Download Are Banks in more Concentrated Markets less

Are Banks in more Concentrated Markets less Stable?
Evidence from the EU-25.
Pieter IJtsma
∗
Laura Spierdijk†
January 26, 2015
Abstract
We use panel data from the EU-25 to estimate the effect of banking market
concentration on financial stability. We start with a replication of a study by Uhde
and Heimeshoff (Journal of Banking & Finance, 2009) and find that their results are
not robust. We proceed by using a more recent sample to estimate the relationship
between banking market concentration and financial stability, and find that the level
of analysis matters. Although a country-level analysis indicates a negative relationship between market concentration and stability, the sign of the estimated effect is
reversed when the analysis is performed at the bank level. We argue that the latter
approach is more appropriate because it allows to control for bank characteristics
and unobserved bank heterogeneity. The evidence supports the view that banking
market concentration has a positive effect on the stability of cooperative banks and
savings banks.
JEL classification: G21, G28, G34, L110, L16.
Keywords: Financial stability, market structure, systemic risk.
∗
Corresponding author. Department of Economics, Econometrics and Finance, Faculty of Economics
and Business, University of Groningen, P.O. Box 800, NL-9700 AV Groningen, The Netherlands. Phone:
+31 50 363 3782. E-mail: [email protected].
†
Department of Economics, Econometrics and Finance, Faculty of Economics and Business, University
of Groningen, P.O. Box 800, NL-9700 AV Groningen, The Netherlands. Phone: +31 50 363 3192. E-mail:
[email protected].
1
1
Introduction
In the last two decades, rapid consolidation has led to a rise in the degree of market
concentration in the banking sector of many developed economies (Group of Ten, 2001;
Vives, 2011). As a result, banking markets tend to be dominated by a small number of
large banks. In 2007, the combined market share of the largest 5 banks was over 50% in
the majority of countries from the EU-15 and consolidation has further increased after the
onset of the 2007-2008 financial crisis (Vives, 2011). Although bank mergers can increase
efficiency through economies of scale and scope, some argue that the associated increase in
market concentration has negative repercussions for financial stability. Recent government
interventions in the financial sector, which have often involved the acquisition of troubled
banks by healthier competitors, might thus have unintended negative consequences for
stability. For this reason, it is of crucial importance to policy-makers and supervisors to
understand the relationship between banking market concentration and financial stability.
The theoretical literature regarding this relationship is inconclusive, however, so that
its nature is ultimately an empirical question. The aim of this paper is therefore to
empirically assess the effect of banking market concentration on financial stability. Its
contribution to the existing literature is threefold: first, we attempt to replicate the results
of an earlier study performed by Uhde and Heimeshoff (2009) and test the robustness of
their findings. Our initial results are qualitatively similar to those of Uhde and Heimeshoff
(2009), who find a strong negative effect of market concentration on financial stability.
We show, however, that these findings are sensitive to outliers and are not robust to the
inclusion of country fixed effects. Second, we introduce a more recent sample and show
that the choice for the level of analysis matters. Whereas our country-level analysis suggets
a negative effect of market concentration on stability, the results are reversed when the
analysis is performed at the bank level. Finally, following Beck et al. (2013), we reconcile
the two approaches through the estimation of a weighted bank-level model. This approach
can be interpreted as a middle way between the country-level and bank-level analyses. The
results provide evidence in favor of the view that banking market concentration positively
affects the stability of cooperative banks and savings banks, whereas no evidence is found
2
for an effect on the stability of commercial banks. These findings are reassuring, since
they indicate that recent government interventions in the financial sector are unlikely to
have unintended negative consequences for financial stability. Overall, our study indicates
the importance of using bank-level data to obtain robust findings about the relationship
between market concentration and financial stability.
The rest of the paper is organized as follows: in section 2, the existing theoretical and
empirical literature is reviewed. Our methodology is elaborated upon in section 3, after
which a description of the data follows in section 4. Section 5 presents the models that
are estimated, as well as the obtained results. Section 6 concludes.
2
Literature review
As was mentioned in the introduction, the theoretical literature is inconclusive about the
effect of banking market concentration on financial stability. Moreover, the empirical
literature does not paint a clear picture either, since different studies give conflicting
evidence. We briefly review the literature below.
Within the theoretical literature, advocates of the concentration-stability view argue
that banks in more concentrated markets tend to be more stable for one of the following reasons. First, the charter value hypothesis maintains that a bank’s charter is more
valuable when it operates in a less competitive environment in which future profits are
expected to be high. To protect their charter, banks in more concentrated markets will engage less in excessively risky lending (Marcus, 1984; Chan et al., 1986; Keeley, 1990; Allen
and Gale, 2000, 2004; Repullo, 2004) and will screen loan applicants better (Cordella and
Yeyati, 2002; Hauswald and Marquez, 2006). Both outcomes are beneficial for financial
stability.1 Second, in more concentrated markets each bank serves and becomes informed
about a larger proportion of borrowers. As a result, banks in more concentrated markets
make more informed decisions and are less exposed to credit risk (Marquez, 2002). Third,
when the failure of a bank threatens the stability of the system, banks in more concen1
A crucial assumption behind this line of reasoning is that concentrated markets are less competitive.
This view is challenged by the contestability theory (Baumol, 1982; Corvoisier and Gropp, 2002) and the
efficiency theory (Demsetz, 1973; Smirlock, 1985; Berger, 1995).
3
trated markets will find it easier to reach an agreement to rescue the troubled bank in
order to and prevent financial contagion. In more diffuse markets, an agreement might
not be reached because of a coordination problem. Hence, contagion is less likely to occur
in concentrated markets (Sáez and Shi, 2004). Finally, some argue that it is easier to
monitor a system with only a few large banks than one with many small banks.2
Proponents of the concentration-fragility view, on the other hand, argue that banking
market concentration is detrimental to financial stability. First, if the level of competition
decreases with the degree of market concentration, banks in more concentrated markets
can charge higher loan rates. This aggravates moral hazard problems on the part of
borrowers, who will be induced to invest in more risky projects. As a result, the riskiness of
the bank’s asset portfolio increases (Boyd and De Nicoló, 2005; De Nicoló and Lucchetta,
2009). Second, banks in concentrated markets are more likely to be too-big-to-fail, which
gives rise to a moral hazard problem on the part of bank managers (Mishkin, 1999).3
Third, the ex-ante risk of financial contagion is higher in more concentrated markets, since
the probability that a particular bank is large enough to impact the rest of the system is
increasing in the degree of market concentration (Nier et al., 2007). Finally, some argue
that the supervision of concentrated banking markets is more difficult because banks in
such markets tend to be larger and more complex than their counterparts operating in
more diffuse markets (De Nicoló et al., 2004; Beck et al., 2006).
Empirical studies of the relationship between banking market concentration and financial stability tend to focus on either the bank level or the country level. Analyses of
the latter type typically look at real episodes of financial crises. Using the Demirgüç-Kunt
and Detragiache (2002) (D-D) crisis indicator, Beck et al. (2006) and Schaeck et al. (2009)
find that higher levels of banking market concentration are associated with a decrease in
the probability of the occurrence of a financial crisis. Von Hagen and Ho (2007) and Boyd
et al. (2010), however, argue that the D-D indicator is problematic because it is based on
2
Another often-mentioned argument in favor of the concentration-stability view is that banks in more
concentrated markets are larger and therefore better able to diversify idiosyncratic risk. However, de
Vries (2005) and Wagner (2010) show that diversification cannot raise the stability of the system as a
whole, even though it may increase the stability of individual banks.
3
Dam and Koetter (2012) provide evidence that bank managers who expect to be bailed out in case
of failure engage in more risk-taking behavior.
4
government responses to a crisis. As such, the indicator is potentially biased and defines
the start of a crisis systematically late. To overcome this problem, Boyd et al. (2010) use
proxies of financial crises that are based on sharp declines in lending. They find a positive
effect of banking market concentration on the probability of a crisis.
Although a focus on real episodes of crises is intuitively appealing, it has the important
drawback that an indicator variable does not provide information about the intensity of
a crisis or about the fragility of the system in the absence of a crisis. Most empirical
studies have therefore been performed at the bank level. De Nicoló and Kwast (2002)
find that the stock return correlations of large banks are increasing in the level of market
concentration, a result which offers support for the concentration-fragility view. As shown
by the overview in Table 1, most studies which use the z-score as a proxy of the failure
probability of individual banks report a negative association between concentration and
stability. The exception is a study by Berger et al. (2009), who find a positive relationship.
