes ﬁrms’ access to investment funds. According to the
model developed in Petersen and Rajan (1995), a bank will establish lending relationships with young ﬁrms with no record of performance, thus bearing initial informational
costs, if it can share in their future stream of proﬁts, should they turn out to be successful. However, in highly competitive credit markets, a bank knows that it may not be able
to maintain a tie with the successful ﬁrms: once these ﬁrms are established they will seek
the lowest-cost supply of credit available in the market. Banks that did not invest initial
5

resources in funding the unknown ﬁrms would have a cost advantage in oﬀering better
credit conditions than the bank attempting to recoup the original cost. In the presence
of this free-riding problem, competition in banking can induce credit rationing in the
sense that potentially high quality (but young and unknown) entrepreneurs may not get
funded (for similar analyses of externalities in information production and credit market
competition see also Lang and Nakamura (1989) and Cetorelli and Peretto (2000)).
This theoretical argument is implicitly based on an assumption of market incompleteness. For example, a possible solution to the free-riding problem under competition
would be to allow banks to hold equity positions. Under this scenario, the bank would
participate in future proﬁt sharing regardless of whether the ﬁrm maintains a lending
relationship. In fact, one can argue that monopoly power gives the bank an implicit
equity stake in the ﬁrm it is ﬁnancing. Regulatory restrictions, however, may prevent a
bank from writing equity contracts. We explore how the degree of regulatory restrictions
aﬀect the empirical relationship between banking market power and industrial growth
in Section VII.
II. Model Specifications
A. Basic Model
The ﬁrst model explores the role of bank concentration for industrial growth at large,
that is, regardless of speciﬁc industry characteristics. We write our basic growth equation
as,
Growthj,k = Constant + Ψ1 · Industry dummiesj
+ Ψ2 · Country controlsk
+ ψ 3 · Industry share of manuf acturing value addedj,k
+ ψ 4 · Bank concentrationk
6

(1)

+ Errorj,k ,

where a subscript j indicates that the variable refers to the j-th industry. Similarly,
a subscript k indicates a variable regarding the k-th country. Uppercase coeﬃcients
indicate vectors.
The industry dummies correct for industry-speciﬁc eﬀects. The country controls,
among which is the level of bank development, are regressors customarily used in crosscountry growth studies that we include to reduce the possibility of model mispeciﬁcation
due to the omission of important variables. The entire vector of country control variables
is described in greater detail in the presentation of the empirical results.
The j industries in the data set all belong to manufacturing.3 Similar to the role
played by per-capita income in standard cross-country growth equations, the industry
j share of total value added in manufacturing in country k, calculated at the beginning
of the period, captures an industry-speciﬁc convergence eﬀect: sectors that have already
grown substantially in the past are unlikely to continue to grow at a high rate in the
future. Therefore, ψ 3 is expected to have a negative sign.
Finally, the level of bank concentration isolates the total eﬀect of bank market structure on industrial growth. As we mentioned above, theory suggests that there are two
opposing eﬀects on growth that we can associate with bank concentration. Therefore,
the sign of ψ 4 is a priori ambiguous.
B. Extended Model
The approach outlined above enables us to identify an economy-wide eﬀect of bank
concentration, common to all industrial sectors. In other words, this would be the
eﬀect we would ﬁnd if we used growth rate averages, aggregated across sectors in each
3

As Rajan and Zingales note, this is done “. . .in order to reduce the dependence on country-speciﬁc
factors, like natural resources. . .” (Rajan and Zingales (1998, p. 567]).

country. The use of industry-speciﬁc information yields instead a deeper exploration and
understanding of the role played by banking market structure for growth. This model
speciﬁcation allows us to decompose the total eﬀect of bank concentration in ﬁrst, an
economy-wide eﬀect and second, a sector-speciﬁc eﬀect.
The extended model speciﬁcation is as follows:
Growthj,k = Constant + Φ1 · Industry dummiesj

(2)

+ Φ2 · Country controlsk
+ φ3 · Industry share of manuf acturing value addedj,k
+ φ4 · Bank concentrationk
+ φ5 · (External dependencej · Bank concentrationk )
+ φ6 · (External dependencej · Bank developmentk )
+ Errorj,k .

In this extended speciﬁcation of the model we include the interaction between the level
of external ﬁnancial dependence of industry j and bank concentration in country k. We
test whether sectors that are more in need of external ﬁnance grow disproportionately
slower or faster if they are in a country with high bank concentration. Following the same
arguments as above, the sign of φ5 is a priori ambiguous. As an additional control, we
also include the interaction between external ﬁnancial dependence and the level of bank
development. The coeﬃcient φ6 of this interaction term, extensively analyzed in Rajan
and Zingales (1998), is expected to be positive. As anticipated in the introduction, we
focus on the external ﬁnancial needs of younger ﬁrms (those less than ten years old).4

Results regarding more mature ﬁrms are, however, presented in the section devoted to robustness
tests.

C. Focusing on the Interaction
Finally, in a third model speciﬁcation we focus our attention on the analysis of the
diﬀerential eﬀect of bank concentration across industries, captured by the interaction
term. Since we ﬁnd that bank concentration actually has beneﬁcial eﬀects on industries
more in need of external ﬁnance, a result not obvious ex-ante, we want to be convinced
that this eﬀect is indeed robust. We thus estimate a third model speciﬁcation, where
we drop the level of bank concentration by itself, keeping only the interaction terms.
The reason for choosing to focus on this speciﬁcation to run robustness tests is that it is
econometrically more sound: since we are not identifying ﬁrst-order eﬀects, we can drop
the vector of country controls, and instead we can include country dummies (in addition
to the industry ones), thus eliminating possible biases caused by omitted country-speciﬁc
regressors. The model speciﬁcation is therefore as follows:
Growthj,k = Constant + Γ1 · Industry dummiesj

(3)

+ Γ2 · Country dummiesk
+ γ 3 · Industry share of manuf acturing value addedj,k
+ γ 4 · (External dependencej · Bank developmentk )
+ γ 5 · (External dependencej · Bank concentrationk )
+ Errorj,k ,

where subscripts and variables are the same as described previously.
III. Data
The empirical analysis relies on our augmented version of the Rajan and Zingales data
set.5 The sample includes 41 countries and for each of them 36 industries, yielding a remarkably large sample size. The 36 industries, as mentioned earlier, are all selected from
5

The data set was kindly made available by the authors.

the manufacturing sector. The relevant growth variable is the average (compounded)
rate of growth of real value added for each industrial sector in each country between
1980 and 1990.
Rajan and Zingales calculate the measure of external ﬁnancial dependence for each
industry for U.S. sectors, arguing that the “dependence of U.S. ﬁrms on external ﬁnance
[is] a good proxy for the demand for external funds in other countries” (Rajan and
Zingales 1998, pp. 563–65)).
Table I about here
We augment the data set to include measures of concentration of the banking sector. Speciﬁcally, for each country we calculate the sum of market shares (measured in
total assets) of the three and of the ﬁve largest banks. The data source is the IBCABankScope 1997 CD, which contains detailed balance sheet information on individual
banking institutions for the period 1989 to 1996. For each country we then compute the
concentration ratios for every year in the sample for which there is exhaustive information (the computation for some countries in some years is made unreliable since only a
small fraction of bank balance sheets were reported.) Averages over time constitute our
measures of bank concentration. Table I contains the list of countries in the data set
and the corresponding measures of bank concentration. As we proceed in the description of the empirical results, we will introduce and describe additional variables used for
robustness tests (all variables are summarized in Table II.)
Table II about here
IV. Empirical Results
We were unable to ﬁnd data to construct the concentration measures going back
earlier than 1989. Since the growth variable refers to the decade 1980 to 1990, we
10

are exposed to the potential problems that an ex-post variable could generate, such as
endogeneity. However, we are conﬁdent that in this case the ex-post determination does
not constitute an important issue. First, the market structure of the banking sector, at
the country level, does not vary substantially over such a short time period. We checked
this by analyzing for each country the pattern of variability of the concentration ratios
calculated for the 1989 to 1996 period. We found that indeed such ratios are remarkably
stable over time. For example, we calculated the range for the three-bank concentration
ratio in each country. The cross-country median range of variation in the 1989 to 1996
period was only three percentage points. Even more telling, in about seventy percent of
the countries (27 out of 41) the three-bank ratio had a variation over time of less than
ﬁve percentage points.6 To our knowledge, there are no reasons to believe that such
stability over the 1989 to 1996 period should not also be found in the contiguous 1980
to 1990 period.
In addition, we also constructed the series of the rank of the three-bank and the ﬁvebank concentration ratios. In other words, we allow for the possibility that the averages
calculated over the 1989 to 1996 period are possibly diﬀerent from the ones we would
have calculated for the 1980 to 1990 period, but we require that countries keep their
relative position in the ranking. This is a less stringent condition than requiring that
concentration ratios remain unchanged. Finally, we calculate a dummy variable (highlow concentration) as an additional alternative to our three-bank and ﬁve-bank ratios.
Following the above reasoning, even though the actual values of concentration in the
earlier period may have been diﬀerent from our indices, and perhaps some rankings may
have changed as well, as long as the range of variation was not so large to make a country
shift from the high to the low concentration cluster (or vice versa), a concentration
measure constructed on 1989 to 1996 averages is very likely to be similar to the one we
6

As a term of comparison, the cross-country average three-bank ratio in our sample is approximately
ﬁfty-ﬁve percent.

