-
Review of Quantitative Finance and Accounting, 20: 385413, 2003c
2003 Kluwer Academic Publishers. Manufactured in The
Netherlands.
Banking Mergers: The Impact of FinancialLiberalization on the
Taiwanese Banking IndustryPEIYI YUDepartment of Banking and
Finance, National Chi Nan University, TaiwanE-mail: peiyi
[email protected]
BAC VAN LUULandesbank Baden-Wurttemberg, 4121 Bond Markets, Am
Hauptbahnhof 2, 70191 Stuttgart, GermanyE-mail: [email protected]
Abstract. The objective of this paper is to examine the nature
of the Taiwanese banking sector and to analyzethe impact of
financial liberalization on the Taiwanese banking industry. We
present empirical evidence to showthat the recent wave of bank
mergers observed in other countries is also suitable for Taiwan.
Based on empiricalresults for overall economies of scale and
expansion path subadditivity, Taiwanese banks should obtain the
benefitof scale economies by merging with other banks rather than
expanding by opening more branches. Furthermore,we show that the
Relative Market Power hypothesiswhich postulates that greater
market shares lead to higherprofitabilityfinds empirical support in
Taiwanese banking data after financial reforms were enacted.
Key words: scale economies, scope economies, cost efficiency,
merger
JEL Classification: C33, G21, G14, L11
Introduction
In the 1980s, technological advances in communication and
information systems, progres-sive elimination of official barriers
to capital flows, and intensification of competition in
anincreasingly deregulated environment were identified as three
major forces to shape the newlandscape for global financial
markets. As a matter of fact, these forces have either directlyor
indirectly accelerated the process of financial liberalization in
Taiwan.1 Moreover, likemany of the governments in Southeast Asia,
Taiwan foresaw that after joining the WorldTrade Organisation
(WTO), the opening of the market and foreign competition could
dec-imate local banks unless it first reformed its financial
markets internally. Thus, Taiwansaggressive banking deregulation
program was launched in the early 1990s. For instance, theTaiwanese
government allowed sixteen private commercial banks to be
established since1991 and domestic banks were granted permission to
conduct stock brokering, trading, andinvestment banking activities
through subsidiaries.2
In this paper, we examine whether the recent wave of bank
mergers observed in othercountries is suitable for Taiwan by
evaluating the competitve forces that impact on theTaiwanese
banking sector. These competitive issues are discussed from three
different
Corresponding author.
-
386 YU AND LUU
perspectives: scale and scope economies, cost efficiency and the
tests of market-power andefficient-structure hypotheses. Overall
economies of scale (OES) are important for bankmanagers and
regulators. For example, if overall scale diseconomies are found to
exist, thenpolicies that encourage mergers or expansion of branch
numbers should be reconsidered. Inthis paper, we choose expansion
path subadditivity as a more appropriate method than thetraditional
scope economy measure for examining the cost structure of banking
markets.If expansion path subadditivity, a concept proposed by
Berger, Hanweck and Humphrey(1987), is found to exist, a
combination of output is produced more cost-efficiently by alarge
bank than any combination of smaller ones. Breaking up large banks
may thus leadto higher costs for consumers. Moreover, from the
existence of OES and expansion pathsubadditivity it can be inferred
that Taiwanese banks should choose to merge with otherbanks rather
than to expand their network by opening more branches, if they want
to obtainthe benefit from scale economies. Finally, we investigate
whether the market-power (MP)and efficient-structure (ES)
hypothesis offer explanations of the observed variation in
bankprofitability. We apply a model3 similar to Bergers (1995) to
study the extent to whichthe ES and MP hypotheses can explain the
Taiwanese banking market, and still add directmeasures of both
X-efficiency and scale efficiency to the empirical analysis.4 Based
onthese studies, we may be able to judge whether the Taiwanese
banking industry shouldbe restructured through consolidation in the
next stage or not. Recently, some authorshave followed the research
agenda suggested by the seminal contribution of Berger
(1995).Altunbas et al. (2001) and Carbo, Gardener and Williams
(2002) estimate scale economies,X-inefficiencies and technical
change for European banks between 1989 and 1997. Bothstudies yield
similar results and find that X-efficiency, i.e. differences in
technological andmanagerial efficiency, accounts for most of the
variation in bank profitability while scaleeffects are
quantitatively less important. Punt and van Rooij (1999) arrive at
a very similarconclusion attaching higher weight to X-efficiency
than other factors in explaining return onequity of banks. These
results stand in contrast to the inconclusive study of Berger
(1995)himself, who states that for the US banking sector it does
not appear that any of the ES orMP hypotheses are of great
importance in explaining bank profits. The unique situation ofthe
Taiwanese financial system, with its rapid pace of change and
reform, provides an idealenvironment for studying the impact of
regulatory reform on scale and scope economies aswell as the
profit-structure relationship.
The remaining sections of this paper are structured as follows.
Section 2 outlines thefunctional form and measurement methodologies
adopted in this study. Section 3 discussesthe data sources and
shows the impact of the 1990s financial liberalization on the
differentcompetitive issues of the Taiwanese banking industry. In
Section 4 we will compare andsummarize our findings and give
suggestions for the future industrial organization of theTaiwanese
banking sector.