An interesting study, which bridges the gap between the country-level and bank-level
approaches, has been performed by Uhde and Heimeshoff (2009). In their paper, Uhde and
Heimeshoff use aggregated data to obtain a country-level z-score, which can be interpreted
as measuring the default probability of a country’s financial sector as a whole. Looking
at the EU-25 in the period between 1997 and 2005, they obtain very strong results which
indicate a negative relationship between market concentration and financial stability. We
therefore take a replication of their study as the starting point of our analysis.
3
Methodology
In accordance with Uhde and Heimeshoff (2009), we apply the z-score technique (Roy,
1952) to aggregate balance sheet data of banks from the EU-25. The z-score is a widely
used measure of financial stability, which combines information about a bank’s profitability, capital buffer and return volatility. It is defined as follows:
µ+k
E(r) + k
=
,
z=p
σ
V ar(r)
5
(1)
where r is the bank’s return on assets, k is its capital-to-assets ratio, µ is the expected
return on assets and σ is the return standard deviation. The z-score is thus a measure of
financial soundness. Indeed, if r ∼ N (µ, σ), we can write:
P (r < −k) = P
r−µ
µ+k
<−
σ
σ
= 1 − Φ(z),
(2)
where P is the probability that the bank will be insolvent after the next period and Φ is
the standard normal cumulative distribution function. With normally distributed returns,
the probability of insolvency is a thus decreasing function of the z-score. Moreover, even
when returns are not normally distributed, Chebyshev’s inequality ensures that:
P (r < −k) ≤
1
.
z2
(3)
The upper bound of the probability of insolvency is thus decreasing in the z-score.
Note from Equation (1) that the calculation of a bank’s z-score at a particular point
in time requires knowledge about the first and second moment of its return distribution in
the next period. Since only realized returns are observed, these moments are proxied by
the bank’s return in the last period and the variance of its return over the sample period,
respectively.4 The z-score of bank b in year t is thus calculated as:
zbt =
rbt + kbt
,
sb (rbt )
(4)
where rbt is the bank’s return in year t, kbt is its capital-to-assets ratio at the end of
year t and sb (rbt ) is the sample standard deviation of its realized returns. Although the
z-score measures the soundness of individual banks, Uhde and Heimeshoff (2009) apply
the measure to aggregated data. They calculate the aggregated (consolidated) z-score of
country i in year t as:
zit =
rit + kit
si (rit )
4
(5)
This approach is common in the literature. Alternatively, some papers use a rolling window to
obtain a time-varying estimate of the return variance. A rolling window requires a relatively long sample,
however, as the estimate becomes unreliable when the rolling window is too short.
6
where rit , kit and si (rit ) are the return, capital-to-assets ratio and return standard deviation of the country’s banking sector as a whole. The consolidated z-score can thus be
interpreted as indicating the solvency of a country’s entire banking sector.5
As measures of market concentration, we use the five-bank concentration ratio (CR5 )
and the Herfindahl-Hirschman index (HHI). The CR5 is defined as the combined market
share in terms of assets of the largest five banks operating in the country. Higher values
thus indicate a more concentrated market. Although the CR5 is a straightforward measure, its drawbacks are that the cut-off point of five banks is arbitrary and that the market
shares of all other banks in the country are ignored. Hence, the CR5 could be the same
for markets with rather different structures (Bikker, 2004). The HHI does not suffer from
an arbitrary cut-off point, but has the drawback that it is sensitive to the entrance of a
large number of small banks (Rhoades, 1995). According to Bikker (2004), differences in
the CR5 are mainly determined by the skewness of the distribution of bank size, whereas
differences in the HHI result mainly from differences in the number of banks operating in
the market. We therefore use both concentration measures to check the sensitivity of the
results. Note that both the CR5 and the HHI are measured at the country level. As is
common in the literature, we thus assume that countries represent banking markets.
In the remainder of this section, we elaborate upon the four components of our analysis.
These consist of a replication of Uhde and Heimeshoff (2009), the analysis of a more recent
sample, as well as the estimation of a bank-level and a weighted bank-level model.
Replication As mentioned above, our starting point is a replication of Uhde and
Heimeshoff (2009). More specifically, we try to replicate some of the estimates of their
base model, which are shown in Table 5 of their paper.6 We thus calculate consolidated
z-score according to Equation (5). In addition, we study the robustness of their results to
the omission of outliers and to the inclusion of country fixed effects, which are included
to control for unobserved time-invariant country heterogeneity.
5
Note that the consolidated z-score should not be interpreted as a measure of the probability that all
banks in a country will become insolvent. Whereas the banking sector as a whole must be insolvent when
all banks are insolvent, the reverse is not true.
6
We only replicate the OLS specifications of Uhde and Heimeshoff (2009), as the instruments in their
2SLS specifications are time-invariant and therefore not compatible with a fixed effects specification.
7
In addition to the consolidated z-score and a measure of market concentration, the
model of Uhde and Heimeshoff (2009) includes the following control variables: the rate
of real GDP growth, the level of GDP per capita, the change in the rate of inflation,
the first lag of the change in the real interest rate, the second lag of the rate of private
sector credit growth relative to GDP, and a moral hazard index. GDP growth and a high
GDP per capita are expected to have a positive effect on financial stability (Laeven and
Majnoni, 2003), whereas the effects of a positive change in the rate of inflation or the
real interest rate are theoretically ambiguous (Uhde and Heimeshoff, 2009). A high rate
of credit growth relative to GDP growth might indicate excessive lending. Its second
lag is thus expected to have a negative sign, as excessive lending in the past will have
a negative impact on stability (Dell’Ariccia and Marquez, 2006). Finally, high values
of the moral hazard index indicate the presence of a generous deposit insurance system,
which is expected to have a negative effect on stability because it gives rise to a moral
hazard problem on the part of bank managers (Demirgüç-Kunt and Detragiache, 2002).
In addition to the country controls, three bank controls are included in the model. These
are the logarithm of the net interest margin, the level of loan loss provisions and the costincome ratio.7 The net interest margin is a measure of profitability, which is expected to
have a positive effect on stability. A high level of loan loss provisions, on the other hand,
indicates high credit risk and is expected to have a negative effect. Finally, increases in
the cost-income ratio indicate decreases in efficiency, which are also expected to negatively
affect financial stability (Uhde and Heimeshoff, 2009).
More recent sample After replicating the analysis of Uhde and Heimeshoff (2009), we
extend their analysis by using a more recent sample period, as the coverage of the database
signficantly improves after 2004. In addition, we include slightly different variables, so as
to be more in line with the literature. First, we transform the z-score by first adding the
sample median and then taking the logarithm.8 Second, we take the levels of the rate of
7
Since we estimate a country-level model, the bank controls are aggregated by country.
This is done because the z-score is highly skewed when expressed in levels. The sample median is
added to avoid losing the observations with a negative z-score (Berger and Mester, 1997), while keeping
the results interpretable.
8
8
inflation and the real interest rate instead of their changes (Beck et al., 2006; De Nicoló
and Loukoianova, 2007; Boyd et al., 2010). Third, in line with Fu et al. (2014) we take
the ratio of loan loss provisions to total assets as a more natural measure of credit risk.
Finally, we take the level of the net interest margin instead of its logarithm, and vice
versa for real GDP per capita.
The possibility of reverse causality became apparent during the recent financial crisis,
when government actions in response to financial instability resulted in a higher degree
of market concentration in a number of countries. We take this into account by using a
2SLS procedure to estimate the model, using lagged values of the concentration measures
as instruments of contemporeaneous values. Furthermore, since the bank controls might
themselves be affected by changes in the degree of market concentration, their inclusion
into the model might understate the strength of the relationship between the degree of
market concentration and the z-score. The reason for this is that the effect of increased
market concentration on financial stability might operate through banks’ profitability,
effiency and credit risk. If this is the case, our estimates only represent the direct effect
of market concentration on financial stability, while we are interested in the total effect.
Leaving out the bank controls from the model, on the other hand, might lead to omitted
variable bias. To overcome this problem, we replace the three consolidated bank-level
variables with their first lags as a robustness check.