would have constructed for the 1980 to 1990 period, had the data been available. In
the empirical analysis we test the robustness of the results to the use of this alternative
measures of market structure.
Table III about here
Still on the issue of endogeneity, one could also argue that bank market structure
simply adjusts to a level which is optimal for a country’s industrial structure. However,
this consideration overlooks the fact that there are political and institutional factors that
distort the natural development of ﬁnancial systems. Interest groups, or governments,
or both, will shape the legal, institutional and economic environment for private gains
that may not necessarily coincide with the proper development, in terms of both size and
structure, of the ﬁnancial industry.7 Moreover, in general, the market structure of the
banking sector is a favorite policy variable for reasons not necessarily related to industry
growth.8 Hence, the objective function of the regulator is such that the “optimal” level
of bank concentration may be unrelated to that requested by the industry structure of
the economy.
Beyond this line of discussion, we resolve the concerns regarding the potential endogeneity of the market structure of the banking sector using instrumental variables
(IV) estimation. We have selected as instruments two variables determining a country’s
institutional characteristics, and two variables proxying for market size. In addition to
institutional factors, it is in fact likely that for reasons of minimum scale economies,
a larger market is able to accommodate a higher number of banks. The institutional
variables are an indicator of the legal origin of a country and one concerning the extent
to which laws are enforced in a country (see La Porta, Lopez-de-Silanes, Shleifer and
7

Rajan and Zingales (1999) extensively elaborate on this argument.
For example, the regulator often controls competition in the banking industry to prevent excessive
surplus extraction or for reasons related to the safety and soundness of the industry.
8

Vishny (1997)).9 The proxies for market size are country total population and total
Gross Domestic Product (GDP), measured in U.S. dollars.10
We perform a Durbin-Wu-Hausman (DWH) test of overidentifying restrictions for
each of the regressions in the paper. The test (see Davidson and MacKinnon (1993,
pp. 237–42)) veriﬁes the null hypothesis that the introduction of IVs has no eﬀect on the
estimates of the regression’s coeﬃcients. There are two terms including the measure of
concentration in our equations, namely, the level of concentration and the interaction of
external dependence and concentration. When both terms are present in the regressions,
instruments must be used for both variables. Therefore, for each of the regressions we
perform a DWH F-test, which is reported as the bottom line in each table. If the Pvalue of the F-test is below ten percent (i.e., the null hypothesis is rejected and the IVs
are jointly accepted), then IV estimates are reported. Otherwise, OLS estimates are
reported. We choose the ten percent signiﬁcance level for prudence, since we want to
correct for possible endogeneity when there is the slightest risk of its occurrence.
Anticipating our results, in the vast majority of the regressions we ﬁnd that the
test fails to reject the null hypothesis and that, even when they are used, IVs do not
particularly alter the results of the OLS estimations. Therefore, bank concentration is
robust to the issue of endogeneity, both in level and in interaction.
A. Basic Model
In table IV we report the results of regressions based on the speciﬁcation in equation
(1), in which we add one country control variable at a time.
Table IV about here
9

For another example of instrumental variable approach to tackle the issue of endogeneity of ﬁnancial
variables in growth regressions, see Levine, Loayza and Beck (2000). A time-series analysis of causality
between ﬁnancial variables and growth is in Demetriades and Hussein (1996).
10
As shown in Table III, the negative correlation between bank concentration and the measures of
market size are indeed quite high, especially that with total GDP. In fact, we also tested whether by
using bank concentration we were not just capturing a market size eﬀect. The results, not reported,
show that bank concentration is robust to the inclusion of market size variables.

The dependent variable is the average annual growth in value added for each sector
in each country, while the three-bank ratio is the measure of bank concentration. Unless
otherwise noted, the dependent variable and the measure of bank concentration will
remain the same in all of the following regressions. The country control variables are
the level of bank development, the logarithm of per capita income in 1980, stock market
capitalization over GDP in 1980, an accounting standards indicator, and a measure of
the level of human capital. The measure of bank development is the commonly used
ratio between domestic credit to the private sector and GDP, and is expected to have
a positive eﬀect. Per-capita GDP captures the convergence eﬀect of the economy as a
whole to its long-run steady state, and is therefore expected to have a negative sign.
Stock market capitalization controls for the relative importance of alternative sources
of external ﬁnance and is expected to have a positive sign. Accounting standards is an
index reﬂecting the quality of disclosure of ﬁrms’ annual reports (see Rajan and Zingales
(1998, p. 571)). The poorer such standards, the higher the information cost that a bank
has to sustain to determine the quality of an entrepreneur. The expected sign for this
term is also positive. The level of human capital, another typical regressor in growth
equations, is measured as average number of school years in population over age 25 (as
in Barro and Lee (1993)), and is expected to have a positive eﬀect as well.
All control variables have the sign that one would expect to ﬁnd in any cross-country
growth equation. Also as expected, the share of total value added in manufacturing
is negative and signiﬁcant. The main result to highlight is that the level of bank concentration has a negative and signiﬁcant coeﬃcient. The eﬀect of bank concentration
is robust to the inclusion of the various control variables. This result lends support to
the prior that a concentrated banking industry imposes a deadweight loss in the credit
market and on the economy as a whole.
In order to gauge the economic relevance of the bank concentration variable, we

perform a standard comparative dynamics exercise. Speciﬁcally, we calculate the total impact on growth if we were going from a country at the 25th percentile of the
distribution of bank concentration to a country at the 75th percentile. The eﬀect on
growth based on the estimated coeﬃcients in the regression in column (b) of Table IV
is a negative 1.5 percentage points.11 Note that the average growth across sectors is
3.6 percent (see Table II). We should not take these number at face value, since this
model speciﬁcation is exposed to the aforementioned omitted variable bias, due to the
non-inclusion of potentially important country variables. However, the overall evidence
suggests that the total eﬀect is, on average, negative and signiﬁcant, both statistically
and economically.
B. Extended Model: Economy-wide and Sector-speciﬁc Eﬀects
In Table V we report the results of regressions based on the speciﬁcation in equation
(2), in which, again, we add one country control variable at a time, and where we include
the interaction between external ﬁnancial dependence and bank concentration, together
with that between external ﬁnancial dependence and bank development, acting as a
control. External ﬁnancial dependence refers to that of the younger ﬁrms in the data
set.
Table V about here
The level of bank concentration by itself maintains a negative and signiﬁcant coeﬃcient in the regressions (a) through (d), while we highlight that the interaction coeﬃcient
11

Column (b) regression has the highest number of controls that still allows us to use the largest sample
size, with 1150 observations. Column (c) and (d) regressions have more controls but the inclusion of
accounting standards and human capital brings the sample size down to 950 observations, with a loss
of nine of the original countries in the data set. The countries are Bangladesh, Costa Rica, Egypt,
Jordan, Kenya, Morocco, Pakistan, Sri Lanka, and Zimbabwe. We choose to present our basic results
based on the larger sample, in order to minimize informational losses (nine countries out of forty-one
represent a rather signiﬁcant twenty-two percent). Moreover, since the nine countries are all developing
countries, we want to avoid sample bias. At any rate, based on the estimates of column (d), the eﬀect
of increasing bank concentration would be a negative 2.4 percentage points.

is instead positive and signiﬁcant. The two eﬀects of bank concentration are robust to
the inclusion of the various control variables, all showing the expected sign.12 The
interaction between external ﬁnancial dependence and bank development becomes insigniﬁcant in the last three columns where, as we mentioned in footnote 11, the inclusion
of accounting standards and human capital makes nine countries drop from the sample.
The combined results are consistent with the theoretical priors. They suggest that
bank concentration has a negative eﬀect on growth that, on average, aﬀects all sectors
indiscriminately. However, when we introduce the dimension of the intensity of external
ﬁnancial dependence, we identify an industry-speciﬁc, positive eﬀect of bank concentration. This eﬀect is consistent with the theoretical prior that bank market power, by
facilitating the formation of lending relationships, enhances the growth potentials for
those sectors that are more in need of establishing such relationships. The result indicates that the impact of bank concentration on growth is not uniform across industries.
Consequently, bank concentration has an important redistribution eﬀect. We elaborate
further on this point and the related policy implications in the following sections.
It is also worthwhile remarking that our results are shielded by a potentially important objection. External ﬁnancial dependence in the data set is measured on U.S.
sectors. However, U.S. sectors with higher external ﬁnancial dependence are those that
grew more over time. We also know that countries at a comparable level of economic
development are similar in terms of their industry structure. Therefore, it is likely that
sectors with high external ﬁnancial dependence grew more not just in the U.S. but also
in those countries similar to the U.S., i.e., the richest countries. Suppose now that,
similar to bank development, bank concentration had a positive, high correlation with
income per capita. In that case, the ﬁnding of a positive and signiﬁcant bank concentra12

We also ran a regression where we included external ﬁnancial dependence by itself as a regressor.
In order to do that, however, we had to exclude the industry dummy variables. Such a regressor was
not signiﬁcant while all other coeﬃcients were unchanged.

