Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
A New Data Set On Competition In National BankingMarkets
BY SOFRONIS CLERIDES, MANTHOS D. DELIS, AND SOTIRIOS KOKAS
We estimate the degree of competition in the banking sectors of 148 countries over theperiod 1997–2010 using three methods: the Lerner index, the adjusted Lerner index, andthe profit elasticity. Marginal cost estimates required for all methods are obtained using aflexible semi-parametric methodology. All three indices show that competitive conditionsin banking deteriorated during the period 1997–2006, improved until 2008, and deterioratedagain thereafter. Levels of competition differ across regions and income groups, but there isgradual convergence over time. Banking system is less competitive in sub-Saharan Africaand low income countries and more competitive in Europe and Central and South Asia andOECD countries.
Keywords: Bank market power, competition, world sample.JEL Classification G2, D40.
I. INTRODUCTION
The issue of banking sector competition has attracted much interest in recentyears, not least because of the financial crisis. Alongside the standard objectiveof achieving effective market competition, the issue has additional significance inbanking because of the sector’s crucial role in the allocation of credit. Economicdevelopment relies upon financial intermediation services that link users withproviders of capital and allow consumption decisions to be smoothed over time.There are at least three key questions of interest in the theoretical and empiricalliterature. First, does the degree of bank competition impact the efficiency of creditallocation (Cetorelli and Strahan, 2006; Bonnacorsi and Dell’Ariccia, 2004)?Second, how does the intensity of competition affect the financial system’s stability(Keeley, 1990; Boyd and De Nicolo, 2005)? Third, what are the macroeconomicoutcomes of banking-sector competition (Cetorelli, 2004)?
To address these important questions, one first needs to come up with mean-ingful measures of the intensity of bank competition. Some studies (e.g., Cetorelliand Strahan, 2006) followed the early industrial organization tradition and usedconcentration measures as a proxy for bank competition. There is however agrowing consensus that concentration measures – useful as they may be in re-vealing important structural characteristics of the industry – are not good proxiesfor bank competition (Beck, Demirguc-Kunt and Levine, 2006; Claessens andLaeven, 2004). Instead, recent literature favors measures like the price-cost mar-gin or Lerner index (Lerner, 1934), the adjusted-Lerner index (Koetter, Kolari andSpierdijk, 2012), the H statistic (Panzar and Rosse, 1987), and the profit elasticity(Boone, Griffith and Harrison, 2005). Computing any of these measures (with the
Corresponding author: Sotirios Kokas, Essex Business School. Email: [email protected].
C© 2015 New York University Salomon Center and Wiley Periodicals, Inc.
268 Sofronis Clerides et al.
exception of the H statistic) requires knowledge of marginal cost. Given that dataon marginal cost are not usually available, an important first step in the construc-tion of a competition index is the estimation of marginal cost using econometricmethods.
The objective of this paper is to present a new set of estimates for bank competi-tion in 148 countries for the period 1997–2010 and to analyze trends and patternsin the evolution of competition during that period. Competition is measured interms of the Lerner index, the adjusted-Lerner index and the profit elasticity. Thepaper contributes to the literature on bank competition in four important ways.First, we use a novel technique to obtain accurate estimates of marginal cost andmarket power for each observation in the sample. The smooth coefficient modelwe employ is a semiparametric approach that allows for a flexible cost structure.Second, we have the broadest coverage compared to existing studies, with 148countries from 1997 to 2010. While most studies focus on a specific region orincome group, our data set allows studying the degree of within-country bankcompetition in a large number of countries. With coverage up to 2010, we are ableto analyze the evolution of bank competition during the financial crisis. Third, wecompute three different indicators favored by the recent banking literature, namelythe Lerner index, the adjusted-Lerner index and the profit elasticity. Fourth, weclean the raw data required to estimate these indices from double-counting stem-ming from mergers and acquisitions, ownership issues, inflexible features of theBankscope database, and other problems.
All indices indicate substantive changes in bank competition over time andacross income and regional classification. The Lerner index, the adjusted-Lernerindex and the profit elasticity produce similar overall patterns of competition overtime. We observe a gradual decline in the intensity of competition from 1997 to2006. The trend reverses for a brief period (2007–2008) but then market powerincreases again in 2009 and 2010. This pattern raises interesting questions aboutthe relationship between competitive conditions and financial stability. Notably,bank competition seems to weaken during the upward phase of the businesscycle and to become more intense when economic conditions worsen. We do notformally explore this link in the current work but leave it open as an importantquestion for future research.
The empirical results also highlight important differences across regions andincome groups. On average, the banking systems of Sub-Saharan Africa are theleast competitive, followed by East Asia and Pacific. Europe and Central andSouth Asia have the most competitive banking sectors. Non-OECD countries witheither high or low income levels have less competitive banking sectors than middle-income countries. OECD countries have lower bank market power overall. Despitethe differences, competition levels across income and regional groups seem to beconverging over time.
The three competition measures largely coincide when it comes to the mainoverall trends but do differ when compared at a finer level. We do not take astand on which index is best. Each measure has its own theoretical interpretation
A New Data Set On Competition In National Banking Markets 269
and its calculation presents different empirical challenges. Rather than propose aspecific measure, we provide (section 2) an extensive discussion of the theoreticaljustification for each index that we hope will be a useful guide to potential usersseeking to identify their preferred measure.
The structure of the paper is as follows. Section II presents an overview of bankcompetition measures. Section III describes our data and the empirical method-ology. Section IV presents the empirical results on the various indices of bankcompetition. Section V concludes.
II. EMPIRICAL METHODS FOR THE ESTIMATION OFCOMPETITION
Estimation of competition has a long history in industrial organization, datingback to Lerner (1934). Lerner defined his “index of the degree of monopoly power”(or “degree of monopoly” for short) as
Lerneri = Pi − mciPi
, (1)
where Pi and mci are firm i’s price and marginal cost respectively. The indexranges between 0 and 1, with zero corresponding to perfect competition and largervalues reflecting a greater degree of market power. The Lerner index remains tothis day the most widely and popular measure of market power and the intensityof competition.1 The main reasons for its popularity are simplicity, an intuitiveinterpretation, and relatively mild data requirements. The Lerner index measuresthe ability of an individual bank to charge a price above marginal cost. Whendata on prices and marginal cost are available, the Lerner index can be computedas a simple ratio. Because marginal cost is rarely available, researchers have toeither estimate a cost function or impose equilibrium conditions derived fromtheoretical models in order to obtain marginal cost estimates. The former methodis usually used in the banking literature, while the latter is preferred by industrialorganization economists.2 In this paper we will follow the banking literatureapproach of estimating a cost function.
Koetter, Kolari and Spierdijk (2012) have argued that the conventional approachof computing the Lerner index assumes both profit efficiency (optimal choice ofprices) and cost efficiency (optimal choice of inputs by firms). As a result, theestimated price-cost margins do not correctly measure the true extent of marketpower. The argument reflects a distinction pointed out by Lerner (1934) himself,who states that “for practical purposes we must read monopoly power not aspotential monopoly, but as monopoly in force” (p. 170; italics as in the original).
1Throughout the paper we will use market power as a measure of competition intensity. This is usuallyan innocuous assumption, although research has shown that the Lerner index does not always point inthe expected direction when competitive conditions change (Stiglitz, 1989; Boone, 2008).2The use of equilibrium conditions is the hallmark of the New Empirical Industrial Organization(NEIO) literature; see Bresnahan (1989) for an overview.
270 Sofronis Clerides et al.
In other words, the Lerner index measures actual (exercised) market power, whileKoetter, Kolari, and Spierdijk (2012) are interested in measuring potential marketpower. To that end, the authors propose an adjustment that results in the efficiency-adjusted Lerner index:
adjusted Lerneri = πi + tci − mci · qiπi + tci , (2)
where π i is the profit of bank i, tci is total cost, mci is marginal cost and qi is totaloutput. Like the standard Lerner index, the adjusted Lerner ranges from 0 to 1,with larger values indicating greater market power.
Boone (2008) proposes a new competition measure called relative profit differ-ences (RPD). He shows that in most theoretical models of oligopolistic competi-tion, the difference in profit between more and less efficient firms is increasing inthe intensity of competition. Consider an increase in competition – due perhapsto a decrease in entry costs or to goods becoming closer substitutes. It will causethe profits of all firms to decline but inefficient firms will suffer a greater declinethan efficient firms. In other words, the profits of an efficient firm will increaserelative to the profits of a less efficient firm as increased competition will penalizethe least efficient firm more severely.