Specifications of models
The earlier banking literature only considers one measure of
bank output5 and uses simplestatistical models resembling
ratio-based analyses to examine scale economies in the bank-ing
industry. Although accounting ratios in banking are typically used
to obtain a partial
-
BANKING MERGERS 387
measure of banking productivity, these measures are
problematic.6 Benston et al. (1982)were the first to use the
conventional translog cost function system to estimate economiesof
scale in banking7 and evaluate the bank output by the production
approach. The con-ventional translog cost function system enables
the cost structure of banks to be modelledwith maximum flexibility
and each of the outputs can be considered explicitly. However,one
limitation to the use of the translog cost function is that the
translog cost functionform is potentially subject to
misspecification (McAllister and McManus, 1993). Since thetranslog
cost function is developed as a local approximation to the
underlying cost func-tion, the Fourier approximation and the Kernel
regression technique can overcome thesedeficiencies to provide a
global approximation by restricting the sample to homogeneousbanks.
However, the Fourier approximation and the Kernel regression
technique require alarge sample to obtain accurate results and in
particular, the Fourier approximation is moresuitable for large
banks8 (McAllister and McManus, 1993; Mitchell and Onvural,
1996).Since our sample of Taiwanese banks is small and very few of
the institutions are large insize, the translog cost function seems
the most appropriate to study the Taiwanese bankingsystem.
Moreover, the ordinary translog functional form cannot be modified
to define zerooutputs since all of the outputs enter in logarithmic
form. Therefore, we replace the originaltranslog cost function by
the Box-Coxs (1964) transformation, which is called the
hybridtranslog cost function.9
Methodology: The hybrid translog cost function system
In this study, bank multi-outputs are measured by the
intermediation approach and themodified model from Molyneux,
Altunbas and Gardener (1997) is used to examine scaleeconomies and
scope economies. In our view, the nature of banks is more
accurately de-scribed as intermediators of financial services
rather than producers of loan and depositaccount services, a view
taken by the production approach. The latter usually defines
banksoutput as the number of deposit or loan accounts or the number
of transactions performed onthese accounts. Benston, Hanweck and
Humphrey (1982) and Pulley and Humphrey (1993)are among the
contributions in the production approach literature. Kolari and
Zardkoohi(1987) argue that the intermediation approach has crucial
advantages over the productionapproach. In their view, banks
compete via nominal amounts, not the number of
accounts.Furthermore, dollar amounts constitute a common
denominator for the many kinds of ser-vices banks provide.
Therefore, the intermediation approach seems to be more
appropriatein a competitive, asset-side driven banking market. We
in fact assume that domestic banksin Taiwan operate in a
competitive environment and all of banks aim to minimize costs
withprofit-maximising behavior.
The hybrid transformation methodology evaluates a translog
functional form where thelogarithms of outputs are replaced by the
Box-Cox (1964) transformation. The Box-Coxhybrid transformation can
be written as follows:
Q(Qi ) =(Qi 1)
for other than zero, and (1)
Q(Qi ) = ln Qi for equal to zero (2)
-
388 YU AND LUU
Greene (1997) pointed out that if a minimum of the sum of
squares in the translog costfunction is found, by repeating this
procedure for different values of (from 1 to +1),the optimal value
of can be found. After determining the optimal value of , the
modelbecomes linear again and the maximum likely estimators of all
the parameters are obtained.By using the Box-Cox transformation,
the hybrid translog cost function used in this studyhas the
form:
ln TC = 0 +2
i=1i Qi +
3i=1
i ln pi + b ln B
+ 12
(2
i=1
2j=1
i j Qi Qj +3
i=1
3j=1
i j ln pi ln p j + bb ln B ln B)
+3
i=1
2j=1
i j ln pi Qj +2
i=1bi ln BQi +
3i=1
bi ln B ln pi + (3)
Where:
ln TC: The natural logarithm of the total costs for interest
costs, labor cost and capital costQi : A vector of outputs with the
Box-Cox transformation (Q1 = total loans, Q2 =
government bonds, total securities and the other investments)ln
pi : The natural logarithm of i th input prices (pi = interest
rate, p2 = wage rate and p3 =
capital price)ln B: The natural logarithm of the number of
branches
, , , , , and are coefficients to be estimated.
According to Shephards Lemma10 (Christensen, Jorgenson and Lau,
1973), the deriveddemand for an input can be inferred by partially
differentiating the cost function withrespect to the input price,
pi . Thus, three cost share equations can be generated from
thehybrid translog cost function (3) as follows:
3i=1
Si =3
i=1i +
3i=1
3j=1
i j ln p j +3
i=1
2j=1
Qj +3
i=1bi ln B + ui (4)
Since the duality theorem requires the cost function to be
linearly homogeneous in inputprices, the following restrictions
have to be imposed on the parameters of the hybrid translogcost
function (3):
3i=1
i = 13
i=1ri j = 0 for all j
(5)3
i=1i j = 0
3i=1
bi = 0 for all j
-
BANKING MERGERS 389
Also the second order parameters of the hybrid translog cost
function (3) must satisfy thesymmetry condition.
i j = j i i j = j i for all i j (6)
The hybrid translog cost function (3) is estimated jointly with
the cost share Eq. (4) usingthe seemingly unrelated regression
estimation (SURE) technique. Since the input cost shareequations
will sum to unity, one cost share equation should be omitted from
the estimatedsystem of equations to avoid the problem of a singular
contemporary covariance matrix ofdisturbances11 (Berndt, Hall and
Hansman, 1974).
Scale and scope economies
Overall economies of scale
The concept of scale economies is based on the shape of the
average cost curve. For instance,economies of scale are present up
to the level where the long-run marginal cost (LMC) curvelies below
the long-run average cost (LAC) curve. If diseconomies of scale
exist, the LMClies above the LAC curve. By following Molyneux,
Altunbas and Gardener (1997) andNoulas, Miller and Ray (1990), we
estimate OES for each bank by evaluating Eq. (7) toexamine how
changes in scale affect total cost.