tion interaction, especially when the bank development interaction is insigniﬁcant, could
simply capture this positive association between external ﬁnancial dependence and the
level of economic development. However, we do not think this is a matter of concern,
since there is actually no correlation in the data between bank concentration and income
per capita (see Table III).
We also add a squared term of bank concentration to check for non-linearity. As
the results in column (e) of Table V show, bank concentration has in fact a slight
inverted-U eﬀect. The interpretation for this result is actually provided by what we
learn from the sector-speciﬁc analysis: intuitively, at low levels of concentration only the
sectors that have the lowest need to establish a lending relationship receive the maximum
beneﬁt, in that the deadweight loss from rent extraction is minimal but the potential
informational gains from establishing lending relationships are also minimal. Conversely,
at high levels of concentration only the sectors that are highly dependent on banks will
beneﬁt. The non-linearity seems to suggest that at intermediate values of concentration
the overall growth potential of the entire economy is the highest, since sectors in an
intermediate range of the distribution of external ﬁnance beneﬁt substantially, and those
at the extremes of the distribution of external ﬁnance are not highly penalized.
We evaluate the magnitude of the total eﬀect of bank concentration on economic
growth, decomposed in the ﬁrst-order, economy-wide eﬀect and the second-order, sectorspeciﬁc ones. More precisely, we calculate the eﬀect on growth of sectors with diﬀerent
levels of external ﬁnancial dependence if we went from a level at the 25th percentile to
one at the 75th percentile of the distribution of bank concentration. As we did above,
we perform this calculation based on the estimation results of the regression in column
(b) of Table V. The economy-wide eﬀect is a negative 2.6 percent. For the sector at the
25th percentile of the distribution of external ﬁnancial dependence (Wood products),

the sector-speciﬁc eﬀect is a positive 0.8 percent.13 The net eﬀect for such sector is thus
around minus 1.7 percentage points.14 If we perform this calculation for other sectors,
the second order eﬀect becomes stronger as the intensity of external ﬁnancial dependence
increases.15 Indeed, the net eﬀect actually turns positive for those sectors at the upper
tail of the distribution.16
V. Statistical Robustness Tests
We now turn to present a large battery of robustness tests, for which we focus on
the speciﬁcation in equation (3). Nevertheless, the reader must rest assured that all the
pertinent robustness tests have also been conducted on the ﬁrst-order eﬀect, also shown
to be very robust. When relevant, we do mention or report the result of tests carried
out on the equation (2) model speciﬁcation.
Table VI about here
Table VI reports in column (a) the estimation results for the benchmark equation
(3). The share of total value added in manufacturing remains negative and signiﬁcant,
and the interaction between external dependence and bank development positive and
signiﬁcant. The coeﬃcient of the interaction between external ﬁnancial dependence and
bank concentration is still positive and signiﬁcant at the ﬁve percent level.
Based on the results of this model speciﬁcation, the sector at the 25th percentile of
the external ﬁnancial dependence distribution would receive a positive eﬀect translating
13

The distribution of external ﬁnancial dependence for young ﬁrms has virtually identical mean and
median. The median is equal to 0.66 and the mean equal to 0.67.
14
Based on the estimates of column (d) regression, the ﬁrst-order eﬀect would be a negative 4.5 percent
and the second-order eﬀect a positive one percent, with a net eﬀect of around minus 3.5 percentage
points. Remember, however, the remarks in Footnote 11 about sample size.
15
For example, for the sector at the 75th percentile of the distribution (Ship), the second order eﬀect,
calculated on column (b) coeﬃcients, is estimated to be a positive 1.9 percent, bringing the net eﬀect
to minus 0.7 percentage points.
16
For example, based on column (b) coeﬃcients, the net eﬀect of bank concentration would be positive
for sectors such as Glass, Professional goods and Drugs.

in 0.9 percent of higher growth in going from a country at the 25th percentile of the
distribution of bank concentration to a country at the 75th percentile. A sector at the
75th percentile would receive instead a positive eﬀect of two percent higher growth.
These results conﬁrm the robustness of those obtained with the model speciﬁcation of
equation (2). Recall, however, that all sectors are also subject to the negative economywide eﬀect, which is not identiﬁable with this model speciﬁcation. The information on
the diﬀerent sector-speciﬁc eﬀect allows us to gauge the impact of bank concentration
on the size distribution across industrial sectors. Speciﬁcally, we learn that the growth
diﬀerential between an industrial sector at the 75th percentile and one at the 25th
percentile of the distribution of external ﬁnancial dependence for younger ﬁrms, if we
were going from a country at the 25th percentile of the distribution of bank concentration
to a country at the 75th percentile is estimated to be about 2 − 0.9 = 1.1 percentage
points on an annual basis. Again recalling that the average growth over all sectors is
3.6 percent, this redistribution eﬀect of bank concentration appears to be economically
signiﬁcant.
A. Do the Results Depend on the Choice of the Concentration Measure?
We ﬁrst check whether the three-bank ratio calculated over the 1989 to 1996 period
is an adequate measure of concentration. In column (b) of Table VI, the concentration
measure is the rank of the three-bank ratio, while in column (c) is the high-low concentration dummy for the three-bank ratio.17 In column (d) we use the ﬁve-bank ratio,
to check that the results would not depend on the arbitrary choice of computing the
concentration measure as the sum of the market shares of the top three banks only. The
strong similarity of the results obtained with these alternative measures suggests that
17
For the calculation of the concentration dummy, countries were divided between those with a value
of the ratio below and those above a value equal to 0.70, which is what would be considered high
concentration, for example, in the U.S. banking industry (see, e.g., Calem and Carlino (1991)). An
alternative speciﬁcation, which gave unchanged results and is not reported in the table, divided the
countries around the median of the distribution (0.57).

the three-bank ratio computed for the 1989 to 1996 period is a reliable measure for our
analysis.
In what follows we therefore continue to present regression results using the threebank concentration ratio calculated over the 1989 to 1996 period as our benchmark
measure of banking market structure.
B. Omitted Variables
We test whether the term of interaction of bank concentration is signiﬁcant when
we omit the bank development interaction from the basic model speciﬁcation. The
results, reported in column (a) of Table VII, show that the concentration variable remains
positive and signiﬁcant at the ten percent level.18
Table VII about here
Subsequently, we check whether the bank concentration interaction variable is still
signiﬁcant after controlling for the relative importance of alternative sources of external
ﬁnance. We therefore start by adding the interaction between external dependence and
the size of stock market capitalization over GDP. The coeﬃcient is expected to have a
positive sign. The results in Table VII, column (b), show that this coeﬃcient is indeed
positive, although not signiﬁcant. The bank concentration term is still positive and
signiﬁcant at the ﬁve percent level, with an estimated coeﬃcient close in magnitude to
the one in the baseline regression (column (b), Table VI).
In column (c) of Table VII we report the results of a regression where we add the interaction between external dependence and the logarithm of per-capita income in 1980.
There is a concern that the interaction term of bank development in our basic speciﬁcation may be proxying for the general level of economic development of a country. The
18

We note a decrease in the estimated coeﬃcient, from 0.063 to 0.048. This is likely to be the result
of an omitted variable bias. Performing a bias analysis (see, e.g., Berndt (1991, p. 322) it is indeed
conﬁrmed that the omission of the bank development variable determines a downward bias on the
coeﬃcient of bank concentration.

simple correlation between bank development and per-capita income, 0.56 (reported in
Table III), may justify this concern. The coeﬃcient of the bank concentration interaction remains positive and is signiﬁcant at the ten percent level. Conﬁrming the existence
of some collinearity, the bank development and the income interaction terms have the
expected sign but neither is signiﬁcant.19
Finally, we add to the basic regression the interaction between external ﬁnancial
dependence and the measure of accounting standards. The expected sign for this term
of interaction is positive. Column (d) in Table VII presents the results of this augmented
speciﬁcation of the model. The coeﬃcient of bank concentration is still positive and
signiﬁcant, even though the size of the estimated coeﬃcient decreases from 0.063 to
0.035. The two coeﬃcients, however, are not immediately comparable, again due to
the fact that by including the accounting standard variable the number of observations
decreases from 1150 to 984.20
C. Outliers
A general concern is that the results based on these growth regressions could be
driven by the exceptional performance of some countries (for example, Southeast Asian
countries) or certain industrial sectors in particular countries, which could not be fully
captured by the inclusions of the country and sector dummies. This should not aﬀect our
analysis, since the sample in the Rajan and Zingales data set does not include countries
such as Taiwan or Hong Kong. In addition, the series of growth in value added censors
from above sectors that, on average, grew more than 100 percent annually in the 1980
to 1990 period. To be sure, we run a regression dropping the censored observations
19

If we run an equation (3) regression where we replace the bank development interaction term with
the income level interaction term, the bank concentration interaction maintains its sign, signiﬁcance
and magnitude as well.
20
If we run the equation (3) regression on the restricted sample excluding records for these countries,
the coeﬃcient of the bank concentration variable is equal to 0.042, signiﬁcant at the one percent level.
As pointed out earlier, we choose to present our basic results based on the larger sample, in order to
minimize informational losses and sample bias.

altogether.21 The results, reported in column (e) of Table VII, show that the bank
concentration interaction becomes signiﬁcant at the one percent level, although with a
smaller coeﬃcient. The coeﬃcient of the bank development interaction, signiﬁcant now
at the ten percent level only, decreases as well.
In addition, we evaluate whether the results are sensitive to high and low values in
the distribution of young ﬁrms’ external ﬁnancial dependence. We use a dummy variable
to separate sectors above the median from those below the median of the distribution
of external ﬁnancial dependence and redeﬁne the bank concentration interaction term
accordingly. The results of this regression are reported in column (f) of Table VII.
They show that the interaction term is positive and highly signiﬁcant while the dummy
variable term is not signiﬁcantly diﬀerent from zero. This is interpreted as conﬁrming
that the eﬀect of bank concentration is stable across the entire sample.
VI. Economic Robustness Tests
A. Tests on Mature Firms
We also investigate the role of bank concentration for the ﬁnancial needs of the more
mature ﬁrms in the data set, that is, establishments more than ten years old. As we
mentioned earlier, the external ﬁnancial needs of this category of ﬁrms are much lower
than for younger ones. Moreover, the problem of information acquisition on established
ﬁrms is less severe than for younger ﬁrms. Therefore, focusing speciﬁcally on mature
ﬁrms, we might expect a less important eﬀect of bank concentration on industrial growth.
In the ﬁrst column of Table VIII we report the results of the third model speciﬁcation,
this time calculating the terms of interaction using the external ﬁnancial dependence of
mature ﬁrms.
Table VIII about here
21

We thank Rob Bliss for suggesting this and other outlier tests.