RPD is a theoretical construct that is difficult to compute in practice. Boone,Griffith and Harrison (2005) have proposed the profit elasticity (PE) as an empiricalanalogue of RPD. The PE is the percentage decrease in profits resulting from aone percent increase in the marginal cost:
profit elasticityi =∂ ln πi∂ ln mci
. (3)
The PE (sometimes called the Boone indicator) should be negative becauseprofits and marginal cost are inversely related. An efficient firm will have a lowerPE (in absolute value) as its profits will be less affected by a change in marginalcost. For example, if PE = −0.2, a 1 percent increase in the marginal cost (dueto a decrease in the efficiency level) will decrease its profits by 0.2 percent. IfPE = −0.5, a 1 percent increase in the marginal cost will decrease its profits by0.5 percent. A large absolute value of the PE can be interpreted as a reduction inthe ability of the bank to contain its losses due to an increase in competition. Theprofit elasticity therefore links bank performance to differences in efficiency (interms of marginal cost).
Boone, Griffith and Harrison (2005) suggest that the PE can be estimated fromthe equation:
ln πi = α + βi ln mci . (4)The coefficient β i is the desired profit elasticity index of market power. Intu-
itively, equation (4) examines the relationship between profits and marginal cost;or, differently phrased, the indirect relationship between profits and efficiency.The profit elasticity can take any value and is therefore a continuous indicator of
A New Data Set On Competition In National Banking Markets 271
market power. It has already been used in empirical applications in various indus-tries (e.g., Delis, 2012; Van Leuvensteijn, Sorensen, Bikker and van Rixtel, 2013;Boone and Van Leuvensteijn, 2010). The index has also received some criticism;for example, Schiersch and Schmidt-Ehmcke (2010) show that the profit elasticitymakes critical assumptions relative to firm size (the biggest firms are assumed tobe the most efficient) and relative to the definition of the extent of the market.
It is worth pointing out that a measure of the profit elasticity can be derivedfrom the adjusted-Lerner index by solving for π in equation (2) and differentiatingwith respect to marginal cost. The result is
profit elasticityi =qi · mci
qi · mci − tci (1 − ad justed Lerneri ) . (5)
Hence, the adjusted Lerner and profit elasticity are two closely related concepts.Many older banking studies measured competition with the so-called H statistic.
The statistic is based on the test for “monopoly” equilibrium proposed by Panzarand Rosse (1987; quotation marks included in the paper’s title). Panzar and Rosse(henceforth PR) show that under very general conditions, the sum of the factorprice elasticities of a monopolist’s reduced form revenue equation must be non-positive. They further show that in commonly used oligopolistic models this sumis strictly positive; they note however that this last result is based on relativelysimple oligopoly models and is therefore less general than the monopoly result.They also point out that a monopoly facing a perfectly elastic demand curve canhave a negative H statistic under some conditions, even though it has no marketpower. They conclude that “a rejection of the hypothesis that H < 0 must meanthat the revenue functions of the observed firms are influenced by the actions ofothers” (p. 447).
PR presented the H statistic to be a test for monopoly behavior and avoidedinterpreting it as a measure of competition.3 Nonetheless, their findings gave riseto a large literature – mostly in banking – that used that H statistic as a measureof competition. This literature has been recently criticized by Bikker, Shaffer,and Spierdijk (2012), who show that much of the analysis using scaled revenueequations leads to misleading results, while even “the unscaled P-R test is aone-tail test of conduct.” They conclude that “the P-R revenue test results in anonordinal statistic for firm conduct that is less informative than prior literaturehas suggested.” Based on these findings and on our own reservations due to thelack of a clear theoretical justification, we decided not to compute the H statisticin this paper.
The Lerner, adjusted-Lerner and profit elasticity indices each have their ownmerits and drawbacks. Simplicity and direct applicability to empirical studies arestrong arguments in favor of the Lerner and adjusted-Lerner indices, while theexistence of a solid theoretical foundation justifies the use of the profit elasticity
3They did show that with constant elasticity demand and a CRS Cobb-Douglas technology, the Hstatistic is equivalent to the Lerner index: L = H/(H-1).
272 Sofronis Clerides et al.
index. We do not take a stand on which measure better reflects competition. We usethe three indices described above to measure banking-sector competition in a broadsample of 148 countries. We make all estimates available to the wider researchcommunity and let researchers decide which indicator is preferable, given theobjectives of their research agenda. In fact, we will claim in the next section thatprobably the most important issue underlying all indices is the robust estimationof marginal cost at the bank-year level, which is required with all approaches.
III. DATA AND METHODOLOGY
DATA
Like many other banking studies, we rely on Bankscope as our primary sourceof bank-level data. Our sample covers the period 1997–2010. We exclude earlieryears because of concerns associated with coverage and accounting issues. Ourfocus is on commercial banks, savings banks and cooperative banks. We excludereal-estate and mortgage banks, investment banks, other non-banking credit institu-tions (mainly operating in Germany), specialized governmental credit institutions,bank-holding and other holding companies.4 Besides bank-holding companies, theexcluded institutions are less dependent on the traditional intermediation functionand have a different financing structure compared to our focus group. In short,our focus in this study is on banks carrying out traditional banking activities.Inclusion of bank-holding companies could lead to double counting, as these arecorporations controlling one or more banks. We always check that we have thesubsidiaries of these companies in the sample to avoid false exclusion of somebanks. Further, we have included in our sample only countries that had at leastthree banks in each year of our panel.
We apply three further selection rules to avoid including duplicates in oursample. This is an essential part of the sample-selection process that is absentfrom most empirical studies using the Bankscope database (for a similar strategywith ours, see Claessens and van Horen, 2014). First, even though we do notinclude bank-holding companies, we still need to exclude double entries betweenparent banks and subsidiaries.5 Bankscope’s consolidation code system allowsdownloading either consolidated or unconsolidated statements, but in some casesinformation on either unconsolidated or consolidated statements of certain banksis not available.6 We use either the consolidated or the unconsolidated statementdepending on which one is available. This is a non-trivial process that requiresthe re-examination of all banks on an individual basis to avoid double-counting.
4The main activities of excluded financial institutions relate to the following: provide mortgages; assistcorporations and governments in a range of services (e.g., M&As, raising capital, etc.); provide creditto public sectors; provide funding for public or municipal projects.5For the determination of the subsidiary we follow the definition generally applied in the literature,i.e. 50% or more of the shares of the subsidiary owned by the parent bank.6A consolidated statement is the statement of a bank integrating the statements of its subsidiaries orbranches. An unconsolidated statement does not integrate subsidiaries.
A New Data Set On Competition In National Banking Markets 273
Notably, there are cases of banks with subsidiaries in domestic or in foreigncountries and one should be very careful to avoid double-counting of subsidiariesthat are established, for example, in a foreign country.7
Second, we account for mergers and acquisitions (M&As). We went throughall the M&As one-by-one and made sure that both banks appear separately inthe sample before the M&A and only the merged entity or the acquiring bank isincluded in the sample after the event. For example, if bank A and bank B mergedin 2005, we create a new entity AB after 2005 and exclude the separate financialaccounts of A and B that might still be reported for some time after the merger. Weidentify M&As and their timing using Bankscope and the websites of the mergingparties.
Third, in the US there are many distinct banks that have the same name but areactive in a different state. To solve this issue, we relate the value of total assetsof, say, bank i in the last year this bank appears in our sample with Bankscope’sidentification number for bank i. This also allows avoiding problems with ourprocedure concerning M&As described above. This data cleaning process yieldsan unbalanced panel with 89,778 observations, corresponding to 12,206 banksoperating in 148 countries between 1997 and 2010.
In order to compute the desired competition indices we first need to calculatethe variables π , p, q and tc and then to estimate a cost function. The latter has thegeneral form:
tcit = f (qit, wl,i t , wk,i t , wd,i t ), (6)
where wl,i t , wk,i t and wd,i t are factor prices for labor, capital and deposits respec-tively. To identify bank inputs and outputs for the cost function we use the inter-mediation approach, which assumes that deposits are inputs used in the productionprocess to produce bank outputs, and has been shown to be the preferred approachby a number of studies (e.g., Berger and Humphrey, 1997; Hughes and Mester,1993). In particular, we measure bank total costs (tc) by real total expenses, andbank output (q) by real total earning assets. This measure for bank output relatesto traditional banking activities, therefore our main indices reflect competitionin these activities. We construct real variables were appropriate, using the GDP
7Let us provide some examples to clarify this point. Assume that bank A1 is the parent bank witha consolidated (C) statement and banks A11, A12 and A13 are subsidiaries and unconsolidated (U)statement. If we include all banks in our sample we will have 3 duplicates. Hence, we need to subtracteither the percentage of the subsidiaries or to exclude the subsidiaries from the sample. The formersolution is not feasible because we do not have enough information for the percentage and the timeduration of the ownership of the subsidiaries. Thus, we resort to the later solution. Two other examplesfor the case of banks with foreign subsidiaries that we account for using the same strategy are (i) B1is a parent bank with a C statement, B11 is a subsidiary bank operating in the domestic market with aC or a U statement and B111 is a sub-subsidiary bank operating in the domestic market and (ii) B1 isa parent bank with C statement, B12 is a subsidiary bank operating abroad with a C or a U statementand B121 is a sub-subsidiary bank operating in the domestic market with a U statement.