OES =2
i=1
ln TC Q (7)
It is only appropriate to use Eq. (7) to estimate OES if other
regressors included in the hybridtranslog cost function remain
unchanged as outputs vary. If OES < 1, there are
increasingreturns to scale, i.e. economies of scale exist. If OES =
1, constant returns to scale exist.If OES > 1, there are
decreasing returns to scale. The existence of scale economies
meansthat the average cost of producing a product, in the long run,
decreases as more of the outputis produced.
Expansion cost subadditivity
Previous studies argue that (Molyneux, Altunbas and Gardener,
1997; Noulas, Miller andRay, 1993; Berger, Hunter and Timme, 1993)
expansion path subadditivity is a more appro-priate method than
traditional scope economy measures for examining the cost structure
ofbanking markets. The reason is that cost subadditivity can
measure the relative efficiencyof large and small firms and
consider both scale and scope economies simultaneously.
By following Berger, Hanweck and Humphreys (1987) definition,
expansion path sub-additivity is explained as whether a bank of a
given size can produce a combination ofoutputs more effectively
than two smaller banks which produce the same combination of
-
390 YU AND LUU
outputs. Expansion path subadditivity can be measured as
follows:
EPSUB(Q A) = T C(QB) + T C(QC ) T C(Q)A
T C(Q A) (8)
whereTwo kinds of output: Q1 and Q2.Two smaller banks: bank B
and bank C , and one large bank: bank A. EPSUB(Q A) means
that cost changes resulting from breaking large bank A into two
smaller bank B and bankC . If the value of the expansion path
subadditivity is positive, breaking up a large bankinto smaller
ones cannot bring about lower costs. Negative values indicate the
oppositesituation. Moreover, the overall economies of scope is a
special case of the expansion pathsubadditivity. If economies of
scope exist,
TC(Q A1 , 0) + T C(0, Q A2 ) T C(Q A) > 0 (9)
Although Eq. (9) is a special case of Eq. (8), Eq. (8) shows
whether cost effective multi-product firms should be larger or
smaller. Equation (9) explains whether the firms shouldspecialize
in production.
Cost efficiency
Leibenstein (1966, 1980) defines that inefficiency comprises
allocative inefficiency and X-inefficiency12 and also argue that
there are important distinctions in the economictheoriesunderlying
X-efficiency and technical efficiency.13
Efficiency measurement techniques
For banking cost studies, the difficulty in measuring efficiency
is the problem of disen-tangling relative cost efficiency
differences from short-term differences due to luck ormeasurement
error that temporarily give banks relatively high or low costs. In
the bankingliterature, numerous methods have been used to solve
this problem with different distri-butional assumptions. In this
case, we choose a version of the distribution-free approach(DFA)
described in Berger (1993), variants of which have also been
applied to banking databy Berger et al. (1993) and Berger and
Humphrey (1992a). Rather than imposing prede-termined distributions
on the relative cost inefficiencies and random error, DFA
identifiesone from the other methods using the basic assumption
that relative cost inefficiency dif-ferences across banks should
persist over time, while random errors should be ephemeraland
average out over time.14
Equation (10) is estimated for each of the n periods by using a
translog cost functionwhich we defined in the previous cases of
scale and scope economies, and ln x + ln v willbe treated as a
composite error term. The cost equations for each of the n periods
of a panel
-
BANKING MERGERS 391
data set can be specified as:
ln OCit = ln Ct (Yit , wi t ) + ln xi + ln i t (10)
where:
OC is operating costsC(Y, w) is a cost function with output
quantity and input price vectorsln x represents relative cost
inefficiency,ln v is a mean-zero random error, andt indexes
time.
All the components in Eq. (10) vary over time except for the
efficiency factor xi , whichis constant for bank i . For example,
bank is efficiency for 1988 will be calculated usingthe average of
each banks cost function residual over 198587 and 198991. The
currentresidual is excluded from the computation of current
relative cost inefficiency given thatcurrent costs are used to
compute current profits. Furthermore, the ln xi t are
transformedinto a normalised relative cost efficiency measure as
follows:
X EFFit = exp(ln xmint ln xi t
) (11)where:
ln xmint indicates the minimum ln xi t for all i for that t .It
may be seen that this is an estimate of xmin/xi , the ratio of
estimated costs for the most
efficient bank in the sample to the predicted costs for bank i
for any given vectors of outputsand input prices. This corresponds
with the conventional notion of efficiency as the ratioof the
minimum resources needed for production to the resources actually
used, and rangesover [0, 1].