The bank concentration term is still positive and signiﬁcant. However, the economic
eﬀect on growth is half the size of that determined on young ﬁrms: the growth diﬀerential
between an industrial sector at the 75th percentile and one at the 25th percentile of the
distribution of external ﬁnancial dependence for mature ﬁrms, if we were going from a
country at the 25th percentile of the distribution of bank concentration, to a country at
the 75th percentile, is estimated to be about 0.5 percentage points on an annual basis.
This is much lower (the diﬀerence is statistically signiﬁcant) than the growth diﬀerential
for young ﬁrms, which, as reported above, is instead 1.1 percentage points.
Among the mature ﬁrms, those that have already grown substantially and are well
established are likely to receive minor beneﬁts from a banking relationship, and are
therefore more likely to be exposed to the rent extraction activity of a concentrated
banking sector. Therefore, we perform an additional test on mature ﬁrms, splitting
the sample between large and small sectors, i.e., those sectors in each country that had
a share of value added in manufacturing above the country median, and those below.
Columns (b) and (c) of Table VIII report the results for the two subsets. The bank
concentration interaction term continues to be positive and signiﬁcant for the sectors
below the median. For sectors above the median the coeﬃcient is positive but no longer
signiﬁcant.
In summary, bank concentration appears to have a positive eﬀect on growth of sectors
that are more in need of external ﬁnance. Consistent with theory, the economic impact is
more pronounced for younger ﬁrms than for more mature ones. The dominating positive
eﬀect of bank concentration seems to disappear only when we focus on a particular subset
of the more mature ﬁrms.
B. Market Contestability
Concentration ratios are widely used in empirical analysis to proxy for competitive

conduct.22 However, the potential weakness of this measure is that if markets are contestable, market structure does not necessarily aﬀect conduct. In a cross-country study,
market contestability can be gauged in terms of the ability of foreign banks to access
domestic markets. We can test whether the actual presence of foreign banks aﬀects the
relationship between bank concentration and industry growth using data on the share
of total domestic assets owned by foreign banks (taken from Demirgüç-Kunt and Levine
(1999)), and on the fraction of foreign banks over the total number of banks (taken from
Claessens, Demirgüç-Kunt and Huizinga (1998)). Admittedly, such measures may not
capture the eﬀect on the conduct of domestic banks of a potential threat of entry, which
is what contestability is more about. On the other hand “the threat of foreign bank
entry may not be credible in the absence of actual entry” (Claessens, Demirgüç-Kunt
and Huizinga (1998, p. 7)).
Data show a limited presence of foreign banks in most of the countries in the sample.
For instance, the median share of assets owned by foreign banks is six percent (the 75th
percentile is 14 percent). At the same time, in terms of the number of foreign banks
over the total, perhaps a better indicator of the potentials for entry, the median fraction
is a more substantial 24 percent. The relatively low weight of foreign banks in most
countries may be due to the existence of administrative barriers to entry that were or
still are in place in developing countries, where hostility toward foreign institutions can
be traced back to the experience of colonialism (Vittas (1992)). Such restrictions are
found in developed countries as well. For example, prior to 1993, countries that are now
members of the European Union (EU) signiﬁcantly restricted the entry of foreign banks.
Such restrictions are still in place with respect to banks from non-EU countries.23
22

Recent developments in empirical industrial organization have proposed alternative measures of
market power, which could be used instead of the traditional concentration ratios (see, e.g., Bresnahan
(1991). Such measures are identiﬁed through econometric estimation of industry’s demand and supply
conditions. The major drawback of such an alternative approach is the need for extensive data, which
is only available for the most developed countries.
23
One can also argue that besides regulatory restrictions, informational barriers play an important

We generate a dummy variable equal to one for countries with both a three-bank ratio
and a foreign share of bank assets above their medians. These are countries where, given
the relatively high presence of foreign banks, a high concentration ratio may overestimate
the actual degree of monopoly power. We run an equation (3) regression where we add
the product of the dummy with the concentration interaction term. The results, in
column (a) of Table IX, show that the concentration interaction alone is still positive
and signiﬁcant, while the dummy term is not.
Table IX about here
This suggests that even if the concentration measure may be biased upward in some
countries, such bias is not driving the main ﬁndings.24
C. Under-reporting Bias
We use data of foreign banking penetration to take into account another potential
source of bias in the concentration variable. As described in Section III, the concentration ratios are calculated using the IBCA data set. Such a data set collects balance
sheet items for all banks that report such information. While the percentage of banks
reporting is very high, it is still possible to introduce a bias due to under-reporting. In
particular, Beck, Demirgüç-Kunt and Levine (1999) observe that “. . .[using these data]
the concentration measure might be biased upwards for developing countries, if foreign
and large banks are more likely to report than domestic and smaller banks.” To adrole as well in preventing a banking market from being contestable. The existence of informational
barriers is discussed, and evidence is provided, e.g., in Berger, Bonime, Covitz and Hancock (1999).
Some casual evidence is also provided by the observation that despite the removal of the aforementioned
regulatory barriers to entry among EU countries, the actual presence of banks from other EU countries
is still very limited, averaging about ﬁve percent of total bank assets across country (European Central
Bank (1999)).
24
As an alternative test, we dropped the records for those countries altogether and ran a regression
on the restricted sample. We also repeated these tests with diﬀerent cutoﬀs in the bank concentration
and foreign share distributions, and we also used the proportion of foreign banks in place of foreign
share. Similar tests of robustness were also performed on the equation (2) model speciﬁcation. The
results, available upon request, are in all instances qualitatively identical, indicating an eﬀect of bank
concentration on industrial growth robust to the issue of market contestability.

dress this problem, we generate a dummy variable equal to one for countries below the
median in per-capita GDP and above the median in the foreign bank share, in order to
isolate those countries where the concentration measure is more likely to be biased due
to under-reporting. We then run regressions adding the product of the dummy with the
bank concentration interaction term. While the bank concentration term alone remains
signiﬁcant, the dummy term is not signiﬁcant, suggesting that the under-reporting bias
is not a problem (see column (b), Table 9).25
D. Using Measures of Bank Eﬃciency
We use interest margins and overhead costs as alternative measures of competition.
Using cross-country data from Demirgüç-Kunt and Huizinga (1998), we ﬁnd that the
concentration measure is not correlated with either variable (see Table III). We run
regressions using either one in place of bank concentration, but we do not ﬁnd signiﬁcant
results. An explanation is that interest margins and overhead costs are measures of
bank eﬃciency that, in a cross-country analysis, are likely to be aﬀected by countryspeciﬁc factors. Consistent with this argument, Demirgüç-Kunt and Huizinga (1998),
in a cross-country analysis ﬁnd that, in fact, factors such as macroeconomic conditions,
bank taxation, deposit insurance, legal structure, and other institutional indicators are
very important in the determination of interest margins and overhead costs. They also
conﬁrm that bank concentration, at the cross-country level, is only mildly related to
interest margins and to overhead costs. Hence, trying to trace information on bank
competitive conduct via interest margins or overhead costs is in this case likely to yield
spurious results. For instance, relatively higher margins in one country do not necessarily
imply relatively higher banking market power.

The same results are obtained using the proportion of foreign banks in place of foreign share.