274 Sofronis Clerides et al.
deflator (obtained from the World Bank).8 Real total earning assets include loans,securities, and other earning assets (such as investments and insurance assets).
We also specify a cost function with the same three inputs but two outputs,namely real total earning asset (q1) and off-balance sheet items (q2). We employthis alternative specification because many banks have a significant volume ofoff-balance sheet items and naturally use their inputs to produce these outputs.Our competition index with this approach reflects competition in the traditionalbanking activities after controlling for the fact that many banks also use theirinputs toward the production of off-balance sheet items.
The three input prices are proxied as follows: the price of labor wl by the ratioof personnel expenses to total assets;9 the price of physical capital wk by the ratioof capital expenditures to fixed assets; and the price of deposits wd as total interestexpenses over total customer deposits. We measure bank profits π using totalprofits before taxes. For the Lerner index and the adjusted-Lerner index we needthe aggregate output price p. This is calculated as the ratio of total income overtotal earning assets (Beck, Jonghe and Schepens, 2013).
As a final step, we clean our sample from outliers in the sense of unreasonablynegative values for total assets and total expenses. Additionally, for the three inputprices we drop 1% of our sample from each end of their distribution. This excludesunreasonably high or low input prices (Delis, 2012; Claessens and Laeven, 2004).Notably, the initial dataset before all the steps of the cleaning process describedabove includes 300,180 observations for 21,445 banks operating in 149 countriesbetween 1997 and 2010. Our final dataset consists of 89,778 observations for12,206 banks operating in 148 countries between 1997 and 2010. Most of theobservations are dropped due to some form of double-counting stemming fromBankscope’s consolidation system and M&As. This shows that the data-cleaningprocess is extremely important in generating sensible indices of bank competition.Table 1 reports the number of banks for all years, Table 2 provides detailedinformation on the construction of the variables, and Table 3 reports summarystatistics.
ECONOMETRIC METHODOLOGY FOR THE ESTIMATION OF MARKET POWER
An important concern for the empirical estimation of market power is the choiceof a proper functional form to obtain observation-specific estimates of marginalcost. The selection of a functional form is crucial because an inappropriate choicewill invalidate any inference. At a minimum, the production technology should beallowed to vary (i) across banks operating in the same industry and (ii) over time.
8As is standard in the macroeconomics literature, for Taiwan we use the GDP deflator of China andfor Netherlands Antilles we use the GDP deflator of Aruba.9We divide by total assets instead of the number of employees because Bankscope has limited in-formation on the number of employees. The related literature follows a similar approach (e.g., Delis,2012; Claessens and Laeven, 2004).
A New Data Set On Competition In National Banking Markets 275
Tabl
e1:
Num
ber
ofba
nks
inth
esa
mpl
e
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
All
coun
trie
s3,
895
4,05
16,
349
6,26
26,
184
6,04
75,
937
6,06
27,
362
7,48
67,
829
7,83
87,
531
6,94
5
Inco
me
grou
psL
I54
6674
8790
9811
112
013
814
016
119
120
718
0L
MI
223
250
282
278
273
283
299
337
391
397
427
440
446
391
OE
CD
3,15
43,
241
5,44
35,
281
5,19
45,
013
4,81
54,
848
5,76
55,
693
5,77
75,
652
5,52
45,
219
HI
110
112
107
116
116
116
114
128
141
142
140
133
113
94U
MI
354
382
443
500
511
537
598
629
927
1,11
41,
324
1,42
21,
241
1,06
1R
egio
nalg
roup
sE
AP
119
118
119
100
9810
311
213
717
218
319
922
022
220
8E
CA
8383
145
164
183
236
290
324
625
791
955
1,00
481
164
9L
AC
222
271
291
333
318
282
284
291
280
293
322
345
357
326
ME
NA
4944
4848
4246
4951
4957
7489
9789
SA76
8386
8993
100
111
114
121
117
125
127
135
123
SSA
8299
110
130
138
145
151
154
181
181
210
237
242
205
Not
es:L
Ire
fers
toth
elo
w-i
ncom
eec
onom
ies,
LM
Ire
fers
toth
elo
wer
-mid
dle-
inco
me
econ
omie
s,O
EC
Dre
fers
toth
eO
EC
Dm
embe
rco
untr
ies,
HI
refe
rsto
high
-inc
ome
econ
omie
sot
her
than
OE
CD
coun
trie
s,an
dU
MI
refe
rsto
uppe
r-m
iddl
e-in
com
eec
onom
ies.
EA
Pre
fers
toE
ast
Asi
anan
dPa
cifi
cco
untr
ies,
EC
Ato
the
Eur
opea
nan
dC
entr
alA
sian
coun
trie
s,L
AC
refe
rsto
Lat
inA
mer
ican
and
Car
ibbe
anco
untr
ies,
ME
NA
refe
rsto
Mid
dle
Eas
tern
and
Nor
thA
fric
anco
untr
ies,
SAre
fers
toSo
uthe
rnA
sian
coun
trie
san
dSS
Are
fers
toSu
b-Sa
hara
nA
fric
anco
untr
ies.
276 Sofronis Clerides et al.
Table 2: Data description
Variable Measure Notation
For aggregate output (1 output)
Earning assets Natural log of deflated total earning assets(measure of a bank’s output)
q
Price of output Total income/ total earning assets P
For disaggregate outputs (2 outputs)
Earning assets Natural log of deflated total earning assets(measure of bank’s output)
q1
Off-balance sheet items Natural log of deflated off-balance sheetitems (measure of bank’s output)
q2
Price of earning asset Total income/ total earning assets P1Price of off-balance sheet Total income/ off-balance sheet P2Expenses Natural log of deflated total interest
expenses and total noninterest expenses(measure of a bank’s total cost)
tc
Price of deposits Total interest expenses/ total customerdeposits
wd
Price of labor Personnel expenses/ total assets wlPrice of physical capital (Overheads-personnel expenses)/ fixed
assetswk
Profits A bank’s deflated total profits before tax π
NPI Natural log of
{1 if π ∈ �+
|π | if π ∈ �− NPI
Positive profits Natural log of
{π if π ∈ �+1 if π ∈ �− π
+
We rely on a semiparametric approach, which has some appealing features thatwe analyze below.
We start from the cost function implied by a standard log-linear productionfunction and impose the usual linear homogeneity restriction in inputs prices (thatis, we normalize total cost and the input prices by the price of deposits beforetaking logs). We end up with the following cost function:
ln tcitc = b1 + b2 ln wl,i tc + b3 ln wk,i tc + a1 ln qitc. (7)
We estimate this equation with a semi-parametric partial linear smooth coeffi-cient (PLSC) model, which uses the local polynomial fitting regression and theGaussian kernel function to obtain estimates of a1 for each bank i at time t incountry c. A thorough discussion of the PLSC model can be found in Fan andZhang (1999) and Mamuneas, Savvides and Stengos (2006). Delis (2012) and
A New Data Set On Competition In National Banking Markets 277
Table 3: Summary statistics
Variable Observations Mean Std. dev. Min. Max.
For aggregate output (1 output)
Earning assets 89,778 11.71 2.02 6.83 21.38Price of output 89,778 0.09 0.07 0.02 0.71
For disaggregate output (2 output)
Earning assets 89,778 11.71 2.02 6.83 21.38Off-balance sheet items 79,875 9.16 2.47 −1.58 21.26Price of earning asset 89,778 0.09 0.07 0.02 0.71Price of off-balance sheet 79875 39.30 811 0.00 110,787Expenses 89,778 12.25 2.10 5.29 22.42Price of deposits 89,778 0.06 0.09 0.00 1.03Price of labor 89,778 0.02 0.01 0.00 0.09Price of physical capital 89,778 1.70 3.71 0.13 56.96Profits 81,939 7.00 2.13 −2.24 16.76NPI 89,778 0.58 2.09 −2.27 16.69Positive profits 89,778 6.39 2.83 −2.24 16.76Lerner index 88,202 0.22 0.12 −0.20 0.95Adjusted-Lerner index 85,832 0.17 0.12 −0.20 0.95Profit elasticity 89,778 −0.42 0.08 −0.96 −0.32Notes: The table reports summary statistics (number of observations, mean, standard de-viation, minimum and maximum) for the variables used in the empirical analysis. Thebank level variables are defined in Table 2. The measures of competition are defined inEqs. (1), (2) and (4) for the Lerner index, the adjusted-Lerner index and the Profit elasticity,respectively, and are estimated using the methodology in Section 3.2.