Bank-level analysis After analyzing the relationship between banking market concentration and consolidated z-scores, we estimate a model at the level of the individual bank,
using bank-specific z-scores as the measure of stability. A bank-level analysis has two
major benefits. The first is that the z-score has a more straightforward interpretation
when applied to bank-level data. Secondly, an analysis at the bank level allows to control
for bank characteristics such as size and bank type. This is important because larger
banks tend to exhibit higher insolvency risk (Boyd and Runkle, 1993; De Nicoló, 2000).
Moreover, commercial banks, savings banks and cooperative banks may be affected differently by changes in the degree of market concentration. First, cooperative and savings
banks are generally more focused on traditional financial intermediation than commercial
9
banks. Hence, they might be more vulnerable to changes in lending rates (Hesse and
Cihák, 2007). Second, commercial banks tend to distribute profits to their shareholders,
whereas cooperative banks and savings banks generally retain profits (Salas and Saurina, 2002; Fonteyne, 2007; Ayadi et al., 2010). Changes in profitability that result from
changes in the degree of market concentration can thus be expected to affect the stability
of cooperative banks and savings banks more heavily than that of commercial banks. Finally, commercial and savings banks tend to maximize profits, whereas cooperative banks
may have other objectives. The charter value hypothesis may thus be less relevant for
cooperative banks. (Ayadi et al., 2010; Fiordelisi and Mare, 2014)
In addition to the control variables used in the country-level analysis, we include the
following bank controls: the natural logarithm of total assets, which is a measure of bank
size, and the loan ratio, which measures the extent to which banks are specialized in
making loans as opposed to obtaining other sources of income. We expect a negative
coefficient for total assets (Boyd and Runkle, 1993; De Nicoló, 2000), while the effect
of changes in the loan ratio is not a-priori clear (Berger et al., 2009; Beck et al., 2013).
In contrast to the country-level analysis, the bank-level model includes unconsolidated
(bank-specific) values of the net interest margin, the loss provisions ratio and the costincome ratio. As explained above, we expect a positive coefficient for the net interest
margin, and negative coefficients for the loss provisions ratio and the cost-income ratio.
With the same reasoning as before, we estimate the model through 2SLS and include
first lags of the bank-level control variables as opposed to contemporaneous values. Finally, we acknowledge that savings banks and cooperative banks might behave differently
in response to changes in the degree of market concentration than their commercial counterparts by estimating the model separately for the subsamples of the three bank types.
Weighted model Although a bank-level analysis has clear benefits, its drawback is
that it gives equal importance to banks of very different sizes. This is problematic for two
reasons. First, the results might be driven by developments in countries with many small
banks, as the high number of observations from these countries gives them a large weight in
the analysis (Beck et al., 2013). Second, policymakers tend to be interested in the stability
10
of the banking sector as a whole, for which larger banks are clearly more relevant than
smaller ones. Hence, a bank-level analysis which gives equal weights to every bank might
give misleading estimates of the effect of market concentration on financial stability. We
therefore reconcile the country-level and bank-level approaches by estimating a weighted
bank-level model in which the observations are weighted according to bank size.9
4
Data
Bank balance sheet data are obtained from Bankscope. In our replication study, we use
data as similar as possible to that of Uhde and Heimeshoff (2009). Although their sample
runs from 1997 to 2005, Bankscope only provided data from 1998 onwards at the time
the data were collected. Hence, our replication sample covers the period from 1998 to
2005. We use data from 2005 onwards in the remainder of the analysis, as the coverage
of Bankscope significantly improves after 2004 (see Table 2).
Figure 1 shows the evolution over time of the consolidated z-scores of the five major
European economies in the period from 1998 to 2005 (panel a), and the period from 2005
to 2013 (panel b). The figures show substantial variation in the z-score, both between
countries and over time. They also illustrate differences in the evolution over time of the
z-score. As shown by the figure, the recent financial crisis resulted in a sharp decrease and
quick recovery of the z-score in France, Germany and the UK. In Italy and Spain, on the
other hand, the effects of the crisis are reflected in a steadier, longer-lasting decrease of
the z-score. Figure 2 shows the degree of banking market concentration in the five major
European economies, as measured by the CR5 (panel a) and the HHI (panel b). The
process of consolidation in these countries, especially after the 2007-2008 financial crisis,
can clearly be seen. The only exception is France, where neither of the concentration
measure show a tendency to increase over time.
Variable definitions and data sources are given in Table 3. Descriptive statistics of
9
Bank size is measured in terms of total assets. We take the average size of the bank over the sample
period because we need the weights to be time-constant, as time-varying weights are not compatible
with the within estimator. The procedure is comparable with that of Beck et al. (2013), who weigh each
observation with the inverse of the number of banks in the corresponding country.
11
the replication sample and of the more recent sample are shown in Tables 4a and 4b,
respectively. To deal with outliers, we drop all observations for which the value of any of
the bank-level variables (with the exception of total assets) is in the first or last percentile.
In addition, we drop observations which represent very clear outliers with respect to one of
the country-level variables.10 Furthermore, we correct some obvious mistakes in the data,
either through intrapolition or by looking at other data sources.11 Finally, we exclude
banks for which less than 5 observations are available.12
5
Empirical results
As was mentioned in setion 3, the first part of our analysis is performed at the country
level and starts with an attempt to replicate the results of Uhde and Heimeshoff (2009).
We then estimate a country-level model with a more recent sample, after which a banklevel model and a weighted bank-level model are estimated. We present the specification
and results of these model below.
Replication Similar to Uhde and Heimeshoff (2009), we estimate a random-effects
model in which the consolidated z-score serves as the dependent variable and a measure
of the degree banking market concentration is the main explanatory variable:
zit = αt + β1 cit +
X
βk xit,k + it .
(6)
k
In Equation (6), zit is the consolidated z-score of country i in year t and cit is the corresponding level of market concentration. The coefficient of interest is thus β1 , which
measures the effect of a unit change in the degree of market concentration on the consolidated z-score. Furthermore, xit,k refers to the value of control variable k in country i
at time t, while βk measures the effect of control variable k on the consolidated z-score.
Finally, the model includes a year-specific constant αt and an error term it , which is
10
A list of outliers is presented in Table 6.
Our dataset and code are avaiable upon request.
12
Estimates of the return standard deviation become unreliable when the number of observations is
too small.
11
12
assumed to consist of a country-specific time-invariant component and a time-varying
component capturing the remaining disturbance.
The estimated coefficients are shown in Table 5.
In accordance with Uhde and
Heimeshoff (2009), we obtain negative and significant estimates of the effect of market
concentration on stability when the CR5 is used as the measure of market concentration.
The estimates are economically meaningful and suggest that a percentage point increase
in the CR5 is associated with a decrease in the consolidated z-score of approximately 0.3.
The coefficient of the HHI is not significant, however. Moreover, the significance of the
coefficient of the CR5 disappears when the clearest outlier (Finland in 2002) is omitted
from the sample, as shown in Table 7a.13 As a further check, we leave out the remaining
outliers and include country fixed effects to control for unobserved time-invariant country
heterogeneity.14 The results are reported in Table 7b, which shows that the estimates of
the effect of market concentration are not significantly different from zero. The rate of
lagged credit growth, as well as two of the three consolidated bank-level controls, have
coefficients significantly different from zero at the 10 percent level. The coefficients have
their expected signs, which were discussed in section 3. Overall, we conclude that the
findings of Uhde and Heimeshoff (2009) are not robust to the exclusion of outliers, or to
the inclusion of country fixed effects.15
More recent sample We extend the country-level analysis by using a sample from the
period between 2005 and 2013. The following fixed-effects model is estimated:
zit∗ = αt + γi + β1 cit +
X
βk xit,k + it ,
(7)
k
where zit∗ is the transformed (consolidated) z-score and γi is a country-specific constant,
which is accounted for by the within estimator. As was explained in section 3, the
13
Since Bankscope covers only one Finnish bank in 2002, the z-score cannot reasonably be assumed to
represent the stability of the Finnish banking sector in that year.
14
A Hausman test rejects the null hypothesis that the fixed and random effects model have the same
coefficients, which indicates that the fixed effects specification is to be preferred.
15
When the fixed effects model is estimated without excluding the outliers, the coefficients remains
insigificant. These results are not reported here.