VII. Refinements and Extensions
A. State-owned Banks
Another potential criticism of our use of the concentration ratios is that in some
countries a large proportion of banks is owned by the government. In such cases, where
the same subject owns many banks, those banks might act as a cartel. As a consequence,
the concentration measure in some countries could underestimate the actual market
power. At the same time, it is also plausible to argue that public banks may not
necessarily be proﬁt maximizers and may not have an incentive to establish lending
relationships with potentially proﬁtable enterprises.
Beck, Demirgüç-Kunt and Levine (1999) and Demirgüç-Kunt and Levine (1999)
collect cross-country information on state ownership in the banking sector. We generate
a dummy variable equal to one if the share of state-owned banks is above a certain
threshold to single out countries with a signiﬁcant government presence in the banking
sector.26 Then we test for a non-linear response to the concentration interaction term in
the case where state ownership is particularly high. The coeﬃcient of the product of the
dummy and the concentration interaction term is negative, signiﬁcant and almost the
same in absolute value as the concentration interaction term alone (column (c) in Table
IX). Hence, the positive eﬀect of bank concentration appears to be oﬀset in countries
with a dominant government presence in the banking industry.
To explore further, we also run regressions where we add the interaction between
external ﬁnancial dependence and state ownership. The concentration interaction term
is still signiﬁcant, with a slightly decreased coeﬃcient. The state ownership interaction is negative and signiﬁcant (column (d) in Table IX). Interestingly enough, in the
extended model speciﬁcation we notice that even the level of bank development turns
26

We set the threshold at 50 percent, 60 percent and 70 percent, obtaining virtually identical results.
The regression described in the text is that with the threshold at 60 percent.

insigniﬁcant when we control for the degree of state ownership (column (e) in Table IX),
thus suggesting a general negative impact associated with the presence of the state in
the credit market.
B. Bank Powers
The eﬀect of bank monopoly power may diﬀer depending on the regulatory environment in which banks operate in a country. As mentioned in Section I, if a bank were
allowed to ﬁnance ﬁrms through equity, then even under perfect competition it would
have an incentive to establish close relationships. Hence, in a world where banks are
less constrained in their ﬁnancing choices, we may expect the positive eﬀect of banking
concentration on growth to be less important.
Barth, Caprio and Levine (2000) compile information across countries on the restrictions for banks in terms of their ability to write contracts. They summarize this
information in a quantitative indicator ranging from one (broadest powers) to four (narrowest powers). We add to the model an interaction between this measure of bank
powers and ﬁnancial dependence. The bank concentration interaction remains highly
signiﬁcant, while the bank development interaction loses signiﬁcance.
Table X about here
The bank powers interaction is not signiﬁcant (column (a) of Table X).27 We also
add both bank powers alone and the interaction to the equation (2) model. Both bank
concentration alone and its interaction are highly signiﬁcant while the bank powers
variables are not signiﬁcant (column (b) of Table X).
As an additional test on the equation (2) model we also include an interaction of
the level of bank concentration with bank powers, to see if the ﬁrst-order negative eﬀect
27

We also generate a dummy variable equal to one if bank powers are very broad (below the median).
We then add the product of the dummy and the concentration-dependence interaction to the regression.
This new term is also not signiﬁcant.

of concentration on industrial growth diﬀers across countries with diﬀerent regulatory
restrictions. One could argue, for example, that if banks were allowed to operate in
multiple lines of business, they would face increasing cross-markets competitive pressure
that could reduce their ability to extract rent. The result (column (c), Table X) shows,
however, that this interaction is not signiﬁcant as well. In sum, there is not evidence from
this data set that bank regulatory restrictions have a direct impact on the relationship
between banking market structure and growth.
C. Does Bank Concentration Lead to Industry Concentration?
Another relevant issue is whether bank concentration causes ﬁnancially dependent
industries to become concentrated, thus enabling ﬁrms to earn monopoly proﬁts. Banks
may act as a barrier to entry by privileging incumbents, with whom they already established relationships, over new entrants.28 Since we measure industrial growth in terms of
value added, under such a scenario we could observe positive growth due to an increase
in proﬁts and not necessarily in production, with important welfare implications.
The Rajan and Zingales data set contains information on growth in the number of
establishments and growth in the average size of existing establishments that can be
used to test this hypothesis. If bank concentration induces industry concentration, and
thus industry monopoly proﬁts, then we should ﬁnd that in those sectors that are more
dependent on external ﬁnance there is a relatively slower growth in the number of new
establishments, and in association with it a relatively faster growth in the average size
of existing establishments. In our model speciﬁcation, this implies that the hypothesis
that bank concentration induces industry concentration is consistent with the ﬁnding
that the interaction between bank concentration and external ﬁnancial dependence is
negative and signiﬁcant in a regression with growth in number of establishments as
28

In his study of Italian industrialization at the turn of the past century, Cohen (1967, p. 363) reports
the relation between a quasi-monopolistic banking industry and “. . .the emergence of concentration of
ownership and control in the new and rapidly growing sectors of the industrial structure.”

dependent variable, and positive and signiﬁcant in a regression with growth in average
size as dependent variable.
We ﬁrst run regressions with growth in number of establishments as the dependent
variable. The results in column (d) of Table X show that while the level of concentration
is negative and signiﬁcant, the interaction term is positive and signiﬁcant. Column (e)
shows that the interaction term remains positive and signiﬁcant even when we drop the
regressors in level and we can introduce country dummies.
We then use growth in average size as the dependent variable. The results of these
regressions are reported in column (f) and (g) of Table X. In both model speciﬁcations
the interaction term is never signiﬁcant. Overall, the evidence does not support the
argument that bank concentration may enhance industry concentration. The results
thus indicate that growth in value added is a good proxy for growth in output.
VIII. Concluding Remarks
Important recent contributions have established with reasonable conﬁdence that ﬁnancial development, characterized by a sizeable banking sector, matters for economic
growth. The next important step in the research agenda involves delving deeper into
the micro details governing the actual functioning of the ﬁnance-growth nexus. Beyond
a “black-box” characterization of the banking sector, implicit in focusing on its relative
size only, there is a much more complex web of banks and other institutions interacting
in the credit markets. The various attributes of such a system are likely to have a qualifying impact on the ﬁnance-growth relationship. The market structure of the banking
industry, reﬂecting its competitive conditions, is, in our opinion, one such attribute.
The ﬁndings in our paper suggest a non-trivial impact of bank concentration on industrial growth. There is evidence that bank concentration has a ﬁrst-order negative
eﬀect on growth. This ﬁnding is consistent with the theoretical prediction that higher
30

bank concentration results in a lower amount of credit available in the economy as a
whole. Regardless of their external ﬁnancial dependence, this eﬀect is common to all
industrial sectors. However, we also ﬁnd evidence that bank concentration has a heterogeneous eﬀect across industries. In particular, sectors that are more dependent on
external ﬁnance enjoy a beneﬁcial eﬀect from a concentrated banking sector. This positive eﬀect could actually more than compensate the direct, negative eﬀect on quantities.
This ﬁnding supports models predicting that concentration of market power in banking
facilitates the development of lending relationships, which have in turn an enhancing
eﬀect on ﬁrms’ growth.
The main insights of this study are that ﬁrst, at least maintaining the focus on the
eﬀects on industrial growth, there does not seem to be a Pareto-dominant policy regarding the optimal banking market structure: competition in banking does not necessarily
dominate monopoly, and vice versa. Second, depending on the level of bank concentration, and ceteris paribus, individual industries will grow at diﬀerent speeds. Therefore,
banking market structure plays an important role in shaping the cross-industry size distribution within a country. Moreover, since bank concentration plays a more substantial
role for growth by facilitating credit access of younger ﬁrms, and to the extent that
investment by younger ﬁrms is more likely to introduce innovative technologies, banking
market structure should have an impact on the pace of technological progress.
The results of the paper have relevance for developing countries, where governmentsponsored programs, such as development banks (see Armendariz de Aghion (1999))
or micro banking (see Morduch (1999)), which aﬀect credit market structure, aim at
providing the conditions for convergence to higher levels of welfare. Yet, our ﬁndings
are also relevant for developed countries where we witness important regulatory reforms
and signiﬁcant structural transformations of the banking industry.

References
Armendariz de Aghion, Beatriz, 1999, Development banking, Journal of Development
Economics 58, 83–100.
Barro, Robert J., and Jong W. Lee, 1993, International comparisons of educational
attainment, Journal of Monetary Economics 32, 363–394.
Barth, James R., Gerard Caprio, and Ross Levine, 2000, Banking systems around
the world: do regulation and ownership aﬀect performance and stability?, in Frederic
Mishkin, ed.: Prudential supervision: What works and what doesn’t (NBER Press, Cambridge, MA).
Beck, Thorsten, Aslı Demirgüç-Kunt, and Ross Levine, 1999, A new database on ﬁnancial development and structure, Working paper, World Bank.
Bencivenga, Valerie R., and Bruce Smith, 1991, Financial intermediation and endogenous growth, Review of Economic Studies 58, 195–209.
Berger, Allen N., Seth D. Bonime, Daniel M. Covitz, and Diana Hancock, 1999, Why
are bank proﬁts so persistent? The roles of product market competition, informational
opacity, and regional/macroeconomic shocks, Working paper, Board of Governors of the
Federal Reserve.
Berndt, Ernst R., 1991, The practice of classic and contemporary econometrics (AddisonWesley, New York, NY).
Black, Sandra E., and Philip E. Strahan, 2000, Entrepreneurship and bank credit availability, Working paper, Federal Reserve Bank of New York.
Bonaccorsi di Patti, Emilia, and Giovanni Dell’Ariccia, 2000, Bank competition and ﬁrm
creation, Working paper, Bank of Italy.

Bresnahan, Timothy, 1991, Empirical studies of industries with market power, in Richard
Schmalensee and Robert D. Willig, eds.: Handbook of industrial organization (North
Holland, Amsterdam).
Calem, Paul, and Gerald Carlino, 1991, The concentration/conduct relationship in bank
deposit markets, Review of Economics and Statistics 73, 268–276.
Cameron, Rondo, 1967, Banking in the early stages of industrialization (Oxford University Press, New York, NY).
Cetorelli, Nicola, and Pietro F. Peretto, 2000, Oligopoly banking and capital accumulation, Working paper, Federal Reserve Bank of Chicago.
Claessens, Stijn, Aslı Demirgüç-Kunt, and Harry Huizinga, 1998, How does foreign entry
aﬀect the domestic banking market?, Working paper, World Bank.
Cohen, Jon, 1967, Financing industrialization in Italy, 1894–1914, Journal of Economic
History 27, 363–382.
Davidson, Russell, and James G. MacKinnon, 1993, Estimation and inference in econometrics (Cambridge University Press, New York, NY).
Demetriades, Panicos O., and Khaled A. Hussein, 1996, Does ﬁnancial development
cause economic growth? time-series evidence from sixteen countries, Journal of Development Economics 51, 387–411.
Demirgüç-Kunt, Aslı, and Harry Huizinga, 1998, Determinants of commercial bank interest margins and proﬁtability: some international evidence, Working paper, World
Bank.
Demirgüç-Kunt, Aslı, and Ross Levine, 1999, Bank-based and market-based ﬁnancial
systems: cross-country comparisons, Working paper, World Bank.