Tests of market-power and efficient-structure hypotheses
The purpose of this part is to distinguish the market-power and
efficient-structure hypothesesas they apply to the Taiwanese
banking market. These hypotheses stress different factors
inexplaining the performance, i.e. the profitability of banks. From
the banking regulators pointof view, there is a fundamental
trade-off between maintaining strong competition on the onehand and
promoting the exploitation of scale economies and financial sector
stability on theother. Both goals may often directly contradict one
another and studying and quantifying theforces that shape the
performance of financial institutions may enable policymakers to
decidewhich of the two goals they should pursue in a given market
structure. In brief, four majorhypotheses have emerged in the
banking literature to explain the profit-structure relationship.The
market-power (MP) hypotheses comprise the
structure-conduct-performance (SCP) andrelative-market-power (RMP)
and the efficient-structure (ES) hypotheses include ESX
(X-efficiency version of the efficient-structure hypothesis) and
ESS (scale efficiency versionof the efficient-structure
hypothesis). The SCP hypothesis states that banks set prices
that
-
392 YU AND LUU
are less favourable to consumers in more concentrated markets
because of competitiveimperfections. The RMP hypothesis suggests
that only banks with large market shares andwell-differentiated
products can exercise market power in pricing these products and
earnsupernormal profits (Shepherd, 1982). Although both hypotheses
appear very similar, theyhave fundamentally different implications
in terms of consumer welfare. While consumersare unambiguously
worse off when SCP hypothesis holds, the case is less clear-cut
with theRMP hypothesis. Under the latter and if SCP is rejected at
the same time, banks are thoughtto have gained their market share
by providing well-differentiated or novel products. UnderESX, banks
with superior management or production technologies have lower
costs andhence can earn higher profits. Moreover, since these more
efficient banks are also assumedto gain large market shares that
may result in high concentration, the positive
profit-structurerelationship is spurious in this case (Peltzman,
1977). Finally, under ESS, all banks haveequally good management
and technology, but some banks simply produce at more
efficientscales than others. This hypothesis also yields a positive
profit-structure relationship as aspurious outcome since these
banks are assumed to have large market shares that may resultin
high levels of concentration (Lambson, 1987).
We apply a model similar to Bergers (1995) and employ direct
measures of both X-efficiency and scale efficiency to the empirical
analysis.
ROE = f1(CONC, MS, Relative cost efficiency, SEFFE, MGTH,Dummy
variables) + (12)
CONC = f2(Relative cost efficiency, SEFFE, MGTH, Dummy
variables) + (13)MS = f3(Relative cost efficiency, SEFFE, MGTH,
Dummy variables) + (14)
Table 1 summarizes the definitions for all variables in this
model of profit-structurerelationship.
Table 1. Definitions for all variables in the model of
profit-structure relationship
Symbol Definitions
ROA Ratio of net before-tax income to assets.ROE Ratio of net
before-tax income to equity.CONC Herfindahl index of concentration
of deposit marketMS Bank is share of total market deposit.Relative
Cost Relative cost efficiency: ratio of the smallest (n 1)-year
average residual ofEfficiency all banks to the banks (n 1)-year
average residual (current years data
excluded). The smallest and largest 1 percent are set equal to
the 1st and 99thpercentiles, respectively.
S-EFF Scale efficiency can be obtained from the previous case of
scale economies.S-EFFe Scale economy efficiency: equals S-EFF if
bank is below efficient scale; equals
1 otherwise.S-EFFd Scale diseconomies efficiency; equals S-EFF
if bank is above efficient scale;
equals 1 otherwise.MGTH Real growth of deposits in banks
markets.Dummies Dummies for (n 1) different bank groups.Source:
This table is made by author.
-
BANKING MERGERS 393
The major Eq. (12) is shown to be a valid reduced form for all
of the hypotheses and anyor all of them may be found to be
consistent with the data. A positive
profit-concentrationrelationship occurs because concentration
(CONC) affects price and price affects profit. Onthe other hand,
under the RMP hypothesis, market share (MS) becomes the key
exogenousvariable since banks with large market shares have
well-differentiated products and are ableto exercise market power
in pricing these products. Furthermore, if only RMP holds, CONCwill
have a zero coefficient because CONC is only spuriously related to
profit through itscorrelation with MS.
By contrast, if ES hypotheses are accepted, the coefficients of
the appropriate efficiencyvariables will be positive and the
coefficients of all the other key variables are either
relativelysmall or zero. An important limitation of the
reduced-form profit equation in (12) is that ittests only one of
the three necessary conditions of the ES hypotheses. More
precisely, inorder to explain the profit-structure relationship
spuriously, two more conditions (Eq. (13)and Eq. (14)) should be
met since both profits and the market structure variables mustbe
positively related to efficiency. For instance, one of the
conditions required is that inEq. (14), i.e. that more efficient
firms have greater market shares. This requirement can beexplained
by the fact that more efficient banks obtain greater market share
through pricecompetition or through acquisition of less efficient
banks.
Estimation and results
Since our sample is small in size, we use pooled time-series and
cross-section data to es-timate the hybrid translog cost function
system. This approach is different from that ofprevious studies,
which use a single year to investigate economies of scale. Although
thepositive serial correlation and heteroscedasticity will still
exist, using panel data enablesus to investigate the relationships
between temporal changes and across-sectional differ-ence. We
employ the seemingly unrelated regression estimation (SURE)
technique, whichis particularly useful with large panel data sets
(Avery, 1977) to estimate several equa-tions simultaneously. In
this specific error components model, the regression errors in
eachequation are assumed to be composed of three independent
componentsone componentassociated with time, another with
cross-sectional units, and a third with each observation.15
u jnt = jn + j t + jnt (15)
The model developed above makes the assumptions that both within
and between equationerror covariances are composed of independent
individual, time period, and observationcomponents and the
covariances of all three components are non-zero.
In this study, the hybrid translog cost function system
comprises the hybrid translog costfunction (3), the two cost share
Eq. (4), two restrictions (5) and the symmetry condition
(6).Moreover, since the Taiwanese banking industry was heavily
regulated until the beginningof the 1990s, a pre- and post-analysis
of the changes and an assessment of their impacton the banking
sector is conducted. For instance, as part of this regulation,
entry into theindustry has been restricted.