13
transformed z-score is defined as:
zit∗ = ln(zit + z m ),
(8)
where z m is the sample median of the z-score. The effect of a unit change in the degree
of banking market concentration on the z-score is thus:
∂zit
dzit ∂z ∗
= ∗ it = β1 (zit + z m ).
∂cit
dzit ∂cit
(9)
For a country with a z-score equal to the median, this gives:
∂zit
∂zit /∂cit
= β1 (zit + zit ) ⇐⇒ 2β1 =
.
∂cit
zit
(10)
Hence, by multiplying the estimate of β1 with 2 we obtain an estimate of the relative
change of the z-score of the median country due to a unit change in the degree of market
concentration.
The results are presented in Table 8, where columns (1) and (3) give the estimates obtained by OLS, while columns (2) and (4) present the estimates obtained by 2SLS. In the
latter, lagged values of the concentration measures are used as instruments of contemporaneous values, so that reverse causality is ruled out.16 The estimates indicate a negative
and significant effect of market concentration on the consolidated z-score when the HHI is
used as the measure of market concentration. This negative relationship is economically
meaningful: a percentage point increase in the HHI is associated with a decrease in the
z-score of the median country of approximately 3 (OLS) to 4 (2SLS) percent. When the
CR5 is used as the concentration measure, however, the estimates are not significant. Of
the control variables, only the net interest margin and the loss provisions ratio have significant coefficients, both of which are of the expected sign. As expected, the estimated
effect of market concentration on financial stability is stronger when the contemporaneous bank controls (panel a) are replaced by their lags (panel b). Note also the sharp
16
The F-statistics in the first stage are 87.29 (CR5 ) and 79.90 (HHI), well above 10. Hence, the
instruments are strong.
14
reduction in the estimated coefficients of the loss provisions ratio, which are significant
at the 1% level when the contemporaneous values are used, but become insignficant once
the first lags are included. The coefficient related to GDP growth, on the other hand,
becomes positive and significant once the bank controls are lagged. Overall, we conclude
that there appears to be some evidence in favor of the concentration-fragility view, but
that the evidence is not robust with respect to the measure of market concentration. We
therefore turn to a bank-level model to shed more light on the issue.
Bank-level model Our bank-level analysis consists of the estimation of the following
fixed effects model:
∗
= αt + γb + β1 cit +
zbt
X
βk xbt,k + bt ,
(11)
k
where subscripts i, b and t are country, bank and time indices, respectively. γb is thus
a bank-specific constant, which is included to control for unobserved time-invariant bank
heterogeneity and which is taken into account by the within estimator. As before, zit∗
refers to the transformed z-score while cit represents one of the two measures of market
concentration. Hence, multiplying the estimate of β1 by 2 gives the estimated the effect
of a unit change in the degree of market concentration on the z-score of a bank with a
z-score equal to the sample median. Finally, xbt,k refers to the value of control variable k
for bank b in year t, while αt is a time-specific constant.
The estimated coefficients are reported in Table 9a. In contrast to the results obtained by the country-level analysis, the estimates of the bank-level model suggest that
an increase in the degree of banking market concentration has a positive effect on financial stability. This effect is statistically significant at the 1% level and is economically
meaningful: a percentage point increase in the CR5 is associated with an approximate
0.5 (OLS) to 0.9 (2SLS) percent increase in the z-score of the median bank, whereas a
percentage point increase in the HHI is estimated to lead to an increase in the z-score of 2
(OLS) to 3 (2SLS) percent. The fact that the estimated effect changes sign when we move
from the country level to the bank level confirms the idea that it is important to control
for bank-level characteristics. Indeed, the coefficient related to bank size is statistically
15
significant at the 1% level and has the expected (negative) sign. For the median bank,
a percent increase in size is associated with a decrease in the z-score of approximately
0.25 percent. All other estimates have their expected sign as well, with the exception of
the coefficient related to GDP growth, which has a negative sign. A possible explanation
could be that banks have an incentive to increase their leverage ratio during times of high
economic growth.
Table 10a shows the results when the sample is split into subsamples according to
bank type. The estimates indicate a strong positive effect of market concentration on
the z-score of cooperative banks and savings banks, whereas the estimated effect is not
significantly different from zero for the subsample of commercial banks. This finding is
consistent with the view that cooperative and savings banks are more affected by changes
in the degree of market concentration than their commercial counterparts (Salas and
Saurina, 2002; Fonteyne, 2007; Hesse and Cihák, 2007; Ayadi et al., 2010).
Weighted model As a final exercise, we estimate the model in Equation (11) after
weighing each observation according to bank size. This weighted model can be interpreted
as a middle way between the bank-level and country-level models. Whereas the banklevel model looks at individual banks in isolation, thereby failing to take into account
differences in the systemic importance of banks, the country-level model looks at the
banking sector as a whole and does therefore not allow to control for bank characteristics.
The weighted model does not suffer from either of these shotcomings and could thus be
seen as combining the best of both approaches.
The estimated coefficients are reported in Table 9b. Although the estimates of the
concentration measures remain positive, they are much smaller and not significantly different from zero. This might be the result of the fact that commercial banks, for which no
significant effect was found in the unweighted model, tend to be larger than cooperative
banks and savings banks and subsequently have a larger weight in the analysis. Indeed,
as shown in Table 10b, the estimated effects of market concentration on the z-score are
significant and positive for the subsamples of cooperative banks and savings banks, but
insignificant for the subsample of commercial banks.
16
Overall, the results thus support the view that market concentration positively affects
the stability of cooperative banks and savings banks, whereas no strong evidence is found
for an effect on commercial banks. These findings contrast with the results of our countrylevel analysis, which suggests a negative overall effect of banking market concentration
on financial stability. Hence, our study highlights the importance of looking beyond
country-level data when analyzing financial stability.
6
Conclusion
This paper has investigated the causal relationship between banking market concentration
and financial stability. We have performed an analysis consisting of four components. In
the first, an attempt was made to replicate and investigate the robustness of the findings
of an earlier study by Uhde and Heimeshoff (2009). Similar to Uhde and Heimeshoff
(2009), we found a statistically significant negative relationship between concentration
and the consolidated z-score, a measure of the stability of a country’s financial sector.
We have shown, however, that this result is not robust to either the omission of outliers
or the inclusion of country fixed effects. In the second part of the analysis, we have used
a more recent sample, which runs from 2005 to 2013. Here, the results again indicate a
negative effect of banking market concentration on the consolidated z-score, although the
results are sensitive to the measure of market concentration.
Since a country-level model does not allow to control for bank characteristics such as
size and bank type, we continued our analysis with the estimation of a bank-level model.
The estimates of this model illustrate the importance of looking at bank-level data when
analyzing financial stability. First, contrary to the estimates of the country-level model,
the results suggest a positive effect of an increase in the degree of market concentration
on the stability of individual banks. Second, they show that market concentration has a
heterogeneous impact on different types of banks: whereas a positive and significant effect
was found for cooperative banks and savings banks, there is no evidence of a relationship
between the degree market concentration and the stability of commercial banks. These
findings are confirmed by the final part of our study, in which a bank-level model was
17
estimated in which the observations are weighted according to bank size.
Overall, our findings are in favor with the view that increased banking market concentration has a positive effect on financial stability. This finding is reassuring for policymakers, as they suggest that recent government interventions to safeguard financial stability are unlikely to give rise to future instability by increasing the degree of market
concentration. More importantly, the results indicate the importance of looking beyond
country-level data when analyzing financial stability. They indicate that the effect of a
change in the degree of market concentration might have a different impact on, say, the
German banking sector (which mainly consists of cooperative banks), compared with that
of the UK (which is dominated by commercial banks). A focus on the country level makes
it difficult to take such subtleties into account.
References
Allen, F. and Gale, D. (2000). Comparing Financial Systems. Cambridge University
Press.
Allen, F. and Gale, D. (2004). Competition and financial stability. Journal of Money,
Credit and Banking, 36:453–480.
Ayadi, R., Llewellyn, D. T., Schmidt, R. H., Arbak, E., and De Groen, W. (2010).
Investigating diversity in the banking sector in Europe: Key developments, performance
and role of cooperative banks. Technical report, Centre for European Policy Studies.
Baumol, W. J. (1982). Contestable markets: An uprising in the theory of industry structure. American Economic Review, 72:1–15.