Demirgüç-Kunt, Aslı, and Vojislav Maksimovic, 1998, Law, ﬁnance, and ﬁrm growth,
Journal of Finance 56, 2107–2137.
European Central Bank, 1999, Possible eﬀects of EMU on the EU banking systems in
the medium to long term (European Central Bank, Frankfurt).
Gerschenkron, Alexander, 1965, Economic backwardness in historical perspective (Praeger,
New York, NY).
Goldsmith, Raymond, 1969, Financial structure and economic development (Yale University Press, New Haven, CT).
Gurley, John G., and Edward S. Shaw, 1967, Financial structure and economic development, Economic Development and Cultural Change 15, 257–278.
King, Robert, and Ross Levine, 1993, Finance and growth: Schumpeter might be right,
Quarterly Journal of Economics 108, 717–737.
La Porta, Rafael L., Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert W. Vishny,
1997, Legal determinants of external ﬁnance, Journal of Finance 52, 1131–1150.
Lang, William W., and Leonard I. Nakamura, 1989, Information losses in a dynamic
model of credit, Journal of Finance, 44, 731–746.
Levine, Ross, Norman Loayza, and Thorsten Beck, 2000, Financial intermediation and
growth: causality and causes, Journal of Monetary Economics, forthcoming.
Levine, Ross, and Sara Zervos, 1998, Stock markets and economic growth, American
Economic Review 88, 537–558.
Mayer, Colin, 1988, New issues in corporate ﬁnance, European Economic Review 32,
1167–1183.
Mayer, Colin, 1990, Financial systems, corporate ﬁnance and economic development,
34

in R. Glenn Hubbard, ed.: Asymmetric information, corporate ﬁnance and investment
(The University of Chicago Press, Chicago, IL).
McKinnon, Ronald I., 1973, Money and capital in economic development, (Brooking
Institution, Washington, DC).
Morduch, Jonathan, 1999, The microﬁnance promise, Journal of Economic Literature
37, 1569–1614.
Pagano, Marco, 1993, Financial markets and growth. An overview, European Economic
Review 37, 613–622.
Petersen, Mitchell A., and Raghuram G. Rajan, 1995, The eﬀect of credit market competition on lending relationships, Quarterly Journal of Economics 110, 407–443.
Rajan, Raghuram G., and Luigi Zingales, 1998, Financial dependence and growth, American Economic Review 88, 559–586.
Rajan, Raghuram G., and Luigi Zingales, 1999, The politics of ﬁnancial development,
Working paper, University of Chicago.
Shaﬀer, Sherrill, 1998, The winner’s curse in banking, Journal of Financial Intermediation 7, 359–392.
Shaw, Edward S., 1973, Financial deepening in economic development (Oxford University Press, New York, NY).
Sylla, Richard, 1969, Federal policy, banking market structure and capital mobilization
in the United States, 1863–1913, Journal of Economic History 29, 657–686.
Vittas, Dimitri, 1992, The impact of regulation on ﬁnancial intermediation, in Dimitri Vittas, ed.: Financial regulation. Changing the Rules of the Games (World Bank,
Washington, DC).

Table I
List of Countries and Bank Concentration Ratios
For each country we calculated the sum of market shares (measured in total assets) of the three and
the ﬁve largest banks. The data on individual banking institutions for each country in the sample, are
from the IBCA-BankScope 1997 CD for the period 1989–1996. The values reported are averages over
the sample period. Note that data about the United States are not used in any of the regressions; we
report them only for sake of completeness.

Country
Australia
Austria
Bangladesh
Belgium
Brazil
Canada
Chile
Colombia
Costa Rica
Denmark
Egypt
Finland
France
Germany
Greece
India
Israel
Italy
Japan
Jordan
Kenya

3-Bank
0.60
0.42
0.62
0.49
0.40
0.57
0.45
0.35
0.71
0.74
0.58
0.85
0.28
0.27
0.79
0.40
0.79
0.24
0.21
0.87
0.59

5-Bank
0.80
0.55
0.75
0.73
0.50
0.84
0.62
0.54
0.82
0.82
0.73
0.98
0.44
0.39
0.91
0.51
0.94
0.38
0.32
0.94
0.72

Country
Korea
Malaysia
Mexico
Morocco
Netherlands
New Zealand
Norway
Pakistan
Peru
Philippines
Portugal
Singapore
South Africa
Spain
Sri Lanka
Sweden
Turkey
United Kingdom
United States
Venezuela
Zimbabwe

3-Bank
0.28
0.44
0.53
0.57
0.77
0.75
0.60
0.71
0.64
0.40
0.46
0.61
0.69
0.34
0.75
0.71
0.41
0.50
0.15
0.47
0.78

5-Bank
0.44
0.54
0.66
0.79
0.88
0.99
0.74
0.90
0.76
0.56
0.63
0.83
0.90
0.50
0.89
0.94
0.56
0.65
0.20
0.62
0.97

Variable
Growth in value addedj,k
Growth in average sizej,k
Growth in numberj,k
F raction of value addedj,k
3 − bank ratiok
5 − bank ratiok
External dependence allj
External dependence youngj
External dependence oldj
Accounting standardsk
Bank developmentk
Log per capita GDPk
Stock market capitalizationk /GDPk
Human capitalk
Interest margink
Overhead costsk
Bank powersk
F oreign bank sharek
N o. of f oreign banksk
State ownershipk
P opulationk
T otal GDPk
Rule of lawk

Num. Obs
1150
990
992
1150
1150
1150
1150
1150
1112
984
1150
1150
1150
1106
1442
1442
1363
1283
1479
1011
1584
1584
1085

Mean
0.036
0.024
0.012
0.015
0.538
0.691
0.338
0.665
0.018
0.610
0.384
7.923
0.202
6.109
3.889
3.654
2.171
0.116
0.258
0.334
71.545
585.202
7.101

Std. Dev.
0.092
0.090
0.068
0.020
0.176
0.184
0.371
0.619
0.299
0.138
0.202
1.302
0.306
2.743
2.419
2.208
0.642
0.147
0.187
0.325
149.282
1378.334
2.612

Min
-0.447
-0.536
-0.414
0.000
0.210
0.320
-0.146
-1.535
-1.330
0.240
0.068
4.793
0.000
1.681
1.400
1.300
1.000
0.000
0.000
0.000
3.000
4.000
2.08

Max
1.000
0.410
0.758
0.224
0.870
0.990
1.491
2.058
0.393
0.830
0.856
9.572
1.624
12.141
13.600
10.200
3.500
0.620
0.850
0.980
962
7783
10

Growth in value added is the average (compounded) rate of growth of real value added for each industrial sector in each of the countries between
1980 and 1990. Growth in the average size of ﬁrms is growth in the ratio of value added to the number of ﬁrms. The fraction of value added
is industry j’s share of manufacturing value added in country k; three-bank and ﬁve-bank are the concentration ratios of the banking sector of
each country as explained in Table I; external ﬁnancial dependence measures refer respectively to the borrowing needs of all establishments (all),
of establishments less than ten years old (young), and of establishments ten years and older (old). Accounting standards is an index ranking the
amount of disclosure of companies’ annual reports for each country. Bank development is the ratio of private domestic credit to GDP. Human
capital is the average for 1980 of the years of schooling attained by the population over 25 years of age. Interest margin is banks’ net interest
income divided by total assets (1988 to 1995). Overhead costs is the ratio of banks’ overhead costs to total assets (1988 to 1995). Bank powers
is a measure of regulatory restrictions on bank activities in the 1990s. Foreign bank share is the ratio of foreign bank assets to total bank assets
(1988 to 1995). Number of foreign banks is the ratio of the number of foreign banks to the total number of banks (1988 to 95). State ownership
is the share of assets of state-owned banks over total commercial bank assets. Total GDP is in 1997 U.S. dollars calculated using the World Bank
Atlas method. Rule of law is an indicator of the extent to which laws are enforced.

Table II
Descriptive Statistics of Data Set

38
Interest
1.000
0.847*
0.066
0.245*
0.352*
0.024

Stock
1.000
0.175*
0.444*
0.169*
0.142*
0.213*
State

1.000
-0.023
-0.049
-0.132*

1.000
0.098
0.395*
0.317*
-0.020

1.000
0.259*
0.172*
0.224*

1.000
0.587*
0.560*
-0.057
-0.015
Overhead

Accounting

GDP

Bank

1.000
0.240*
0.125*

F oreign

1.000
-0.270*
-0.288*

* indicates that the correlation is diﬀerent from zero at the 1 percent signiﬁcance level.