Delis, Iosifidi and Tsionas (2012) use this model to estimate marginal cost; thesummary that follows is based on their discussion.
We can write the econometric form of the total cost equation as
Yi = E(Yi |Wi ) + ei = Xiβ1 + Viβ2(Zi ) + ei . (8)In this equation, β2 is a function of one or more variables of dimension k
added to the vector Z, which is an important element of the analysis and will bediscussed below. The presence of a linear part in Eq. (8) is in line with the idea ofa semiparametric model as opposed to a fully nonparametric one. The coefficientsof this part are estimated in a first step as averages of the polynomial fitting byusing an initial bandwidth chosen by cross-validation. In the second step we usethese average estimates to re-define the dependent variable as
Y ∗i ≡ Yi − Xi β̂i = Viβ2 (z) + e ∗i . (9)
278 Sofronis Clerides et al.
The coefficient β2(z) is a smooth but unknown function of z. We estimate β2(z)using a local least squares approach.10 Using the PLSC technique, we allow themodel to be linear in the regressors but their coefficients are allowed to change“smoothly” with the value of other variables z.
We can now re-write Eq. (7) in econometric form as
ln tcitc = b1 + a1(zitc) ln qitc + b2,i tc ln wl,i tc + b3,i tc ln wk,i tc + eitc, (10)where e is a stochastic disturbance and z is the smoothing variable required by thePLSC approach. The model is semiparametric in the sense that it treats only thecoefficient a1 as a function of z. From Eq. (10) we can obtain the marginal costfor each bank i at time t in country c as ∂tcitc/∂qitc = a1 (zitc) (tcitc/qitc).11
A critical issue in the estimation process is the choice of the variable(s) thatcomprise Z. The best candidates are variables that shift a1 and exhibit variationacross banks, countries and time. We need variables that shift marginal cost andthe natural candidates to use are the input prices; they are observation-specific byconstruction and they clearly affect marginal cost. Any cost function derived froma cost minimization problem will be a function of, among other thing, input prices.Numerous studies that have estimated cost functions using a translog specificationfind that input prices are indeed important determinants of marginal cost. Delis,Iosifidi and Tsionas (2012) propose using the linear combination of input pricesas the z variable and show that it works well in this context. We follow theirsuggestion and define the smoothing variable as zitc = ln wl,i tc + ln wk,i tc. As arobustness check we experimented with various other combinations (such as theproduct of the input prices or linear combinations with different weights) andfound that our results are not sensitive to the choice of z function.
Once we have our estimates of marginal cost we use them to calculate theLerner index from Eq. (1) and the adjusted-Lerner index from Eq. (2), as well asto estimate the profit elasticity from Eq. (4). A problem that arises with the latteris the fact that we have to take the logarithm of profits, and profits can be negative.Bos and Koetter (2011) propose a solution using the following transformation forbank profits (π )12:
π+ ={
π if π ∈ �+1 if π ∈ �− and NPI =
{1 if π ∈ �+
|π | if π ∈ �− . (11)
10Mamuneas, Savvides and Stengos (2006) discuss in detail how this function can take specificparametric formulations (such as linear) that can be tested against the general unknown specification.They also provide formulae for the local least squares criterion.11We use ln tcitc = b1 + a1(zitc) ln q1,i tc + a2 ln q2,i tc + b2,i tc ln wl,i tc + b3,i tc ln wk,i tc + eitc for themodel with the off-balance sheet items. Given that we focus on the market power stemming fromtraditional banking activities, marginal cost is still derived from ∂tc/∂q1,i tc = a1(zitc)(tcitc/q1,i tc).12There are two alternative approaches. The first is to truncate profits and exclude banks that incurlosses and the second is to rescale profits by adding the maximum loss observed in the sample plus asmall number to each observation on bank profits. The obvious drawback of the first approach is thatit introduces selection bias. The rescaling approach is a good approximation when the shift is smallrelative to the magnitude of the shifted variable but will bias the estimated coefficient if the shift islarge, which may be the case here. We experimented with both of these alternative approaches but theyproduced coefficients which were either insignificant or unreasonable.
A New Data Set On Competition In National Banking Markets 279
Table 4: Correlation between indices of weighted market power by marketshare
Lerner index Adjusted -Lerner index Profit elasticity
Lerner index 1.00Adjusted-Lerner index 0.86* 1.00Profit elasticity 0.33* 0.31* 1.00
Notes: The * mark denotes statistical significance at the 1% level.
This is an indicator that captures both profits and losses since π+ can be usedas a left-hand side variable in the regression equation and the negative profitindicator (NPI) as a right-hand side variable. Hence, the actual estimated equationequivalent to that of Eq. (4) is
ln π+i tc = a + β(zitc) ln mcitc + γ ln NPIitc + uitc. (12)Estimation of equation (10) and subsequently of equation (12) via the PLSC
technique presents some considerable interrelated advantages. Most importantly,the semiparametric nature of the method implies that no assumption regarding thefunctional form of the underlying cost function is made globally. This is importantas it usually quite difficult for the researcher to be certain about the validity of thechosen functional form. The PLSC method allows for considerable flexibility inthe estimation of the cost function because of variation in the structural variablesemployed. In their survey paper, Reiss and Wolak (2007) are very skeptical aboutusing a specific functional to estimate cost functions without a prior analysis ofthe data, since an “incorrect” cost function can bias the estimation and inferenceof marginal cost in an unknown direction and magnitude. The flexibility of thesemiparametric technique takes into account the heterogeneity in the productiontechnology across banks and countries and hence allows the use of large inter-national samples of banks employing different production technologies. Finally,Mammen, Rothe and Schienle (2012) establish theoretically that the two-step esti-mation procedure reduces measurement errors stemming from unobserved countryfactors. Delis, Iosifidi and Tsionas (2012) and Wheelock and Wilson (2012) pro-vide empirical support to these arguments by showing that estimation of marginalcost using semiparametric and nonparametric methods produces significantly bet-ter results than parametric techniques and commonly used functional forms likethe translog.
IV. ESTIMATES OF BANK COMPETITION
In Table 4 we report the pairwise correlation coefficients between the Lerner,the adjusted-Lerner and the profit-elasticity measures (the calculation weighs ob-servations by bank size, measured as total earning assets). The Lerner and theadjusted-Lerner indices are highly correlated, while the correlation coefficients
280 Sofronis Clerides et al.
Tabl
e5:
Ave
rage
esti
mat
esof
mar
ket
pow
er(w
eigh
ted
bym
arke
tsh
ares
)us
ing
the
Ler
ner
inde
x
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Afg
hani
stan
0.38
10.
237
0.08
40.
362
0.29
90.
147
0.25
2A
lban
ia0.
120
0.32
10.
210
0.20
10.
184
0.21
50.
293
0.27
40.
317
0.30
30.
318
0.35
90.
223
Alg
eria
0.15
30.
165
0.06
50.
153
0.22
90.
387
0.24
40.
459
0.59
00.
648
0.53
30.
624
0.52
80.
513
0.37
8A
ndor
ra0.
255
0.29
60.
354
0.35
90.
305
0.37
30.
459
0.50
50.
505
0.50
70.
439
0.28
10.
386
Ang
ola
0.27
50.
313
0.28
10.
397
0.49
80.
427
0.41
20.
267
0.45
90.
492
0.42
70.
467
0.39
3A
ntig
uaan
dB
arbu
da0.
051
0.09
00.
123
0.13
30.
266
0.33
40.
344
0.19
2A
rgen
tina
0.21
70.
170
0.18
90.
218
0.13
60.
121
0.01
90.
167
0.25
70.
285
0.24
50.
209
0.32
50.
318
0.20
5A
rmen
ia0.
182
0.23
50.
215
0.18
80.
280
0.34
80.
375
0.38
90.
374
0.36
40.
354
0.32
90.
226
0.28
40.
296
Aus
tral
ia0.
253
0.24
80.