Beck, T., De Jonghe, O., and Schepens, G. (2013). Bank competition and stability:
Cross-country heterogeneity. Journal of Financial Intermediation, 22:218–244.
Beck, T., Demirgüç-Kunt, A., and Levine, R. (2006). Bank concentration, competition,
and crises: First results. Journal of Banking & Finance, 30:1581–1603.
Berger, A. N. (1995). The profit-structure relationship in banking–tests of market-power
and efficient-structure hypotheses. Journal of Money, Credit and Banking, 27:404–431.
Berger, A. N., Klapper, L. F., and Turk-Ariss, R. (2009). Bank competition and financial
stability. Journal of Financial Services Research, 35:99–118.
Berger, A. N. and Mester, L. J. (1997). Inside the black box: What explains differences
in the efficiencies of financial institutions? Journal of Banking & Finance, 21:895–947.
18
Bikker, J. A. (2004). Competition and Efficiency in a Unified European Banking Market.
Edward Elgar.
Boyd, J. H. and De Nicoló, G. (2005). The theory of bank risk taking and competition
revisited. Journal of Finance, 60:1329–1343.
Boyd, J. H., De Nicoló, G., and Jalal, A. M. (2009). Bank competition, risk and asset
allocations. Working paper 143, IMF.
Boyd, J. H., De Nicoló, G., and Loukoianova, E. (2010). Banking crises and crisis dating:
Theory and evidence. Working paper, CESifo.
Boyd, J. H. and Runkle, D. E. (1993). Size and performance of banking firms: Testing
the predictions of theory. Journal of Monetary Economics, 31:47–67.
Chan, Y.-S., Greenbaum, S. I., and Thakor, A. V. (1986). Information reusability, competition and bank asset quality. Journal of Banking & Finance, 10:243–253.
Cordella, T. and Yeyati, E. L. (2002). Financial opening, deposit insurance, and risk in a
model of banking competition. European Economic Review, 46:471–485.
Corvoisier, S. and Gropp, R. (2002). Bank concentration and retail interest rates. Journal
of Banking & Finance, 26:2155–2189.
Dam, L. and Koetter, M. (2012). Bank bailouts and moral hazard: Evidence from Germany. Review of Financial Studies, 25:2343–2380.
De Nicoló, G. (2000). Size, charter value and risk in banking: An international perspective.
Discussion paper, Federal Reserve System.
De Nicoló, G., Bartholomew, P., Zaman, J., and Zephirin, M. (2004). Bank consolidation,
internationalization, and conglomeration: Trends and implications for financial risk.
Financial Markets, Institutions & Instruments, 13:173–217.
De Nicoló, G. and Kwast, M. L. (2002). Systemic risk and financial consolidation: Are
they related? Journal of Banking & Finance, 26:861–880.
De Nicoló, G. and Loukoianova, E. (2007). Bank ownership, market structure and risk.
Working paper, IMF.
De Nicoló, G. and Lucchetta, M. (2009). Financial intermediation, competition, and risk:
A general equilibrium exposition. Working paper, IMF.
de Vries, G. C. (2005). The simple economics of bank fragility. Journal of Banking &
Finance, 29:803 825.
Dell’Ariccia, G. and Marquez, R. (2006). Lending booms and lending standards. Journal
of Finance, 61:2511–2546.
Demirgüç-Kunt, A. and Detragiache, E. (2002). Does deposit insurance increase banking system stability? an empirical investigation. Journal of Monetary Economics,
49(7):1373–1406.
19
Demsetz, H. (1973). Industry structure, market rivalry, and public policy. Journal of Law
and Economics, 16:1–9.
Fiordelisi, F. and Mare, D. S. (2014). Competition and financial stability in European
cooperative banks. Journal of International Money and Finance, 45:1–16.
Fonteyne, W. (2007). Cooperative banks in Europe – policy issues. Working paper, IMF.
Fu, X. M., Lin, Y. R., and Molyneux, P. (2014). Bank competition and financial stability
in Asia Pacific. Journal of Banking & Finance, 38:64–77.
Group of Ten (2001). Report on consolidation in the financial sector.
Hauswald, R. and Marquez, R. (2006). Competition and strategic information acquisition
in credit markets. Review of Financial Studies, 19:967–1000.
Hesse, H. and Cihák, M. (2007). Cooperative banks and financial stability. Working
paper, IMF.
Jiménez, G., Lopez, J. A., and Saurina, J. (2013). How does competition affect bank
risk-taking? Journal of Financial Stability, 9:185–195.
Keeley, M. C. (1990). Deposit insurance, risk, and market power in banking. The American Economic Review, 80:1183–1200.
Laeven, L. and Majnoni, G. (2003). Loan loss provisioning and economic slowdowns: too
much, too late? Journal of Financial Intermediation, 12:178–197.
Marcus, A. J. (1984). Deregulation and bank financial policy. Journal of Banking &
Finance, 8:557 – 565.
Marquez, R. (2002). Competition, adverse selection, and information dispersion in the
banking industry. Review of Financial Studies, 15:901–926.
Mishkin, F. S. (1999). Financial consolidation: Dangers and opportunities. Journal of
Banking & Finance, 23:675–691.
Nier, E., Yang, J., Yorulmazer, T., and Alentorn, A. (2007). Network models and financial
stability. Journal of Economic Dynamics and Control, 31:2033–2060.
Repullo, R. (2004). Capital requirements, market power, and risk-taking in banking.
Journal of Financial Intermediation, 13:156–182.
Rhoades, S. A. (1995). Market share inequality, the HHI, and other measures of the
firm-composition of a market. Review of Industrial Organization, 10:657–674.
Roy, A. D. (1952). Safety first and the holding of assets. Econometrica, 20:431–449.
Sáez, L. and Shi, X. (2004). Liquidity pools, risk sharing, and financial contagion. Journal
of Financial Services Research, 25:5–23.
Salas, V. and Saurina, J. (2002). Credit risk in two institutional regimes: Spanish commercial and savings banks. Journal of Financial Services Research, 22:203–224.
20
Schaeck, K., Cihak, M., and Wolfe, S. (2009). Are competitive banking systems more
stable? Journal of Money, Credit and Banking, 41:711–734.
Smirlock, M. (1985). Evidence on the (non) relationship between concentration and
profitability in banking. Journal of Money, Credit and Banking, 17:69–83.
Uhde, A. and Heimeshoff, U. (2009). Consolidation in banking and financial stability in
Europe: Empirical evidence. Journal of Banking & Finance, 33:1299 – 1311.
Vives, X. (2011). Competition policy in banking. Oxford Review of Economic Policy,
27:479–497.
Von Hagen, J. and Ho, T.-K. (2007). Money market pressure and the determinants of
banking crises. Journal of Money, Credit and Banking, 39:1037–1066.
Wagner, W. (2010). Diversification at financial institutions and systemic crises. Journal
of Financial Intermediation, 19:373–386.
21
Figure 1: Consolidated z-scores of the banking sector in the five major European
economies in the period between 1998 and 2005 (a), and the period between 2005 and
2013 (b).
(a)
(b)
22
Figure 2: Five-bank concentration ratio (a) and Herfindahl-Hirschman index (b) of the
banking sector in the five major European economies.
(a)
(b)
23
24
Country
Bank
Country
Bank
Bank
Bank
Country
Country
Country
Bank
Bank
Bank & Country
Demirgüç-Kunt and Detragiache (2002)
De Nicoló et al. (2004)
Beck et al. (2006)
De Nicoló and Loukoianova (2007)
Berger et al. (2009)
Boyd et al. (2009)
Schaeck et al. (2009)
Uhde and Heimeshoff (2009)
Boyd et al. (2010)
Jiménez et al. (2013)
Fiordelisi and Mare (2014)
Fu et al. (2014)
Crisis dummy
Z-score
Crisis dummy
Z-score
Z-score
Z-score / Loan losses
Crisis dummy
Consolidated z-score
“Crisis indicators”
Non-performing Loans
Z-score
Crisis dummy & z-score
Dependent variable
Positive
Negative
Positive
Negative
Positive
Positive
Positive
Negative
Negative
Non-linear
Ambiguous
Negative
Effect
No
No
Yes
No
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
Instrument(s)
The reported effects are significant at the 10% level at the least. The last column indicates whether the study controls
for reverse causality by instrumenting the concentration measure with an exogenous variable.