-
394 YU AND LUU
The data resources and definitions of variables
The data resources
In this study, the major data resources were banks balance
sheets and income statementsobtained from the Central Bank of China
(CBC).16 The data on branch numbers for Tai-wanese banks were also
gathered from the Central Bank of China (CBC).17 Other
relevantinformation which was not available in the Central Bank of
China (CBC) was obtainedfrom the following sources. Personnel
expense was obtained from the Bureau of Mone-tary Affairs.18 The
general index of consumer price in the Taiwan area in each year
wasavailable from Directorate-General of Budget, Accounting and
Statistics, Executive Yuan,Republic of China.19 The number of total
employees for each bank was obtained from theinternational bank
database BankScope. Given the chosen intermediation approach, we
usetwo categories of outputs, three kinds of input variables and
one control variable in ourfollowing models. All variables in this
study are measured in NT million dollars. Data fromincome
statements are gathered from 1st of January to 31st of December for
each year. Datafrom balance sheets and the other official reports
are obtained on 31st of December for eachyear. Finally, each
variable should be deflated by the general index of consumer price
inTaiwan for each year to correct for price inflation. All
variables in this paper are defined inAppendix 1.
The sample period of study
Since financial deregulation in the early 1990s offers the
potential for a pre- and post-analysis of the changes, we separate
the whole sample period into two shorter periods:19851991 and
19931997. 1992 is omitted because the reforms were enacted during
thisyear and some of new established banks did not have data for
the whole year. The em-pirical analysis ends in 1997, as another
shift in the banking sector regime took placeafterwards. After
1997, Taiwan started privatizing some of the large
government-ownedbanks. Since the latter were subject to numerous
restrictions in their operations, such aslimitations on the number
of senior management, restrictive pay and bonus stipulations andthe
inability to fire poor performers, they were perceived as
slumbering giants. TaipeiBankstarted privatizing in 1997 and
finished its stock offering in July 1997. Until October 2002,five
more formerly government-owned20 banks completed their
privatization. Thus weconsider the steps taken by the government to
float the large public banks to constitutea substantial change in
the external environment and restrict the post-reform analysis
tothe period before 1997. Moreover, the question of whether to pool
the data or not natu-rally arises with panel data. In our case, we
partition each period data into two subsam-ples to carry out Chows
breakpoint test. All the tests for poolability are summarized
inTable 2.
Table 2 shows that Chows breakpoint test does not reject
poolability across time pe-riods, under the null hypothesis: H0 = t
= for t = 1, . . . , T . It can be conc-luded that parameters in
our model are constant, and stable in the
estimatedrelationship.
-
BANKING MERGERS 395
Table 2. Chows breakpoint test performed for poolability
Sample Period Test Statistics Critical Value Decision
19851991 1.95 F0.01(21, 105) = 2.03 Not rejected19931997 1.90
F0.01(21, 143) = 1.99 Not rejected
The constitution of the sample in this study
Before 1991, the Taiwanese government imposed many restrictions
on the Taiwanese bank-ing market and there were only twenty-four
domestic banks in the market. From 1991 to1992, the Taiwanese
government started to relax some financial restrictions imposed on
thebanking market, and sixteen new banks were established during
these two years. The sampleis extended to include 38 domestic banks
in the period between 1993 and 1997. Foreignbanks are excluded in
our sample, although one main rationale of allowing competitionfrom
foreign institutions is the expected gain in the operation
efficiency and service qualityin the local financial market.21
However, statistics on the market share of foreign banks
indifferent areas of the banking business indicate that derivatives
trading, foreign exchangetrading and guarantee constitute main
business of foreign banks.22 In 1994, Taiwan sig-nificantly revised
the Guidelines for the Reviewing of Foreign Banks Applications for
theEstablishment of Branch and Representative Offices in accordance
with the GATT. Thoserevisions aimed at according foreign banks
national treatment to compete on an equal foot-ing with local
banks. They did not remove all restrictions on foreign banks,
however, so thatthe latter were not allowed to acquire more than
50% of local banks until 1999. Because ofthe different business
focus of foreign banks and the diverging restrictions on expansion
bymergers, we drop them from the sample.
Furthermore, based on bank asset size and business similarity,
we can divide the wholesample of domestic banks into four different
small subgroups: Government-owned banks(GOB), local banks (LB),
old-private banks (OPB) and new-private banks (NPB). GOBsare
subject to government control in their day-today operations and
have average assets ofaround 25 billion US-$, while the three other
groups are much smaller with average assetsof roughly 5 billion
US-$. OPBs and LBs were established before financial reform,
wherethe latter faced geographical restrictions regarding their
branch network. Those restrictionswere abolished by the financial
reform in 1991, which also enabled NBPs to be established.In this
paper, the empirical results of scale and scope economies will be
compared betweenthese four groups. These four groups of banks are
listed in Appendix 2.