211
0.28
5−0
.085
0.22
50.
250
0.23
30.
218
0.16
50.
250
0.25
10.
209
Aus
tria
0.14
70.
122
0.13
20.
146
0.14
50.
154
0.18
90.
185
0.18
20.
174
0.16
60.
151
0.20
60.
260
0.16
8A
zerb
aija
n0.
533
0.37
00.
377
0.53
50.
436
0.38
20.
375
0.43
50.
441
0.38
80.
388
0.41
10.
380
0.27
50.
409
Bah
amas
,The
0.15
90.
173
0.21
00.
272
0.29
40.
214
0.32
10.
356
0.39
30.
388
0.42
10.
333
0.39
10.
390
0.30
8B
ahra
in0.
205
0.17
70.
175
0.16
10.
177
0.23
90.
223
0.28
40.
265
0.20
10.
189
0.23
20.
211
Ban
glad
esh
0.03
0−0
.033
0.07
00.
114
0.13
40.
142
0.13
80.
164
0.21
40.
189
0.21
10.
256
0.27
50.
339
0.16
0B
elar
us0.
092
0.20
90.
112
0.17
80.
120
0.18
30.
168
0.15
00.
182
0.21
10.
186
0.17
40.
241
0.24
60.
175
Bel
gium
0.10
30.
138
0.14
50.
162
0.16
60.
150
0.16
10.
158
0.12
20.
143
0.07
1−0
.016
0.07
90.
155
0.12
4B
erm
uda
0.09
70.
114
0.11
80.
156
0.12
00.
194
0.21
00.
131
0.26
90.
266
0.27
40.
128
0.21
10.
229
0.18
0B
oliv
ia0.
138
0.18
60.
206
0.17
90.
194
0.23
90.
203
0.14
50.
177
0.22
10.
238
0.30
00.
261
0.27
40.
211
(Con
tinu
ed)
A New Data Set On Competition In National Banking Markets 281
Tabl
e5:
(Con
tinue
d)
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Bos
nia
and
Her
zego
vina
0.21
50.
237
0.23
30.
256
0.18
30.
230
0.25
10.
229
Bot
swan
a0.
246
0.30
70.
248
0.32
40.
326
0.33
80.
353
0.33
70.
357
0.32
80.
269
0.29
40.
309
0.33
60.
312
Bra
zil
0.13
70.
160
0.15
50.
132
0.14
40.
161
0.22
80.
219
0.24
40.
274
0.27
80.
195
0.29
40.
259
0.20
6B
ulga
ria
0.30
90.
283
0.33
90.
360
0.37
20.
378
0.38
50.
338
0.32
30.
343
0.34
3B
urki
naFa
so0.
277
0.38
60.
337
0.27
00.
236
0.35
00.
348
0.31
70.
342
0.30
60.
308
0.24
60.
266
0.34
60.
310
Cam
bodi
a0.
478
0.46
90.
337
0.38
60.
436
0.43
60.
450
0.48
40.
517
0.37
90.
363
0.43
0C
amer
oon
0.58
00.
499
0.45
10.
420
0.38
50.
479
0.43
20.
426
0.43
50.
390
0.31
40.
345
0.43
0C
anad
a0.
135
0.10
80.
179
0.16
80.
166
0.19
40.
202
0.22
90.
187
0.21
50.
190
0.15
20.
258
0.30
40.
192
Cay
man
Isla
nds
0.17
60.
176
Chi
le0.
161
0.16
00.
204
0.20
60.
238
0.28
30.
194
0.15
00.
160
0.22
80.
308
0.21
70.
411
0.38
30.
236
Chi
na0.
405
0.38
30.
254
0.27
50.
259
0.34
60.
379
0.39
90.
385
0.39
00.
429
0.40
70.
417
0.44
90.
370
Col
ombi
a0.
146
0.08
10.
030
0.08
50.
146
0.15
20.
244
0.28
30.
322
0.27
90.
312
0.31
80.
341
0.37
90.
223
Cos
taR
ica
0.07
30.
084
0.07
60.
182
0.18
50.
183
0.23
50.
220
0.21
40.
226
0.21
30.
175
0.14
50.
222
0.17
4C
ote
d’Iv
oire
0.37
90.
386
0.32
20.
300
0.26
30.
241
0.23
00.
273
0.26
60.
276
0.30
30.
286
0.27
70.
263
0.29
0C
roat
ia0.
209
0.16
70.
169
0.22
60.
202
0.21
50.
251
0.27
10.
282
0.25
70.
268
0.25
30.
274
0.30
10.
239
Cub
a0.
824
0.76
10.
731
0.68
90.
569
0.70
30.
785
0.78
70.
701
0.61
10.
470
0.55
70.
536
0.65
10.
670
Cyp
rus
0.15
50.
151
0.28
40.
107
0.11
10.
143
0.17
60.
208
0.18
80.
253
0.28
40.
202
0.23
30.
249
0.19
6C
zech
Rep
ublic
0.18
00.
158
0.16
70.
166
0.16
20.
239
0.26
70.
298
0.34
30.
328
0.32
80.
277
0.44
00.
444
0.27
1D
enm
ark
0.16
50.
175
0.14
10.
147
0.25
10.
265
0.39
00.
180
0.18
40.
161
0.13
50.
104
0.21
80.
213
0.19
5D
omin
ican
Rep
ublic
0.18
90.
180
0.16
60.
190
0.19
00.
198
0.17
50.
115
0.18
40.
202
0.22
00.
226
0.22
00.
266
0.19
4E
cuad
or0.
050
−0.1
240.
297
0.12
70.
113
0.18
50.
197
0.22
70.
268
0.27
60.
268
0.24
10.
234
0.26
50.
187
Egy
pt,A
rab
Rep
.0.
065
0.06
50.
314
0.23
80.
171
ElS
alva
dor
0.11
90.
169
0.16
60.
178
0.24
40.
288
0.28
20.
304
0.32
60.
365
0.35
90.
365
0.38
00.
447
0.28
5
(Con
tinu
ed)
282 Sofronis Clerides et al.
Tabl
e5:
(Con
tinue
d)
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Est
onia
0.26
20.
029
0.01
40.
052
0.20
40.
271
0.32
80.
347
0.34
10.
364
0.32
30.
313
0.28
60.
373
0.25
1E
thio
pia
0.27
00.
257
0.34
40.
285
0.40
60.
331
0.57
40.
573
0.57
40.
612
0.53
80.
616
0.65
00.
595
0.47
3Fi
nlan
d0.
055
0.00
00.
338
0.35
40.
266
0.20
70.
174
0.18
80.
194
0.11
80.
267
0.28
00.
203
Fran
ce0.
100
0.10
70.
128
0.11
20.
132
0.15
20.
168
0.20
50.
220
0.22
10.
197
0.17
20.
229
0.24
80.
171
Gam
bia,
The
0.49
50.
569
0.55
10.
552
0.52
90.
530
0.43
70.
401
0.41
70.
272
0.33
00.
253
0.31
70.
435
Geo
rgia
0.33
50.
362
0.31
80.
339
0.34
10.
341
0.31
60.
351
0.33
30.
282
0.26
20.
230
0.23
50.
311
Ger
man
y0.
171
0.15
10.
164
0.13
90.
132
0.15
70.
175
0.18
90.
185
0.20
40.
166
0.15
30.
193
0.23
40.
172
Gha
na0.
160
0.44
20.
419
0.13
70.
412
0.41
40.
435
0.48
30.
442
0.29
30.
274
0.24
10.
324
0.34
4G
reec
e0.
169
0.20
10.
404
0.21
50.
000
0.04
40.
112
0.13
60.
183
0.21
60.
173
0.10
40.
184
0.15
10.
164
Gua
tem
ala
0.08
80.
124
0.12
60.
186
0.22
80.
246
0.25
10.
242
0.25
30.
248
0.25
70.
204
Hai
ti0.
123
0.11
90.
116
0.17
20.
156
0.10
80.
224
0.09
90.
145
0.17
10.
178
0.19
70.
183
0.18
30.
155
Hon
dura
s0.
338
0.26
20.
186
0.12
90.
165
0.19
70.
256
0.18
00.
205
0.24
00.
250
0.27
20.
233
0.20
80.
223
Hon
gK
ong
SAR
,Chi
na0.
238
0.18
70.
243
0.27
30.
165
0.35
10.
389
0.42
90.
300
0.27
60.
260
0.17
60.
299
0.34
30.
281
Hun
gary
0.15
30.
144
0.08
70.
122
0.16
30.
181
0.22
60.
219
0.24
50.