Level
Paper
Table 1: Overview of empirical analyses of the effect of banking market concentration on financial stability.
25
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
Austria
120
127
147
161
169
209
235
241
258
260
242
244
247
244
236
Belgium
50
48
43
45
45
45
40
42
42
35
36
37
38
38
33
Cyprus
19
15
15
15
17
15
15
14
12
13
14
13
12
10
9
Czech Republic
16
17
18
16
15
15
17
18
18
18
17
18
17
19
19
Denmark
57
57
63
57
55
53
66
74
80
79
101
101
97
90
76
Estonia
4
4
4
4
4
4
5
5
6
6
5
5
6
7
7
Finland
4
4
3
2
1
3
6
9
7
10
12
13
14
19
27
France
206
205
201
189
170
160
166
211
201
199
199
189
206
211
204
Germany
1911 1863 1740 1616 1488 1373 1352 1636 1645 1634 1592 1603 1633 1655 1625
Greece
6
3
2
2
3
4
17
19
18
18
18
18
19
14
13
Hungary
17
20
22
19
21
20
22
24
24
22
23
23
23
21
23
Ireland
11
10
9
8
5
3
7
13
14
12
10
11
13
12
10
Italy
139
139
108
101
67
36
29
610
614
621
615
596
594
588
551
Latvia
8
8
9
10
11
14
17
17
17
17
16
17
18
17
18
Lithuania
6
7
8
7
7
8
8
9
10
10
10
12
12
11
9
Luxembourg
93
98
87
74
71
68
67
66
67
77
75
72
69
67
63
Malta
5
4
5
4
4
6
5
5
6
6
7
7
9
9
8
Netherlands
21
18
16
16
17
15
23
23
25
26
27
30
29
29
32
Poland
17
19
16
14
13
15
31
34
31
30
36
41
41
42
40
Portugal
14
12
12
8
5
5
6
18
23
24
26
25
27
27
24
Slovakia
13
9
11
11
12
12
12
13
11
12
13
12
12
12
12
Slovenia
15
15
15
12
10
11
12
17
17
16
17
19
18
18
17
Spain
36
24
21
20
16
11
50
172
175
151
158
171
166
162
136
Sweden
5
7
8
86
86
85
77
83
80
77
69
70
71
74
75
United Kingdom
80
76
70
67
67
64
84
103
99
97
99
100
109
117
116
Total
2873 2809 2653 2564 2379 2254 2369 3476 3500 3470 3437 3447 3500 3513 3383
The number of banks reported includes commercial banks, savings banks and cooperative banks which Bankscope designates as
“institutions” (consolidation codes C1, C2, U1 and A1)
Country\Year
Table 2: Number of banks covered by Bankscope.
.
172
31
8
18
68
7
22
187
1442
9
18
8
489
17
8
45
8
29
34
23
11
16
75
72
102
2919
2013
26
Combined market share in assets of largest 5 banks in the country
Sum of squared market shares in assets of all banks in the country
Explanatory variables
CR5
HHI
Bank controls
Assets
Loan ratio
Net interest margin
Loss provisions
Loss provisions ratio
Cost-income ratio
Total assets in US$
Ratio of outstanding loans to total assets (%)
Bank’s net interest revenu as a share of interest-bearing assets (%)
Loan loss provisions in US$
Loan loss provisions / total assets (%)
Ratio of overhead costs to total revenue (%)
Rate of real GDP growth (%)
Ratio of nominal GDP to population
Rate of growth of GDP deflator (%)
3-month money market interest rate minus GDP deflator (%)
Rate of growth of ratio of private sector credit to GDP (%)
First principal component of PCA of various deposit
insurance system features
(Capital ratio + roaa)/sd(roaa)
Return on average assets before taxes
Ratio of equity to total assets
Standard deviation of roaa over the sample period
Dependent variable
Z-score
Roaa
Capital ratio
Sd(roaa)
Country controls
GDP growth
GDP per capita
Inflation
Real interest rate
Credit growth
Moral hazard index
Definition
Variable
Bankscope
Bankscope
Bankscope
Bankscope
Bankscope
Bankscope
World Development Indicators
World Development Indicators
World Development Indicators
Eurostat, OECD
World Development Indicators
(Demirgüç-Kunt and Detragiache, 2002)
ECB statistics
ECB statistics
Author’s calculations
Bankscope
Bankscope
Author’s calculations
Source
Table 3: Variable definitions and data sources.
27
200
197
197
200
200
200
161
200
200
200
200
200
Z-score
CR5
HHI
GDP per capita
GDP growth
Δ inflation
Δ real interest rate
Credit growth
Moral hazard index
Net interest margin
Loss provisions
Cost-income ratio
22.09
0.58
0.11
20.49
3.63
-0.23
-0.36
4.26
1.32
2.92
1.12
0.67
Mean
18.36
0.20
0.08
13.65
2.35
2.01
2.36
10.92
1.40
1.49
3.25
0.29
Std. Dev.
-1.44
0.19
0.01
2.75
-1.55
-8.61
-8.01
-37.15
-0.79
0.13
-0.65
0.32
Min.
112.88
0.99
0.41
80.93
10.97
5.74
11.47
74.23
4.58
6.68
20.21
3.80
Max.
225
225
225
225
225
225
211
222
225
225
225
12067
12067
12057
12061
11983
12027
Bank level
Z-score
Total assets
Loan ratio
Net interest margin
Loss provisions ratio
Cost-income ratio
N
Country level
Z-score
CR5
HHI
GDP per capita
GDP growth
Inflation
Real interest rate
Credit growth
Net interest margin
Loss provisions ratio
Cost-income ratio
Variable
146.15
35.80
59.79
2.32
0.43
0.65
15.20
0.60
0.11
33.55
1.39
2.32
0.32
4.89
2.13
0.66
0.59
Mean
233.12
191.00
18.86
1.40
1.01
0.22
16.56
0.18
0.07
20.11
4.13
2.58
2.55
7.67
1.00
1.04
0.15
Std. Dev.
25.34
0.02
0.00
-3.52
-17.09
0.01
-2.36
0.22
0.02
7.17
-17.95
-3.83
-9.66
-23.31
0.65
-0.67
0.26
Min.
(b) More recent sample (2005-2013).
4946.77
3070.00
99.97
71.43
29.68
7.00
99.50
0.98
0.40
112.03
12.23
20.30
14.81
29.74
4.99
6.88
2.15
Max.
GDP per capita in thousands of US$. Loan loss provisions and total assets in billions of US$. GDP growth, inflation, the real interest rate, credit growth, the
net interest margin, the loss provisions ratio and the loan ratio in percentage terms. First lag of real interest rate and second lage of credit growth rate reported.
N
Variable
(a) Replication sample (1998-2005).
Table 4: Descriptive statistics of the two samples used in the analysis.
28
(t-2)
(log)
(t-1)
-0.218
(1.008)
-0.091
(0.623)
-0.189
(0.541)
-0.093
(0.146)
6.734
(5.517)
-0.766
(1.270)
10.797
(11.911)
-1.878
(2.376)
-32.842**
(13.252)
(1)
OLS
0.186
(0.223)
-0.088
(1.098)
-0.058
(0.614)
-0.173
(0.546)
-0.084
(0.128)
6.692
(5.075)
-0.758
(1.231)
10.545
(14.296)
-1.999
(2.598)
-31.074**
(13.086)
(2)
OLS
-0.145
(1.041)
-0.086
(0.531)
-0.200
(0.489)
-0.090
(0.140)
6.619
(6.331)
-0.776
(1.954)
11.179
(12.168)
-1.865
(2.433)
-58.968
(37.537)
(4)
OLS
-2.263
(2.197)
-0.073
(0.660)
-0.365
(0.355)
0.015
(0.305)
-0.070
(0.089)
-24.110**
(12.097)
(5)
OLS
Country fixed effects
No
No
No
No
Time dummies
Yes
Yes
Yes
Yes
Observations
159
159
159
159
Countries
23
23
23
23
R2
0.073
0.133
0.018
0.161
Estimated coefficients of Equation (6). Panel-bootstrapped standard
errors in parentheses: *** p<0.01, **p<0.05, * p<0.1. The numbering
of the colums refers to Table 5 of Uhde and Heimeshoff (2009). The
sample runs from 1998 to 2005 and includes all countries from the EU-25
with the exception of Malta and Cyprus, which drop out due to data
availability. Data sources and definitions are given in Table 3.