Variables
Interest margink
Overhead costsk
State ownershipk
F oreign presencek
Bank powersk
3 − bank ratiok

Variables
Stock marketk /GDPk
Log per capita GDPk
Accounting standardsk
Bank developmentk
3 − bank ratiok
5 − bank ratiok

1.000
-0.250*

P owers

1.000
0.972*

3 − Bank

1.000

3 − Bank

1.000

5 − Bank

Stock market/GDP is the ratio between stock market capitalization and GDP in 1980; Log per capita GDP is the log of per capita GDP in
1980; Accounting standards is a measure of the quality of accounting and disclosure levels of businesses; Bank development is the ratio of private
domestic credit to GDP; three-bank and ﬁve-bank are the concentration ratios of the banking sector of each country as explained in Table I.
Interest margin is banks’ net interest income divided by total assets (1988 to 1995). Overhead costs is the ratio of banks’ overhead costs to total
assets (1988 to 1995); State ownership is the share of assets of state-owned banks over total commercial bank assets; foreign presence is the ratio
of foreign bank assets to total bank assets (1988 to 1995); Bank powers is a measure of regulatory restrictions on bank activities in the 1990s.
Foreign bank share is the ratio of foreign bank assets to total bank assets (1988 to 1995).

Table III
Correlations between Concentration Measures and Selected Variables

39
0.127
1150
0.60

(a)
-0.875***
(0.260)
0.074***
(0.016)
-0.038**
(0.017)
-0.016***
(0.003)

0.137
1150
0.13

(b)
-0.876***
(0.260)
0.066***
(0.016)
-0.048***
(0.017)
-0.016***
(0.003)
0.031***
(0.006)

0.201
984
1.57

(d)
-0.303**
(0.141)
0.058***
(0.014)
-0.117***
(0.016)
-0.025***
(0.004)
0.029***
(0.007)
0.109***
(0.027)
0.004**
(0.002)
0.223
950
8.73***

* indicates rejection of the null at the 10 percent signiﬁcance level, ** indicates 5 percent signiﬁcance level, and *** indicates 1 percent signiﬁcance
level.

R2
Observations
Durbin − W u − Hausman

Human capitalk

Accounting standardsk

Stock market capitalizationk /GDPk

Log of per capita GDPk

Bank concentrationk

Bank developmentk

Regressors
F raction of value addedj,k

The dependent variable in all columns is the average (compounded) rate of growth of real value added for each industrial sector in each country
between 1980 and 1990. The fraction of value added is industry j’s share of manufacturing in country k. Bank development is the ratio of private
domestic credit to GDP. Bank concentration is the three-bank ratio. Human capital is the average for 1980 of the years of schooling attained
by the population over 25 years of age. Missing values in the accounting standards variable restrict our sample when such variable is used as a
regressor (the countries that are dropped are: Bangladesh, Costa Rica, Jordan, Kenya, Morocco, Pakistan, Sri Lanka, and Zimbabwe). Similarly,
the inclusion of the measure of human capital implies that records about Egypt are dropped from the sample. This explains the diﬀerence in the
number of observations. Heteroskedasticity-consistent standard errors are reported in parentheses. The Durbin-Wu-Hausman statistic tests the
null hypothesis that the use of instrumental variables does not change the estimation outcome. We report IV estimates when the test is rejected
at the 10 percent level or less. Instruments are: rule of law, legal origin, total GDP, and population. Industry dummy variables are included in all
regressions, but we do not report their coeﬃcient estimates. Note that country dummy variables are not included in these regressions.

Table IV
The Average Effect of Bank Concentration on Industrial Growth

40
0.133
1150
0.69

0.059**
(0.030)
-0.016***
(0.003)

(a)
-0.888***
(0.260)
0.043**
(0.018)
0.047**
(0.024)
-0.077***
(0.017)

0.144
1150
0.16

0.060**
(0.029)
-0.016***
(0.003)
0.031***
(0.006)

(b)
-0.889***
(0.259)
0.035*
(0.018)
0.047**
(0.024)
-0.088***
(0.017)

0.204
984
1.04

0.040***
(0.015)
-0.023***
(0.003)
0.023***
(0.007)
0.067***
(0.024)

0.048**
(0.021)
-0.025***
(0.004)
0.029***
(0.007)
0.109***
(0.027)
0.004**
(0.002)
0.226
950
4.40**

(d)
-0.305**
(0.139)
0.053***
(0.019)
0.008
(0.018)
-0.149***
(0.021)

(e)
-0.274**
(0.137)
0.085***
(0.022)
0.009
(0.019)
0.243*
(0.136)
-0.367***
(0.133)
0.049**
(0.021)
-0.028***
(0.004)
0.020**
(0.008)
0.123***
(0.029)
0.005**
(0.002)
0.192
950
7.27***

* indicates rejection of the null at the 10 percent signiﬁcance level, ** indicates 5 percent signiﬁcance level, and *** indicates 1 percent signiﬁcance
level.

R2
Observations
Durbin − W u − Hausman

Human capitalk

Accounting standardsk

Stock market capitalizationk /GDPk

Log of per capita GDPk

External dependencej · Bank concentrationk

Squared bank concentrationk

Bank concentrationk

External dependencej · Bank developmentk

Bank developmentk

Regressors
F raction of value addedj,k

The dependent variable in all columns is the average (compounded) rate of growth of real value added for each industrial sector in each country
between 1980 and 1990. The fraction of value added is industry j’s share of manufacturing in country k; external ﬁnancial dependence refers to the
borrowing needs of establishments less than ten years old. Bank development is the ratio of private domestic credit to GDP. Bank concentration
is the three-bank ratio. Human capital is the average for 1980 of the years of schooling attained by the population over 25 years of age. As
mentioned in the note to Table IV, missing values in the accounting standards variable restrict our sample when such variable is used as a regressor
(the countries that are dropped are: Bangladesh, Costa Rica, Jordan, Kenya, Morocco, Pakistan, Sri Lanka, and Zimbabwe). Similarly, the
inclusion of the measure of human capital implies that records about Egypt are dropped from the sample. This explains the diﬀerence in the
number of observations. Heteroskedasticity-consistent standard errors are reported in parentheses. The Durbin-Wu-Hausman statistic tests the
null hypothesis that the use of instrumental variables does not change the estimation outcome. We report IV estimates when the test is rejected
at the 10 percent level or less. Instruments are: rule of law, legal origin, total GDP, and population. Industry dummy variables are included in all
regressions, but we do not report their coeﬃcient estimates. Note that country dummy variables are not included in these regressions.

Table V
Economy-Wide and Sector-Specific Effects of Bank Concentration on Industrial Growth

Rank
(b)
-0.906***
(0.285)
0.046**
(0.022)
0.088**
(0.041)
0.287
1150
0.41

High/Low
(c)
-0.903***
(0.286)
0.031*
(0.019)
0.019*
(0.010)
0.286
1150
0.73

5-Bank
(d)
-0.903***
(0.285)
0.045**
(0.021)
0.085**
(0.035)
0.288
1150
0.22

* indicates rejection of the null at the 10 percent signiﬁcance level, two asterisks indicates 5 percent signiﬁcance level, and three asterisks indicates
1 percent signiﬁcance level.

R2
Observations
Durbin − W u − Hausman

External dependencej · Bank concentrationk

External dependencej · Bank developmentk

Regressors
F raction of value addedj,k

3-Bank
(a)
-0.905***
(0.285)
0.049**
(0.022)
0.063**
(0.029)
0.288
1150
0.48

The dependent variable in all regressions is the average (compounded) rate of growth of real value added for each industrial sector in each of the
countries between 1980 and 1990. The fraction of value added is industry j’s share of manufacturing value added in country k; external ﬁnancial
dependence refers to the borrowing needs of establishments less than ten years old. Banking development is the ratio of private domestic credit
to GDP. Column (a) uses the three-bank ratio (as deﬁned in the text) as a measure of banking concentration. In column (b) bank concentration
is the rank of the three-bank ratio, column (c) uses a high-low concentration dummy, while column (d) has the ﬁve-bank concentration ratio. The
high-low dummy is equal to one for the countries with a high value of concentration (≥ 0.7) and zero otherwise. Alternative estimation, with the
dummy for countries below or above the median (0.57), yielded virtually unchanged results and is not reported in the table. Other regressors
included are country dummies and industry dummies, but we do not report their coeﬃcient estimates. Heteroskedasticity-consistent standard
errors are reported in parentheses. The Durbin-Wu-Hausman statistic tests the null that the use of instrumental variables does not change the
estimation outcome. We report IV estimates when the test is rejected at the 10 percent level or less. Instruments are: rule of law, legal origin,
total GDP, and population.

Table VI
Tests of Robustness, Focusing on the Interaction of Bank Concentration and Young Firms’
Financial Needs

42
0.284
1150
0.69

0.048*
(0.026)

(a)
-0.887***
(0.287)

0.288
1150
0.45

0.289
1150
0.58

Omitted variables
(b)
(c)
-0.914*** -0.905***
(0.286)
(0.285)
0.045**
0.028
(0.022)
(0.030)
0.059**
0.058*
(0.029)
(0.033)
0.012
(0.010)
0.005
(0.005)

0.416
984
0.00

0.032
(0.022)

(d)
-0.296**
(0.147)
0.005
(0.012)
0.035**
(0.014)

0.327
1148
0.14

-0.017
(0.038)
0.288
1150
0.50

Outliers
(e)
(f)
-0.816*** -0.905***
(0.260)
(0.286)
0.033*
0.049**
(0.017)
(0.023)
0.036*** 0.076***
(0.014)
(0.028)

* indicates rejection of the null at the 10 percent signiﬁcance level, ** indicates 5 percent signiﬁcance level, and *** indicates 1 percent signiﬁcance
level.