243
0.25
00.
192
0.22
30.
313
0.19
7Ic
elan
d0.
167
0.17
50.
200
0.06
80.
145
0.21
00.
231
0.26
90.
336
0.36
30.
331
0.42
60.
337
0.48
90.
268
Indi
a0.
121
0.14
60.
120
0.15
80.
158
0.20
90.
244
0.30
30.
282
0.26
60.
241
0.18
60.
194
0.21
10.
203
Indo
nesi
a0.
134
0.04
30.
030
0.10
70.
129
0.16
00.
228
0.32
50.
248
0.25
60.
295
0.31
10.
315
0.35
60.
210
(Con
tinu
ed)
A New Data Set On Competition In National Banking Markets 283
Tabl
e5:
(Con
tinue
d)
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Iraq
0.46
30.
316
0.38
9Ir
elan
d0.
177
0.17
50.
253
0.21
50.
148
0.13
50.
228
0.21
70.
144
0.13
20.
146
0.14
60.
196
0.20
50.
180
Isra
el0.
153
0.06
40.
092
0.12
40.
084
0.10
20.
116
0.17
70.
150
0.19
80.
206
0.14
10.
197
0.10
80.
136
Ital
y0.
157
0.20
00.
143
0.20
30.
183
0.21
80.
218
0.17
90.
241
0.25
80.
240
0.19
80.
238
0.23
60.
208
Jam
aica
0.12
80.
158
0.20
10.
289
0.21
60.
271
0.23
30.
267
0.27
80.
271
0.30
10.
293
0.33
40.
249
Japa
n0.
246
0.24
60.
259
0.25
90.
250
0.23
00.
266
0.26
10.
282
0.28
50.
286
0.24
20.
191
0.23
30.
253
Jord
an0.
152
0.18
20.
173
0.14
70.
239
0.23
70.
325
0.36
20.
490
0.40
00.
363
0.34
90.
370
0.41
90.
301
Kaz
akhs
tan
0.24
50.
310
0.30
60.
246
0.34
70.
366
0.35
90.
393
0.35
60.
329
0.34
00.
243
0.23
00.
077
0.29
6K
enya
0.15
30.
262
0.27
00.
311
0.32
10.
318
0.38
00.
371
0.36
10.
391
0.36
90.
344
0.32
60.
384
0.32
6K
orea
,Rep
.0.
071
0.11
50.
219
0.17
90.
266
0.31
10.
316
0.33
10.
310
0.29
10.
271
0.19
10.
221
0.25
80.
239
Kuw
ait
0.09
20.
239
0.28
70.
299
0.36
70.
444
0.51
70.
555
0.56
50.
470
0.39
30.
384
Kyr
gyz
Rep
ublic
0.17
60.
323
0.11
60.
371
0.37
50.
460
0.36
50.
397
0.45
40.
319
0.35
90.
327
0.33
7L
aoPD
R0.
232
0.01
90.
000
0.25
20.
478
0.55
50.
669
0.29
20.
285
0.35
30.
314
Lat
via
0.28
00.
214
0.25
70.
280
0.27
10.
303
0.33
70.
356
0.36
20.
327
0.30
50.
241
0.24
70.
227
0.28
6L
eban
on0.
168
0.14
90.
141
0.14
40.
127
0.14
10.
163
0.14
20.
151
0.14
90.
144
0.17
90.
190
0.22
60.
158
Lib
ya0.
535
0.57
60.
535
0.05
00.
401
0.52
30.
597
0.69
10.
248
0.46
2L
ithua
nia
0.26
90.
154
0.24
20.
151
0.18
30.
217
0.18
40.
252
0.28
90.
306
0.31
10.
245
0.17
80.
205
0.22
8L
uxem
bour
g0.
103
0.09
50.
115
0.13
40.
118
0.13
40.
151
0.18
90.
207
0.19
80.
184
0.13
70.
242
0.28
50.
164
Mac
aoSA
R,C
hina
0.12
70.
132
0.16
60.
184
0.19
00.
290
0.35
40.
396
0.36
60.
296
0.28
00.
325
0.39
50.
423
0.28
0M
aced
onia
,FY
R0.
498
0.35
30.
346
0.29
70.
303
0.26
50.
317
0.31
70.
359
0.35
90.
365
0.31
40.
261
0.24
20.
328
Mad
agas
car
0.55
50.
565
0.50
70.
377
0.32
10.
356
0.45
10.
458
0.47
10.
492
0.44
10.
337
0.27
10.
260
0.41
9
(Con
tinu
ed)
284 Sofronis Clerides et al.
Tabl
e5:
(Con
tinue
d) 1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Mal
awi
0.42
00.
460
0.44
30.
390
0.26
30.
357
0.36
00.
371
0.39
00.
491
0.52
50.
438
0.42
20.
360
0.40
6M
alay
sia
0.27
70.
246
0.27
10.
362
0.34
40.
355
0.35
10.
352
0.35
50.
353
0.36
00.
366
0.36
20.
409
0.34
0M
ali
0.25
20.
266
0.29
80.
253
0.32
40.
307
0.33
50.
304
0.31
10.
367
0.32
50.
304
0.32
10.
286
0.30
4M
alta
0.21
40.
217
0.24
90.
226
0.22
50.
239
0.27
30.
307
0.34
50.
339
0.33
60.
292
0.31
00.
362
0.28
1M
auri
tani
a0.
574
0.50
50.
313
0.34
00.
186
0.46
30.
466
0.27
50.
277
0.43
10.
333
0.37
8M
auri
tius
0.17
40.
198
0.18
00.
183
0.20
40.
326
0.27
90.
324
0.33
00.
279
0.26
20.
284
0.30
40.
399
0.26
6M
exic
o0.
011
0.00
20.
063
0.01
70.
280
−0.0
25−0
.023
0.04
6M
oldo
va0.
353
0.38
80.
401
0.41
30.
380
0.38
40.
408
0.35
10.
289
0.34
10.
340
0.28
40.
222
0.30
90.
347
Mon
golia
0.31
60.
220
0.27
20.
255
0.22
60.
263
0.21
40.
167
0.20
00.
219
0.20
70.
190
0.22
9M
onte
negr
o0.
000
0.27
50.
238
0.16
10.
204
0.25
60.
205
0.19
70.
231
0.19
6M
oroc
co0.
217
0.23
70.
217
0.29
40.
310
0.32
90.
329
0.37
50.
305
0.33
70.
336
0.35
90.
354
0.36
40.
312
Moz
ambi
que
0.26
30.
236
0.31
90.
259
0.27
90.
272
0.19
40.
238
0.25
90.
340
0.36
80.
375
0.38
50.
356
0.29
6N
amib
ia0.
183
0.02
30.
490
0.42
50.
270
0.25
50.
256
0.24
90.
241
0.27
00.
266
Nep
al0.
355
0.24
70.
319
0.36
20.
357
0.34
80.
231
0.25
80.
273
0.31
10.
292
0.33
30.
326
0.28
30.
307
Net
herl
ands
Ant
illes
0.11
40.
142
0.21
00.
130
0.12
90.
145
Net
herl
ands
0.12
60.
127
0.14
30.
204
0.21
30.
109
0.09
40.
160
0.15
40.
135
0.17
70.
183
0.14
90.
256
0.15
9N
ewZ
eala
nd0.
121
0.08
50.
230
0.20
70.
226
0.27
20.
249
0.20
00.
211
0.19
60.
173
0.20
40.
198
Nic
arag
ua0.
201
0.22
00.
237
0.29
50.
327
0.34
20.
370
0.37
90.
367
0.30
4N
iger
0.26
10.
399
0.06
60.
206
0.14
50.
206
0.14
30.
233
0.30
40.
322
0.26
50.
352
0.33
60.
328
0.25
5N
iger
ia0.
228
0.29
00.
304
0.27
60.
296
0.26
80.
275
0.26
40.
313
0.31
70.
309
0.32
50.
195
0.22
40.
277
Nor
way
0.16
90.
061
0.14
60.
157
0.15
50.
128
0.15
90.
219
0.26
50.
230
0.17
60.
146
0.26
60.
263
0.18
1
(Con
tinu
ed)
A New Data Set On Competition In National Banking Markets 285
Tabl
e5:
(Con
tinue
d)
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Om
an0.
309
0.27
40.
283
0.25
80.
301
0.39
20.
398
0.42
80.
423
0.42
00.
378
0.42
90.
464
0.36
6Pa
kist
an0.
040
0.02
3−0
.014
0.04
50.
119
0.18
50.
259
0.27
00.
395
0.36
80.
321
0.27
70.