Moral hazard index
Cost-income ratio
Loan loss provisions
Net interest margin
Credit growth
Δ real interest rate
Δ inflation
GDP growth
GDP per capita
HHI
CR5
Dependent variable:
z-score
Table 5: Results of replication of Uhde and Heimeshoff (2009).
Year
1998
1999
1999
2002
2002
2003
2004
2005
2007
2008
2010
2010
2012
2013
Country
Estonia
Estonia
Slovenia
Denmark
Finland
Malta
Malta
Malta
Latvia
Latvia
Latvia
Lithuania
Greece
Slovenia
Replication
Replication
Replication
Replication
Replication
Replication
Replication
Replication
Recent
Recent
Recent
Recent
Recent
Recent
Sample
Δreal interest rate (t-1)
Δreal interest rate (t-1)
Cost-income ratio
Credit growth (t-2)
Z-score
Z-score
Z-score
Z-score
Inflation
Inflation
Real interest rate (t-1)
Real interest rate (t-1)
Credit growth (t-2)
Cost-income ratio
Variable
11.47
9.78
2.06
74.23
61.54
112.88
102.09
95.01
20.30
14.38
14.81
11.01
20.53
2.15
Value
Table 6: Overview of outliers identified in the two samples.
29
(t-2)
(log)
(t-1)
-0.071
(1.166)
-0.135
(0.561)
-0.075
(0.476)
-0.058
(0.132)
3.116
(4.836)
-0.769
(0.697)
4.291
(11.621)
-1.544
(2.450)
-12.914
(13.115)
(1)
OLS
0.202
(0.204)
0.053
(1.205)
-0.102
(0.667)
-0.052
(0.593)
-0.048
(0.141)
2.871
(4.231)
-0.761
(0.913)
3.722
(13.847)
-1.695
(2.381)
-11.829
(13.585)
(2)
OLS
-0.036
(1.143)
-0.129
(0.610)
-0.077
(0.531)
-0.055
(0.145)
2.945
(4.983)
-0.775
(1.398)
4.255
(12.757)
-1.413
(2.389)
-10.477
(39.666)
(4)
OLS
-2.002
(2.200)
0.050
(0.553)
-0.216
(0.301)
0.015
(0.228)
-0.048
(0.078)
-11.730
(12.367)
(5)
OLS
(t-2)
Cost-income ratio
Loan loss provisions
Net interest margin
Credit growth
Δ real interest rate
Δ inflation
GDP growth
GDP per capita
HHI
CR5
Dependent variable:
z-score
(log)
(t-1)
0.183
(0.293)
-0.118
(0.144)
-0.199
(0.164)
-0.090*
(0.053)
3.210*
(1.669)
-0.814**
(0.364)
3.810
(4.072)
-1.614
(12.849)
(1)
OLS
0.124
(0.191)
0.251
(0.311)
-0.102
(0.146)
-0.183
(0.159)
-0.085*
(0.049)
2.967*
(1.757)
-0.807*
(0.461)
3.318
(4.359)
-2.302
(11.494)
(2)
OLS
0.194
(0.306)
-0.114
(0.133)
-0.194
(0.154)
-0.087*
(0.049)
3.073*
(1.711)
-0.811*
(0.467)
3.632
(4.643)
10.787
(34.466)
(4)
OLS
0.317
(0.235)
-0.192
(0.197)
-0.119
(0.176)
-0.096*
(0.052)
1.270
(14.562)
(5)
OLS
(b) Remaining outliers omitted and country fixed effects included.
Country fixed effects
No
No
No
No
Country fixed effects
Yes
Yes
Yes
Yes
Time dummies
Yes
Yes
Yes
Yes
Time dummies
Yes
Yes
Yes
Yes
Observations
158
158
158
158
Observations
155
155
155
155
Countries
23
23
23
23
Countries
23
23
23
23
R2
0.022
0.097
0.001
0.123
R2
0.364
0.370
0.365
0.185
Estimated coefficients of Equation (6). Panel-bootstrapped standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1. The column numbers refer to those
of Table 5 in Uhde and Heimeshoff (2009). The sample runs from 1998 to 2005 and includes all countries from the EU-25 with the exception of Malta and
Cyprus, which drop out due to data availability. Data sources and definitions are given in Table 3.
Moral hazard index
Cost-income ratio
Loan loss provisions
Net interest margin
Credit growth
Δ real interest rate
Δ inflation
GDP growth
GDP per capita
HHI
CR5
Dependent variable:
z-score
(a) Finland in 2002 omitted.
Table 7: Results of robustness check of replication of Uhde and Heimeshoff (2009).
30
(t-2)
0.093
(0.118)
0.003
(0.005)
-0.007
(0.007)
0.001
(0.005)
-0.002
(0.001)
0.073
(0.044)
-4.811***
(1.857)
-0.170
(0.123)
-0.143
(0.286)
(1)
OLS
0.081
(0.122)
0.002
(0.005)
-0.007
(0.007)
0.001
(0.005)
-0.002
(0.001)
0.071
(0.044)
-4.852**
(1.936)
-0.170
(0.127)
-0.225
(0.334)
(2)
2SLS
-1.469**
(0.648)
0.004
(0.088)
0.002
(0.005)
-0.007
(0.006)
-0.000
(0.004)
-0.001
(0.001)
0.069**
(0.032)
-4.933***
(1.670)
-0.138
(0.108)
(3)
OLS
-2.176**
(0.948)
-0.049
(0.098)
0.002
(0.004)
-0.007
(0.007)
-0.001
(0.004)
-0.001
(0.002)
0.066**
(0.029)
-5.027***
(1.784)
-0.122
(0.108)
(4)
2SLS
(t-2)
Cost-income ratio
(t-1)
(t-1)
(t-1)
Loan loss provisions
Net interest margin
Credit growth
(t-1)
(log)
Real interest rate
Inflation
GDP growth
GDP per capita
HHI
CR5
Dependent variable:
transformed z-score
0.287*
(0.156)
0.014***
(0.004)
-0.001
(0.006)
0.002
(0.006)
-0.002
(0.002)
0.074*
(0.044)
-0.462
(1.622)
0.052
(0.105)
-0.091
(0.336)
(1)
OLS
0.247
(0.157)
0.014***
(0.004)
-0.002
(0.007)
0.002
(0.006)
-0.002
(0.002)
0.071
(0.047)
-0.474
(1.707)
0.054
(0.110)
-0.339
(0.446)
(2)
2SLS
(b) Lagged bank controls.
-1.511*
(0.803)
0.179
(0.139)
0.013***
(0.004)
-0.001
(0.006)
0.001
(0.006)
-0.001
(0.002)
0.072*
(0.038)
-0.376
(1.593)
0.065
(0.107)
(3)
OLS
-2.502**
(1.202)
0.098
(0.156)
0.013***
(0.004)
-0.000
(0.006)
-0.000
(0.005)
-0.001
(0.002)
0.070**
(0.034)
-0.322
(1.782)
0.074
(0.104)
(4)
2SLS
Country fixed effects
Yes
Yes
Yes
Yes
Country fixed effects
Yes
Yes
Yes
Yes
Time dummies
Yes
Yes
Yes
Yes
Time dummies
Yes
Yes
Yes
Yes
Observations
202
202
202
202
Observations
180
180
180
180
Countries
25
25
25
25
Countries
25
25
25
25
R2
0.481
0.481
0.551
0.534
R2
0.418
0.411
0.481
0.454
Estimated coefficients of Equation (7). Multiplying any coefficient by 2 gives the estimated effect of a unit increase in the corresponding explanatory variable on
the z-score of the median country. Panel-bootstrapped standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1. In the 2SLS specifications, lags of the
concentration indices are used as instruments of contemporaneous values. The sample runs from 2005 to 2013 and includes all countries from the EU-25. Data
sources and definitions are given in Table 3.
Cost-income ratio
Loss provisions ratio
Net interest margin
Credit growth
(t-1)
(log)
Real interest rate
Inflation
GDP growth
GDP per capita
HHI
CR5
Dependent variable:
transformed z-score
(a) Contemporaneous bank controls.