OES and expansion path subadditivity
Estimation of the hybrid translog cost function system
We show all coefficients derived from the hybrid translog cost
function system and thetwo cost share equations from (3) to (6) in
the following Table 3. In brief, our empiricalresults in Table 3
are similar to the findings of the banking studies reviewed earlier
(Mester,
-
396 YU AND LUU
Table 3. Empirical results of the hybrid translog cost function
system
19851991 19931997Coefficient = 0.1 = 0.5
Constant 3.0675 6.1665(0.3383) (0.1915)
Q1 0.2197 0.0048(0.0786) (0.0005)
Q2 0.0150 0.0031(0.0555) (0.0009)
ln p1 0.1906 0.4790(0.0619) (0.0403)
ln p2 1.2106 0.2180(0.1112) (0.0270)
ln B 0.3961 0.7037(0.1960) (0.1699)
Q1 Q1 0.0308 2.70E-07(0.0123) (5.97E-07)
Q1 Q2 0.0209 6.63E-07(0.0084) (8.70E-07)
Q2 Q2 0.0156 4.75E-07(0.0082) (2.16E-06)
ln p1 ln p2 0.0616 0.0204(0.0047) (0.0036)
ln p1 ln p3 0.0645 0.0363(0.0086) (0.0069)
ln p2 ln p3 0.0037 0.0181(0.0056) (0.0030)
ln B ln B 0.0227 0.1054(0.1842) (0.0817)
ln p1 Q1 0.0603 7.87E-05.0054) (2.83E-05)
ln p2 Q1 0.0137 2.46E-05(0.0024) (1.02E-05)
ln p1 Q2 0.0324 0.0001(0.0048) (6.45E-05)
ln p2 Q2 0.0096 4.33E-05(0.0016) (2.56E-05)
ln B Q1 0.0536 0.0007(0.0306) (0.0002)
ln B Q2 0.0488 0.0008(0.0067) (0.0003)
ln B ln p1 0.0238 0.1444(0.0147) (0.0064)
ln B ln p2 0.2502 0.0683(0.0353) (0.0075)
Adjusted R squared of the 0.9968 0.9992hybrid translog cost
function
Approximate standard error in parentheses.Significantly
different from zero at 10% level.Significantly different from zero
at 5% level.Significantly different from zero at 1% level.
-
BANKING MERGERS 397
1987; Molyneux, Altunbas and Gardener, 1997) since all the
coefficients of input prices arestatistically significant. There
are two factors of input, capital and labor, whose prices
aredenoted by P1 and P2 respectively and two kinds of output, total
loans and investments, Q1and Q2. Moreover, we also consider the
number of branches B as one factor impacting ontotal cost. The
coefficients on the output variables and input factors are
significant, exceptfor total investments before financial
liberalization. Post-reform, however, investments alsobecome highly
significant in determining total costs, which indicates that banks
increasinglyengaged in securities dealing, brokerage, underwriting,
as well as investment management.We also find that before the
financial reform, labor has the most influential role in
determin-ing total cost, hence we argue that banks did not have an
incentive or the ability (e.g. if thegovernment imposes
restrictions on laying off employees) to control labor cost
efficiently.This result is similar to the finding obtained for the
Italian banking system (Molyneux,Altunbas and Gardener, 1997). The
results subsequent to the financial reform suggest thatthe cost of
interest input becomes more important than the other two kinds of
inputs. Thecoefficients of interest cost and labor cost are both
significant, but the magnitude of theinterest cost has increased.
Finally, the number of branches B are significant in
explainingtotal cost pre- and post-reform alike. The remaining
variables are merely cross-products ofthe main factors.
Empirical results of OES
We summarize the empirical results of OES for the Taiwanese
banking industry in Ta-ble 4. The average value of OES obtained
from the hybrid translog cost function system isaround 0.3470. This
means that Taiwanese banks were able to obtain the benefit from
OESbefore financial reform. This value is similar to the result for
Spain whose value of OESis nearly equal to 0.3695 (Molyneux,
Altunbas and Gardener, 1997). Evidence from theEuropean banking
sector using Fourier flexible functional form and stochastic cost
frontiermethodologies indicate that scale economies range between 7
to 10%, while X-efficiencymeasures appeared to be much larger at
about 22% (Carbo, Gardener and Williams, 2002;Altunbas et al.,
2001). From Table 4, we also find that after 1992, OES still exist
in the
Table 4. Empirical results of OES for the Taiwanese banking
industry from1985 to 1997
Model I of the Hybrid Conventional TranslogTranslog Cost
Function System Cost Function System
19851991 0.3470 0.5629(0.0420) (0.0855)
19931997 0.0030 0.7017(0.0011) (0.0568)
Approximate standard error in parentheses.Significantly
different from one at 10% level.Significantly different from one at
5% level.Significantly different from one at 1% level.
-
398 YU AND LUU
Taiwanese banking market. However the value of OES had gone down
dramatically nearlyto 0.0030. Since the average value of OES is
extremely small comparing with the resultsobtained from the
previous conventional translog cost function studies, it might be
usefulto reestimate the OES by using the conventional translog cost
function system. Moreover,we show these results estimated by using
the conventional translog cost function system inTable 4. The
values of OES obtained from the conventional translog cost function
systemare much bigger than the results estimated by the hybrid
translog cost function system, buttheir values are still smaller
than one and indicate that OES exist in the Taiwanese
bankingmarket. For the pre-analysis, the average value of OES is
0.5629, although the value isincreased to 0.7017 after financial
reform.
Based on Table 4, we argue that the specification of the hybrid
translog cost function andtwo-stage estimation procedure sometimes
may be causing problems for the OES estimates.Comparing the results
before and after financial reform, both types of cost functions
indicatethat economies of scale still exist post financial
liberalization. However, the estimates forOES based on the hybrid
translog cost function are very low with a value of 0.003,
implyingthat total cost would increase only by 0.003% if output
were raised by 1%. The potentialbenefits from increasing overall
scale hence are unusually large. This is particularly odd
asinflation-adjusted average assets of domestic banks actually
increased by 30% in 19931997compared to 19851991. A previous study
on European banking markets by Molyneux,Altunbas and Gardener
(1997) using hybrid translog cost functions reports values for
OESranging from 0.37 to 0.74. Our results for the conventional
translog cost functionfall withina similar range before and after
financial reform and thus seem to be a more reasonablemeasure of
true scale economies. The two stage estimation method used for
deriving thehybrid translog cost function may be less robust to
structural breaks such as the financialreforms enacted in Taiwan in
1992 and the subsequent establishment of new banks.