288
0.27
60.
204
Pana
ma
0.19
60.
134
0.31
70.
259
0.25
50.
300
0.36
30.
322
0.30
60.
275
0.32
00.
311
0.30
50.
313
0.28
4Pa
pua
New
Gui
nea
0.25
00.
259
0.08
80.
401
0.64
10.
520
0.50
40.
611
0.61
40.
530
0.49
00.
446
Para
guay
0.27
80.
181
0.10
40.
041
0.09
20.
015
−0.1
140.
052
0.14
00.
131
0.13
30.
208
0.16
80.
216
0.11
8Pe
ru0.
219
0.20
30.
184
0.16
00.
174
0.25
90.
295
0.31
50.
357
0.36
40.
351
0.38
70.
438
0.39
00.
293
Phili
ppin
es0.
264
0.27
20.
177
0.00
10.
065
0.21
40.
298
0.23
70.
239
0.24
80.
239
0.19
30.
278
0.32
50.
218
Pola
nd0.
170
0.17
50.
162
0.16
50.
166
0.16
90.
137
0.17
40.
190
0.23
90.
246
0.21
50.
232
0.24
10.
192
Port
ugal
0.11
90.
131
0.10
40.
168
0.30
50.
202
0.23
00.
294
0.19
80.
162
0.13
80.
082
0.08
70.
065
0.16
3Q
atar
0.24
20.
318
0.47
10.
522
0.51
40.
551
0.43
50.
398
0.37
00.
375
0.42
0R
oman
ia0.
233
0.21
50.
214
0.19
90.
247
0.19
00.
202
0.26
20.
236
0.22
10.
209
0.22
40.
234
0.27
80.
226
Rus
sian
Fede
ratio
n0.
207
0.06
10.
410
0.37
70.
454
0.34
40.
310
0.33
90.
307
0.29
70.
282
0.27
20.
239
0.20
20.
293
Rw
anda
0.18
70.
205
0.25
70.
109
0.00
40.
320
0.35
20.
343
0.24
90.
343
0.23
7Sa
nM
arin
o0.
185
0.26
20.
400
0.39
70.
328
0.33
50.
435
0.50
60.
504
0.46
00.
382
0.19
50.
366
Saud
iAra
bia
0.26
30.
261
0.25
40.
247
0.31
10.
405
0.49
00.
501
0.49
00.
488
0.34
00.
225
0.36
20.
288
0.35
2Se
nega
l0.
356
0.42
80.
351
0.34
40.
364
0.35
20.
345
0.34
20.
330
0.34
00.
307
0.32
70.
297
0.28
10.
340
Serb
ia0.
374
0.47
20.
362
0.33
60.
217
0.24
90.
228
0.23
40.
176
0.29
4Se
yche
lles
0.19
80.
508
0.55
90.
567
0.59
50.
594
0.37
70.
528
0.49
1Si
erra
Leo
ne0.
190
0.40
00.
646
0.53
50.
481
0.47
40.
519
0.47
20.
386
0.28
70.
188
0.24
70.
328
0.39
6Si
ngap
ore
0.24
80.
232
0.36
20.
353
0.29
70.
230
0.41
40.
361
0.30
90.
331
0.37
60.
489
0.43
80.
342
Slov
akR
epub
lic0.
092
0.03
20.
029
0.14
20.
158
0.18
30.
216
0.24
60.
267
0.29
10.
284
0.30
40.
322
0.39
00.
211
Slov
enia
0.21
40.
213
0.22
40.
238
0.18
80.
210
0.21
40.
252
0.26
60.
252
0.24
90.
184
0.23
70.
269
0.22
9So
uth
Afr
ica
0.10
50.
163
0.16
70.
179
0.20
40.
300
0.21
10.
177
0.15
50.
233
0.22
20.
199
0.21
70.
229
0.19
7Sp
ain
0.13
00.
161
0.22
80.
181
0.17
90.
196
0.23
80.
275
0.24
20.
246
0.22
90.
207
0.29
20.
305
0.22
2Sr
iLan
ka0.
149
0.17
70.
114
0.10
20.
094
0.15
00.
224
0.21
00.
210
0.19
60.
171
0.14
60.
171
0.23
20.
168
(Con
tinu
ed)
286 Sofronis Clerides et al.
Tabl
e5:
(Con
tinue
d) 199
719
9819
9920
0020
0120
0220
0320
0420
0520
0620
0720
0820
0920
10M
ean
Suda
n0.
395
0.26
60.
246
0.25
80.
145
0.31
70.
180
0.29
10.
257
0.27
70.
171
0.22
30.
193
0.21
40.
245
Swed
en0.
186
0.16
80.
156
0.18
20.
183
0.16
90.
206
0.27
70.
234
0.22
40.
178
0.16
00.
223
0.24
40.
199
Switz
erla
nd0.
168
0.13
20.
126
0.15
60.
124
0.16
50.
179
0.18
00.
122
0.12
50.
039
0.03
60.
129
0.17
90.
133
Syri
anA
rab
Rep
ublic
0.00
00.
064
0.30
90.
568
0.56
90.
567
0.34
6Ta
iwan
0.15
90.
165
0.22
70.
349
0.28
30.
307
0.27
80.
248
0.21
80.
294
0.34
20.
261
Tanz
ania
0.47
10.
439
0.39
00.
423
0.39
50.
392
0.35
70.
343
0.40
1T
haila
nd0.
171
0.01
10.
045
0.10
60.
148
0.23
30.
290
0.37
50.
375
0.28
80.
289
0.33
40.
369
0.38
90.
245
Togo
0.11
10.
191
0.21
60.
446
0.22
50.
129
0.27
60.
315
0.25
90.
307
0.25
90.
282
0.24
40.
344
0.25
8T
rini
dad
and
Toba
go0.
195
0.19
30.
231
0.26
60.
284
0.30
20.
360
0.34
70.
309
0.32
10.
338
0.34
50.
313
0.44
20.
303
Tun
isia
0.56
20.
557
0.45
80.
302
0.29
20.
267
0.18
90.
208
0.22
10.
285
0.29
50.
323
0.33
10.
346
0.33
1T
urke
y0.
022
0.03
40.
143
0.04
6−0
.017
0.11
20.
190
0.24
00.
286
0.22
60.
227
0.20
90.
335
0.32
00.
169
Uga
nda
0.40
10.
360
0.36
80.
341
0.36
7U
krai
ne0.
229
0.26
90.
316
0.21
10.
229
0.18
20.
245
0.23
30.
221
0.24
30.
220
0.31
40.
250
0.21
40.
241
Uni
ted
Ara
bE
mir
ates
0.30
70.
298
0.31
40.
295
0.34
00.
462
0.50
70.
516
0.51
60.
359
0.34
60.
372
0.45
30.
468
0.39
7U
nite
dK
ingd
om0.
182
0.18
40.
177
0.24
30.
110
0.16
90.
282
0.29
20.
254
0.24
10.
236
0.10
30.
294
0.30
80.
220
Uni
ted
Stat
es0.
239
0.22
90.
252
0.22
40.
266
0.33
20.
355
0.32
10.
304
0.26
80.
227
0.23
90.
344
0.35
20.
282
Uru
guay
0.07
20.
076
0.07
30.
097
0.03
70.
248
0.01
30.
241
0.09
00.
190
0.26
90.
363
0.18
10.
250
0.15
7U
zbek
ista
n0.
378
0.30
70.
301
0.37
10.
364
0.32
10.
223
0.18
10.
239
0.27
50.
283
0.22
90.
212
0.24
80.
281
Ven
ezue
la,R
B0.
291
0.28
30.
217
0.18
20.
226
0.30
10.
327
0.34
30.
276
0.29
30.
281
0.26
30.
265
0.30
60.
275
Vie
tnam
0.37
90.
346
0.31
40.
345
0.26
40.
292
0.27
30.
349
0.33
60.
282
0.27
70.
208
0.19
80.
205
0.29
1Y
emen
,Rep
.0.
055
0.22
60.
231
0.20
00.
272
0.24
20.
204
Zam
bia
0.04
70.
172
0.11
70.
194
0.23
40.
224
0.10
10.
233
0.29
60.
340
0.34
00.
299
0.33
70.
288
0.23
0Z
imba
bwe
0.29
70.
299
0.29
8M
ean
0.21
20.
207
0.22
60.
230
0.22
80.
255
0.28
00.
298
0.29
90.
301
0.29
40.
277
0.29
00.