Table 8: Results of country-level analysis with the more recent sample.
31
(t-1)
0.565***
(0.036)
-0.005***
(0.001)
0.005***
(0.001)
-0.004***
(0.001)
-0.020
(0.028)
-0.124***
(0.009)
0.000
(0.000)
0.037***
(0.003)
-0.042***
(0.002)
-0.032*
(0.018)
0.252***
(0.028)
(1)
OLS
0.602***
(0.038)
-0.005***
(0.001)
0.005***
(0.001)
-0.003**
(0.001)
-0.048*
(0.027)
-0.122***
(0.009)
0.000*
(0.000)
0.037***
(0.003)
-0.039***
(0.002)
-0.025
(0.018)
0.441***
(0.036)
(2)
2SLS
0.945***
(0.154)
0.562***
(0.036)
-0.005***
(0.001)
0.005***
(0.001)
-0.005***
(0.001)
-0.003
(0.028)
-0.125***
(0.009)
0.000
(0.000)
0.037***
(0.003)
-0.043***
(0.002)
-0.035**
(0.018)
(3)
OLS
1.454***
(0.207)
0.587***
(0.037)
-0.005***
(0.001)
0.005***
(0.001)
-0.004***
(0.001)
-0.014
(0.028)
-0.125***
(0.009)
0.000
(0.000)
0.037***
(0.003)
-0.042***
(0.002)
-0.032*
(0.018)
(4)
2SLS
Cost-income ratio
(t-1)
(t-1)
(t-1)
Loss provisions ratio
Net interest margin
(t-1)
(log) (t-1)
(t-2)
Loan-asset ratio
Total assets
Credit growth
(t-1)
(log)
Real interest rate
Inflation
GDP growth
GDP per capita
HHI
CR5
Dependent variable:
transformed z-score
-0.005
(0.117)
-0.002
(0.003)
-0.000
(0.002)
-0.005*
(0.003)
-0.036
(0.063)
-0.050**
(0.024)
0.002***
(0.001)
0.037***
(0.010)
-0.061***
(0.015)
0.059
(0.053)
0.136
(0.089)
(1)
OLS
-0.001
(0.118)
-0.002
(0.003)
-0.000
(0.002)
-0.005*
(0.003)
-0.035
(0.064)
-0.050**
(0.024)
0.002***
(0.001)
0.037***
(0.010)
-0.061***
(0.015)
0.059
(0.053)
0.175
(0.114)
(2)
2SLS
(b) Weighted model.
0.267
(0.299)
-0.009
(0.119)
-0.002
(0.003)
-0.001
(0.002)
-0.006**
(0.003)
-0.034
(0.061)
-0.052**
(0.024)
0.002***
(0.001)
0.037***
(0.009)
-0.060***
(0.015)
0.061
(0.054)
(3)
OLS
0.266
(0.389)
-0.009
(0.121)
-0.002
(0.003)
-0.001
(0.002)
-0.006**
(0.003)
-0.034
(0.060)
-0.052**
(0.024)
0.002***
(0.001)
0.037***
(0.009)
-0.060***
(0.015)
0.061
(0.053)
(4)
2SLS
Bank fixed effects
Yes
Yes
Yes
Yes
Bank fixed effects
Yes
Yes
Yes
Yes
Time dummies
Yes
Yes
Yes
Yes
Time dummies
Yes
Yes
Yes
Yes
Observations
21,036
21,036
21,036
21,036
Observations
21,036
21,036
21,036
21,036
Banks
2,951
2,951
2,951
2,951
Banks
2,951
2,951
2,951
2,951
R2
0.368
0.365
0.367
0.366
R2
0.258
0.258
0.255
0.255
Estimated coefficients of Equation (11). Panel (a) gives the coefficients of the bank-level model, whereas panel (b) gives the coefficients when the observations
are weighted according to bank size. Multiplying any coefficient by 2 gives the estimated effect of a unit increase in the corresponding explanatory variable on
the z-score of the median bank. Clustered standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1. In the 2SLS specifications, lags of the concentration
indices are used as instruments of contemporaneous values. The sample runs from 2005 to 2013 and includes all commercial banks, cooperative banks and savings
banks from the EU-25 of which at least 5 observations are available. Data sources and definitions are given in Table 3 and descriptive statistics in Table 4b.
Cost-income ratio
(t-1)
(t-1)
Loss provisions ratio
Net interest margin
(t-1)
(log) (t-1)
(t-2)
Loan-asset ratio
Total assets
Credit growth
(t-1)
(log)
Real interest rate
Inflation
GDP growth
GDP per capita
HHI
CR5
Dependent variable:
transformed z-score
(a) Bank-level model.
Table 9: Results of bank-level model and weighted model.
32
Yes
Yes
-0.105
(0.085)
Yes
Yes
-0.129
(0.122)
(2)
2SLS
Yes
Yes
3,777
622
0.133
Yes
Yes
-0.011
(0.083)
(2)
2SLS
-0.241
(0.303)
Yes
Yes
(3)
OLS
Yes
Yes
3,777
622
0.133
-0.012
(0.161)
Yes
Yes
(3)
OLS
-0.328
(0.438)
Yes
Yes
(4)
2SLS
Yes
Yes
3,777
622
0.133
-0.072
(0.217)
Yes
Yes
(4)
2SLS
Yes
Yes
12,469
1,660
0.523
Yes
Yes
0.578***
(0.054)
(6)
2SLS
Yes
Yes
0.378***
(0.075)
(5)
OLS
Yes
Yes
0.502***
(0.089)
(6)
2SLS
1.167*
(0.618)
Yes
Yes
(7)
OLS
Yes
Yes
12,469
1,660
0.521
1.925***
(0.446)
Yes
Yes
(7)
OLS
(b) Weighted model.
Yes
Yes
12,469
1,660
0.525
Yes
Yes
0.410***
(0.041)
(5)
OLS
2.031***
(0.529)
Yes
Yes
(8)
2SLS
Yes
Yes
12,469
1,660
0.519
2.742***
(0.566)
Yes
Yes
(8)
2SLS
Yes
Yes
0.656***
(0.203)
(9)
OLS
Yes
Yes
4,790
669
0.504
Yes
Yes
0.699***
(0.089)
(9)
OLS
Yes
Yes
0.612**
(0.250)
(10)
2SLS
Yes
Yes
4,790
669
0.497
Yes
Yes
1.138***
(0.093)
(10)
2SLS
2.475***
(0.915)
Yes
Yes
(11)
OLS
Yes
Yes
4,790
669
0.498
2.567***
(0.486)
Yes
Yes
(11)
OLS
2.291***
(0.856)
Yes
Yes
(12)
2SLS
Yes
Yes
4,790
669
0.494
4.073***
(0.807)
Yes
Yes
(12)
2SLS
Bank fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Time dummies
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
3,777
3,777
3,777
3,777
12,469
12,469
12,469
12,469
4,790
4,790
4,790
4,790
Banks
622
622
622
622
1,660
1,660
1,660
1,660
669
669
669
669
R2
0.445
0.444
0.443
0.443
0.507
0.502
0.489
0.474
0.413
0.413
0.413
0.413
Estimated coefficients of Equation (11) when the sample is split into subsamples including only commercial banks (columns 1-4), cooperative banks (5-8) or
savings banks (9-12). Coefficients of the (unweighted) bank-level model in panel (a) and those of the weighted model in panel (b). The control variables are the
same as those in Table 9. Multiplying any coefficient by 2 gives the estimated effect of a unit increase in the corresponding explanatory variable on the z-score
of the median bank. Clustered standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1. In the 2SLS specifications, lags of the concentration indices are
used as instruments of contemporaneous values. The subsamples run from 2005 to 2013 and includes all commercial banks, cooperative banks or savings banks
from the EU-25 of which at least 5 observations are available. Data sources and definitions are given in Table 3 and descriptive statistics in Table 4b.
Country controls
Bank controls
HHI
CR5
(1)
OLS
Yes
Yes
3,777
622
0.133
Bank fixed effects
Time dummies
Observations
Banks
R2
Dependent variable:
transformed z-score
Yes
Yes
-0.018
(0.058)
(1)
OLS
Country controls
Bank controls
HHI
CR5
Dependent variable:
transformed z-score
(a) Bank-level model.
Table 10: Results of bank-level model for different subsamples according to bank type.