According to the empirical results of OES, we discover that OES
actually exist in Tai-wanese banking industry. However, there are
two ways for banks to obtain the benefit ofscale economies as
follows:
1. Taiwanese banks can achieve OES by increasing branch numbers
or2. Taiwanese bank can increase their size by mergers.
To find out which way can benefit the Taiwanese banking sector,
we examine the presenceof expansion path subadditivity in the
following part.
Empirical results of expansion path subadditivity
We can use Eq. (8) to estimate expansion path subadditivity and
divide all representativebanks into two smaller bank groups by
taking the mean value of outputs. In our case, we usethe average
output prices for the whole group to divide the whole sample into
two subsam-ples. The empirical results of expansion path
subadditivity are given in the following Table 5.
From Table 4, our estimated values of expansion path
subadditivity are always positive.This result suggests that the
Taiwanese banking market exhibits expansion path subadditivityand
breaking up large banks into smaller ones can lead to higher costs.
From the point of
-
BANKING MERGERS 399
Table 5. Expansion path subadditivity from 1985 to 1997 (%)
Model I
1985 1.23121986 1.23561987 1.23711988 1.23651989 1.25481990
1.27941991 1.3093The financial environment of Taiwanese banking
market was
dramatically changed around 19921993 1.59231994 1.59011995
1.58801996 1.58391997 1.5803
view of cost efficiency and scale, it may be concluded that it
is better to have fewer banksin the Taiwanese banking market. This
strong conclusion has to be balanced against thepotential drawbacks
of reduced competition or increased need for government
regulationwhen banks become larger. Compared with other previous
studies, Noulas, Miller andRay (1993) also find positive values of
expansion path subadditivity for medium-sized USbanks and Molyneux,
Altunbas and Gardener (1997) find that the France, Germany and
Italybanking markets are natural monopolies and have a tendency for
banks to become large.Finally, based on the empirical results of
expansion path subadditivity and OES obtained,we can infer that
Taiwanese banks should choose to merge with other banks rather than
toexpand their network by opening more branches to obtain the
benefit from OES.
Cost efficiency
In our case, relative cost efficiency will include allocative
cost efficiency and X-efficiency. Inthis part, the theoretical
models are applied to the same data set as in the cases of scale
andscope economies. However, in this part, the entire data set is
reconsidered for five groups: thegovernment-owned specialised banks
(GOSB), the government-owned commercial banks(GOCB), local banks
(LB), old-private banks (OPB) and new-private banks (NPB).
Average cost efficiency
We obtain relative cost efficiency by estimating the previous
hybrid translog cost functionsystem with a common effect. The table
below summarizes the results of relative costefficiency for
different Taiwanese bank groups.
From Table 6, we observe that after financial reform, different
bank groups have a similarrelative cost efficiency. However, the
range of relative cost efficiencies is very wide forindividual
banks, even those within the same group (please refer Appendix 3.
For example,the most efficient bank, YP belongs to the most
inefficient group of new private banks.
-
400 YU AND LUU
Table 6. Empirical results of relative cost efficiency for
different Taiwanese bank groups
19861991 (before financial reform) 19941997 (after financial
reform)& Model I & Model I
The government-owned 0.7416 0.7441specialised banks (GOSB)
(0.0384) (0.0459)
The government-owned 0.8205 0.7971commercial banks (GOCB)
(0.1097) (0.0883)
Local banks (LB) 0.7468 0.7892(0.0476) (0.0653)
Old-private banks (OPB) 0.7680 0.7983(0.0267) (0.0575)
New-private banks (NPB) 0.7176(0.1069)
Source: Calculated from the data which are collected by the
Central Bank of China (CBC).
From Table 7, we observe that for some banks, the explicit
repeal of regulations mayresult in an increase in allocative
efficiency, while a general increase in the level of competi-tion
permitted increases in technical efficiency. However, we also find
that for some banks,the results obtained after financial reform are
reversed. Furthermore, the rankings of banksaccording to their
relative cost efficiency are changed dramatically and their
rankings do notdepend on which bank group they belong to. This
stands in contrast to recent results fromChen and Yeh (1999), who
employ a Data Envelopment Analysis to evaluate the
relativeefficiency of the Taiwanese banking sector. They find that
publicly-owned banks predom-inantly manage their resources less
efficiently than private banks. Among the inefficientbanks they
identified, technical inefficiency is said to be the main source of
inefficiencyrather than scale factors. From Appendix 3 it can be
seen that both the most efficient bankand the most inefficiency
bank are in the NPB (new private bank) group. This result maybe
explained by the different management strategies which banks choose
to implement tocounter the increasing competitive pressure, such as
entering new markets or the creationof innovative products.
The results of tests of market-power and efficient-structure
hypotheses
Firstly, the panel data indicates that the sample of banks is
homogeneous so that the degreeof pooling is valid (Please refer to
Appendix 4). Then, we describe empirical results of thebasis model
and investigate the market-power (MP) and efficient-structure (ES)
hypothesesas alternative explanations of the observed variation in
bank profitability.