305
0.26
6
Not
es:T
heta
ble
repo
rts
aver
age
estim
ates
ofm
arke
tpow
er(w
eigh
ted
bym
arke
tsha
res)
byco
untr
yan
dye
ar.A
vera
ges
are
obta
ined
from
the
bank
-lev
eles
timat
esof
mar
ketp
ower
usin
gth
eL
erne
rin
dex,
asth
isis
defi
ned
inE
q.(1
).H
ighe
rva
lues
refl
ecth
ighe
rm
arke
tpow
er(l
ower
com
petit
ion)
.
A New Data Set On Competition In National Banking Markets 287
Tabl
e6:
Ave
rage
esti
mat
esof
mar
ket
pow
er(w
eigh
ted
bym
arke
tsh
ares
)us
ing
the
adju
sted
-Ler
ner
inde
x
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Afg
hani
stan
0.41
20.
129
0.01
70.
400
0.25
90.
027
0.20
7A
lban
ia0.
515
0.25
80.
133
0.12
50.
187
0.15
20.
295
0.24
70.
293
0.24
80.
227
0.27
30.
211
Alg
eria
0.00
90.
028
0.04
20.
041
0.05
80.
230
0.14
50.
098
0.20
70.
460
0.42
00.
512
0.49
60.
468
0.23
0A
ndor
ra0.
220
0.25
40.
315
0.32
40.
299
0.36
60.
443
0.47
90.
438
0.44
90.
374
0.23
60.
350
Ang
ola
0.24
80.
296
0.06
50.
046
0.27
00.
133
0.36
40.
194
0.37
50.
435
0.36
10.
405
0.26
6A
ntig
uaan
dB
arbu
da0.
051
0.09
00.
123
0.13
80.
249
0.27
60.
301
0.17
5A
rgen
tina
0.18
30.
098
0.09
70.
115
0.04
20.
038
−0.0
130.
104
0.19
40.
197
0.17
70.
155
0.25
60.
288
0.13
8A
rmen
ia0.
082
0.18
00.
135
0.10
50.
144
0.34
10.
320
0.39
10.
373
0.34
50.
322
0.30
10.
140
0.25
30.
245
Aus
tral
ia0.
281
0.23
00.
137
0.16
0−0
.125
0.23
80.
241
0.22
10.
208
0.13
90.
167
0.20
40.
175
Aus
tria
0.09
70.
055
0.09
20.
089
0.07
70.
081
0.11
60.
135
0.13
40.
184
0.14
50.
074
0.09
90.
148
0.10
9A
zerb
aija
n0.
487
0.31
10.
187
0.42
40.
325
0.25
80.
259
0.30
30.
300
0.30
30.
325
0.27
30.
174
0.06
20.
285
Bah
amas
,The
0.15
20.
172
0.18
90.
249
0.27
30.
175
0.28
40.
328
0.38
30.
372
0.39
90.
289
0.35
70.
331
0.28
2B
ahra
in0.
167
0.08
80.
112
0.12
00.
129
0.12
60.
214
0.37
20.
278
0.20
20.
148
0.11
80.
173
Ban
glad
esh
0.04
0−0
.023
0.03
50.
062
0.09
20.
092
0.09
00.
115
0.09
70.
163
0.17
40.
216
0.24
00.
293
0.12
0B
elar
us−0
.124
0.01
5−0
.034
0.07
7−0
.054
0.13
10.
115
0.11
60.
158
0.15
20.
164
0.12
40.
177
0.17
80.
085
Bel
gium
0.08
30.
058
0.11
40.
118
0.12
40.
111
0.14
90.
142
0.14
00.
161
0.07
2−0
.068
0.02
80.
127
0.09
7B
erm
uda
0.09
50.
105
0.12
00.
153
0.11
10.
170
0.20
40.
307
0.32
10.
281
0.27
30.
237
0.21
00.
240
0.20
2B
oliv
ia0.
110
0.07
60.
085
−0.0
410.
027
0.04
60.
097
0.07
90.
124
0.19
60.
220
0.22
10.
232
0.21
50.
121
Bos
nia
and
Her
zego
vina
0.09
10.
128
0.14
00.
144
0.07
40.
071
0.09
60.
106
Bot
swan
a0.
211
0.28
90.
284
0.26
90.
318
0.32
60.
329
0.31
40.
339
0.30
80.
239
0.27
40.
266
0.31
00.
291
Bra
zil
0.08
70.
063
0.06
50.
107
0.10
30.
104
0.16
60.
161
0.18
70.
168
0.23
10.
089
0.20
80.
200
0.13
9B
ulga
ria
0.29
50.
239
0.34
60.
336
0.30
90.
332
0.34
70.
303
0.18
90.
182
0.28
8B
urki
naFa
so0.
303
0.38
60.
298
0.19
10.
196
0.32
90.
272
0.23
20.
160
0.20
80.
165
0.12
70.
223
0.30
20.
242
Cam
bodi
a0.
445
0.00
00.
246
0.27
40.
285
0.27
30.
372
0.46
20.
503
0.30
90.
302
0.31
6C
amer
oon
0.28
60.
308
0.27
90.
309
0.38
10.
341
0.29
90.
320
0.32
30.
289
0.26
40.
304
0.30
9
(Con
tinu
ed)
288 Sofronis Clerides et al.Ta
ble
6:(C
ontin
ued)
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
Mea
n
Can
ada
0.12
00.
072
0.15
10.
125
0.12
20.
090
0.16
90.
216
0.17
60.
206
0.18
40.
127
0.19
30.
272
0.15
9C
aym
anIs
land
s0.
138
0.13
8C
hile
0.12
40.
105
0.06
80.
148
0.18
30.
249
0.17
10.
155
0.18
70.
238
0.27
70.
133
0.28
00.
289
0.18
6C
hina
0.39
20.
352
0.18
50.
201
0.21
40.
269
0.29
40.
329
0.32
70.
328
0.37
80.
331
0.37
50.
416
0.31
4C
olom
bia
0.07
70.
023
0.00
0−0
.016
0.06
30.
092
0.18
00.
241
0.25
50.
217
0.22
70.
208
0.22
40.
284
0.14
8C
osta
Ric
a0.
066
0.07
90.
071
0.16
40.
167
0.14
20.
210
0.20
30.
194
0.19
00.
188
0.14
80.
095
0.16
10.
148
Cot
ed’
Ivoi
re0.
235
0.19
30.
198
0.18
40.
137
0.08
80.
091
0.15
40.
191
0.24
80.
273
0.25
50.
272
0.23
10.
196
Cro
atia
0.13
50.
093
0.10
50.
198
0.18
90.
215
0.23
20.
236
0.25
60.
231
0.23
40.
220
0.18
70.
204
0.19
5C
uba
0.82
20.
761
0.72
60.
684
0.55
40.
681
0.77
80.
770
0.68
30.
607
0.45
40.
559
0.51
60.
611
0.65
8C
ypru
s0.
126
0.11
70.
273
0.16
50.
068
−0.0
220.
035
0.06
20.
122
0.19
60.
258
0.17
20.
138
0.13
00.
131
Cze
chR
epub
lic0.
055
−0.0
020.
016
0.10
30.
146
0.24
90.
295
0.31
60.
330
0.32
10.
307
0.21
70.
355
0.36
00.
219
Den
mar
k0.
147
0.14
60.
120
0.11
80.
186
0.20
60.
339
0.19
40.
205
0.19
10.
147
0.02
30.
048
0.10
80.
156
Dom
inic
anR
epub
lic0.
162
0.13
80.
139
0.14
20.
146
0.16
90.
122
0.09
30.
155
0.18
90.
213
0.20
70.
184
0.23
30.
164
Ecu
ador
0.01
9−0
.052
0.03
90.
081
0.12
50.
143
0.13
80.
160
0.19
90.
216
0.20
10.
211
0.17
20.
188
0.13
1E
gypt
,Ara
bR
ep.
0.23
70.
046
0.28
50.
222
0.19
7E
lSal
vado
r0.
129
0.13
70.
086
0.09
10.
138
0.18
10.
185
0.20
50.
205
0.28
80.
271
0.25
10.
176
0.33
40.
191
Est
onia
0.25
70.
000
0.03
4−0
.052
0.11
60.
218
0.29
50.
332
0.34
90.
364
0.34
50.
222
0.00
10.
086
0.18
3E
thio
pia
0.15
70.
205
0.23
20.
213
0.34
50.
247
0.52
90.
491
0.55
90.
597
0.52
70.
610
0.63
70.
591
0.42
4Fi
nlan
d0.
000
0.37
80.
335
0.39
30.
261
0.20
70.
168
0.19
00.
191
0.11
30.
197
0.25
10.
2