R2
Observations
Durbin − W u − Hausman

Highj · External dependencej · Bank concentrationk

External dependencej · Accounting Standardsk

External dependencej · Log of per capita GDPk

External dependencej · Stock marketk /GDPk

External dependencej · Bank concentrationk

External dependencej · Bank developmentk

Regressors
F raction of value addedj,k

The dependent variable in all regressions is the average (compounded) rate of growth of real value added for each industrial sector in each country
between 1980 and 1990. The fraction of value added is industry j’s share of manufacturing in country k; external ﬁnancial dependence refers to the
borrowing needs of establishments less than ten years old. Banking development is the ratio of private domestic credit to GDP. Bank concentration
is the three-bank ratio. Accounting standards is an index ranking the amount of disclosure of companies’ annual reports for each country. Because
of missing values in accounting standards, the inclusion of accounting standards in column (d) causes all records referring to the following countries
to be dropped: Bangladesh, Costa Rica, Jordan, Kenya, Morocco, Pakistan, Sri Lanka, and Zimbabwe. Column (e) is our baseline regression (as
in column (a) of Table VI), but we exclude the observations where the dependent variable was censored at growth = 100 percent per year. The
regression in column (f) includes High, a dummy variable equal to one when a sector’s external ﬁnancial need is above the median (0.60). Other
regressors included are country dummies and industry dummies, but we do not report their coeﬃcient estimates. Heteroskedasticity-consistent
standard errors are reported in parentheses. The Durbin-Wu-Hausman statistic tests the null hypothesis that the use of instrumental variables
does not change the estimation outcome. We report IV estimates when the test is rejected at the 10 percent level or less. Instruments are: rule of
law, legal origin, total GDP, and population.

Table VII
Tests of Robustness, Continued

Table VIII
Regressions with Financial Needs of Old Establishments
The dependent variable in all regressions is the average (compounded) rate of growth of real value
added for each industrial sector in each country between 1980 and 1990. In all regressions, the fraction
of value added is industry j’s share of manufacturing in country k; external ﬁnancial dependence refers
to the borrowing needs of establishments more than ten years old. Bank development is the ratio of
private domestic credit to GDP. Bank concentration is the three-bank ratio. The ﬁrst column is a
regression of the entire sample; the second regression only considers sectors whose share of value added
in manufacturing in their country is above the median, while the third column reports a regression of
sectors whose share of value added is below the median. Other regressors included are country dummies
and industry dummies, but we do not report their coeﬃcient estimates. Heteroskedasticity-consistent
standard errors are reported in parentheses. The Durbin-Wu-Hausman statistic tests the null that the
use of instrumental variables does not change the estimation outcome. We report IV estimates when
the test is rejected at the 10 percent level or less. Instruments are: rule of law, legal origin, total GDP,
and population.

Regressors
F raction of value addedj,k
External dependencej · Bank developmentk
External dependencej · Bank concentrationk
R2
Observations
Durbin − W u − Hausman

All ﬁrms
(a)
-0.898***
(0.282)
0.114***
(0.037)
0.100**
(0.048)
0.282
1112
0.60

Large sectors
(b)
-0.661***
(0.235)
0.303***
(0.108)
0.023
(0.090)
0.507
535
2.56

Small sectors
(c)
-5.945***
(1.790)
0.091**
(0.045)
0.120*
(0.066)
0.351
577
0.08

* indicates rejection of the null at the 10 percent signiﬁcance level, ** indicates 5 percent signiﬁcance
level, and *** indicates 1 percent signiﬁcance level.

44
0.253
952
1.57

0.044**
(0.019)
-0.015
(0.027)

0.030*
(0.017)

Contestability
(a)
-0.802**
(0.335)

0.253
952
0.36

-0.020
(0.027)

0.037**
(0.018)

0.022
(0.018)

Under-reporting
(b)
-0.810**
(0.334)

-0.056**
(0.025)
0.238
741
0.27

0.236
741
0.38

-0.035***
(0.012)

0.173
635
2.29

0.014
(0.011)
-0.023***
(0.009)

State ownership
(c)
(d)
(e)
-1.003*** -1.008*** -0.371***
(0.364)
(0.366)
(0.137)
0.020
(0.020)
0.046*
0.034
0.011
(0.024)
(0.023)
(0.020)
-0.060***
(0.015)
0.048***
0.035**
0.041***
(0.016)
(0.016)
(0.015)

* indicates rejection of the null at the 10 percent signiﬁcance level, ** indicates 5 percent signiﬁcance level, and *** indicates 1 percent signiﬁcance
level.

R2
Observations
Durbin − W u − Hausman

High Statek · Ext. dep.j

Ext. dep.j · Statek

Statek

High F oreign & Low per capita GDPk · Ext. dep.j

High F oreign & High Bank Concentrationk · Ext. dep.j

Ext. dep.j · Bank concentrationk

Bank concentrationk

Ext. dep.j · Bank developmentk

Bank developmentk

Regressors
F raction of value addedj,k

The dependent variable in all columns is the average (compounded) rate of growth of real value added for each industrial sector in each country
between 1980 and 1990. The fraction of value added is industry j’s share of manufacturing in country k; external ﬁnancial dependence refers to the
borrowing needs of establishments less than ten years old. Bank development is the ratio of private domestic credit to GDP. Bank concentration
is the three-bank ratio. High Foreign & High Bank Concentration is a dummy variable equal to one for countries with both a three-bank ratio
and a foreign share of bank assets above their median. High Foreign & Low per-capita GDP is a dummy variable equal to one for countries below
the median in per-capita GDP and above the median in the foreign bank share. State measures the share of assets of state-owned banks over
total commercial bank assets. High State is a dummy variable equal to one if the share of state-owned banks is above 60 percent. Industry- and
country-dummy variables are included in columns (a) through (d), but we do not report their coeﬃcient estimates. In column (e) also per-capita
GDP, stock market capitalization, and accounting standards are used as country controls (because the country dummies are excluded), but we
do not report their estimates. Heteroskedasticity-consistent standard errors are reported in parentheses. The Durbin-Wu-Hausman statistic tests
the null hypothesis that instrumental variables do not change the estimation outcome. We report IV estimates when the test is rejected at the 10
percent level or less. Instruments are: rule of law, legal origin, total GDP, and population.

Table IX
Contestability, Under-Reporting, and State Ownership

Bank powers
(b)
(c)
-0.387*** -0.382***
(0.139)
(0.140)
0.014
0.003
(0.017)
(0.018)
-0.111*** -0.113***
(0.022)
(0.027)
0.039***
0.052**
0.036*
(0.015)
(0.021)
(0.021)
-0.004
(0.005)
-0.001
0.000
-0.007
(0.005)
(0.005)
(0.005)
0.012
(0.014)
0.381
0.203
0.200
1035
984
984
0.34
2.34*
4.03***
(a)
-0.360**
(0.146)
0.012
(0.012)

0.074
876
25.08***

Growth
(d)
0.038
(0.137)
0.025
(0.020)
-0.143***
(0.032)
0.058**
(0.029)

0.442
961
4.47**

0.048**
(0.021)

in No.
(e)
-0.212
(0.172)
0.023
(0.015)

0.207
876
7.57***

0.444
990
2.52

Growth avg. size
(f)
(g)
-0.572** -0.940***
(0.244)
(0.361)
-0.010
0.005
(0.015)
(0.017)
0.015
(0.031)
-0.009
0.021
(0.026)
(0.018)

* indicates rejection of the null at the 10 percent signiﬁcance level, ** indicates 5 percent signiﬁcance level, and *** indicates 1 percent signiﬁcance
level.

R2
Observations
Durbin − W u − Hausman

Bank concentrationk · Bank P owersk

External dependencej · Bank P owersk

Bank P owersk

External dependencej · Bank concentrationk

Bank concentrationk

External dependencej · Bank developmentk

Regressors
F raction of value addedj,k

The dependent variable in columns (a) through (c) is the average (compounded) rate of growth of real value added for each industrial sector in
each country between 1980 and 1990. The dependent variable in columns (d)–(e) is growth in number of ﬁrms, and in columns (f)–(g) is growth
in average ﬁrm size. The fraction of value added is industry j’s share of manufacturing in country k; external ﬁnancial dependence refers to the
borrowing needs of establishments less than ten years old. Bank development is the ratio of private domestic credit to GDP. Bank concentration
is the three-bank ratio. Bank powers is a measure of regulatory restrictions on bank activities in the 1990s. Heteroskedasticity-consistent standard
errors are reported in parentheses. The Durbin-Wu-Hausman statistic tests the null that the use of instrumental variables does not change the
estimation outcome. We report IV estimates when the test is rejected at the 10 percent level or less. Instruments are: rule of law, legal origin, total
GDP, and population. Industry- and country-dummy variables are included in columns (a), (e), and (g), but we do not report their coeﬃcient
estimates. In columns (b), (c), (d) and (f), per-capita GDP, stock market capitalization, accounting standards, and bank development are used as
country controls (because the country dummies are excluded), but we do not report their estimates.

Table X
Bank Powers and Industry Concentration

Full text of Working Papers (Federal Reserve Bank of Chicago) : Banking Market Structure, Financial Dependence and Growth : International Evidence From Industry Data, Working Paper 1999-08

FRASER