Empirical results before financial reform
In Table 8, the coefficient of concentration (CONC) in the major
Eq. (12) is negativeand statistically significant at the 1%
critical level. This means that there is a
negativeprofit-concentration relationship in the Taiwanese banking
industry, which contradicts thestructural-conduct-performance (SCP)
hypothesis. Since the market share (MS) coefficientin Eq. (12) is
positive, but not significant, the relative market power (RMP)
hypothesis
-
BANKING MERGERS 401
Table 7. Rank of asset and relative cost efficiency for each
bank from 19861997
19861991 19941997
Relative Rank of Relative Rank ofRank Cost Relative Cost Rank of
Cost Cost Relative
Asset of Asset Efficiency Efficiency Asset Asset Efficiency
Efficiency
A 223735.8 9 0.7039 19 380119.0 12 0.7579 19B 208569.6 10 0.7404
15 404278.5 9 0.6928 29C 698042.2 2 0.8378 4 1486397.0 1 0.7586 18D
241936.8 8 1 1 428851.5 8 0.7624 16E 501024.8 5 0.7807 9 1090496.0
3 0.7815 13F 797036.8 1 0.6307 21 1462416.0 2 0.6286 35G 503190.0 4
0.8069 5 815007.3 4 0.8724 5H 542943.0 3 0.7745 12 806957.5 5
0.8400 10I 493078.2 6 0.8430 3 763248.5 6 0.8418 9J 124909.8 12
0.7549 13 346554.5 13 0.8782 3K 174470.8 11 0.7807 10 388858.8 11
0.8019 11L 51660.6 17 0.7983 7 190100.0 17 0.7560 17M 62733.8 16
0.7381 16 180944.3 18 0.7530 22N 344359.6 7 0.8504 2 701428.8 7
0.8759 4O 79542.2 13 0.7803 11 227947.5 14 0.7680 15P 64510.0 14
0.7833 8 206337.8 16 0.7972 12Q 62944.2 15 0.7186 18 219412.0 15
0.7405 23R 29589.2 18 0.7507 14 123779.3 20 0.7105 27S 26687.8 19
0.7336 17 99786.3 33 0.7549 21T 8027.2 20 0.6612 20 35913.5 38
0.8709 6U 4088.6 21 0.7999 6 42995.5 37 0.8822 2YA 109731.3 25
0.7083 28YB 105654.3 27 0.8448 8YC 107387.5 26 0.7138 26YD 105589.0
28 0.7768 14YE 88994.3 35 0.8542 7YF 91435.8 34 0.7556 20YG
114039.8 23 0.7375 24YH 114068.0 22 0.6696 31YI 115746.0 21 0.6127
36YJ 104275.8 29 0.6591 33YK 102162.0 32 0.6468 34YL 124227.0 19
0.6708 30YM 110794.5 24 0.7239 25YN 103121.3 30 0.6671 32YO
102876.5 31 0.5955 37YP 82849.3 36 0.9970 1YQ 396985.5 10 0.5661
38Source: Calculated from the data which are collected by the
Central Bank of China (CBC).
does not explain the profit-structure relationship well,
although efficiency variables arecontrolled for in the equation. We
also observe that the coefficient of the relative costefficiency
(Efficiency) is positive (insignificant) in the major Eq. (12), but
not positivelyrelated to market share in the market share Eq. (14).
Hence the ESX hypothesis cannot
-
402 YU AND LUU
Table 8. The sample banks with the period of 19861991 are
estimated by the FGLS procedures
Variable ROE (Eq. (12)) CONC (Eq. (13)) MS (Eq. (14))
Constant 0.9495 0.0806 0.3142(0.4850) (0.0216) (0.0654)
Concentration rate (CONC) 5.0050(1.3871)
Market share (MS) 1.3975(0.8716)
Relative cost efficiency (Efficiency) 0.3409 0.0490
0.0025(0.2578) (0.0125) (0.0452)
Scale economy efficiency (SEFFE) 0.4412 0.0097 0.0593(0.2018)
(0.0123) (0.0272)
Scale diseconomy efficiency (SEFFD) 0.3391 0.0223 0.2318(0.2614)
(0.0113) (0.0352)
Real growth of deposit market (MGTH) 0.0551 0.0351
0.0423(0.1889) (0.0054) (0.0179)
Adjusted R-squared 0.2438 0.1557 0.9100Approximate standard
error in parentheses.Significantly different from zero at 10%
level.Significantly different from zero at 5% level.Significantly
different from zero at 1% level.
significantly contribute to the explanation of the
profit-structure relationship. It seems thatsuperior management
techniques or technological differences between banks do not playan
important role in explaining differences in profitability.
Turning to the scale efficiency results, since the scale
efficiency coefficient (SEFFE) isnegative and significant at 5%
critical level in the major Eq. (12), we may conclude thatthe ESS
hypothesis contradicts the profit-structure relationship of the
Taiwanese bankingmarket before financial reform. This may happen
when banks try to have higher profits by ex-ploiting economies of
scale. At the same time, the benefit is depreciated by the
invisible cost,such as a change in management strategies or a
change in relative cost efficiency following amerger.
Empirical results after financial reform
From the empirical results in Table 9, the market share (MS)
coefficient in the majorEq. (12) is positive and significant at the
1% critical level. This suggests that the relativemarket power
(RMP) hypothesis can explain part of the profit-structure
relationship ofthe Taiwanese banking market, while efficiency
variables are controlled for. According tothe relative market power
hypothesis (RMP), Taiwanese banks should obtain relative
largemarket share by producing well-differentiated products in
order to attract more consumersif they want to earn supernormal
profit.
In order to explain the profit-structure relationship
spuriously, both profits and the marketstructure variables
(concentration and market share) must be positively related to
efficiency.However, in the major Eq. (12), the coefficient of the
relative cost efficiency (Efficiency) isnegative (significant at 5%
critical level) and not positively related to market share in
the
-
BANKING MERGERS 403
Table 9. The sample banks with the period of 19931997 are
estimated by the FGLS procedures
Variable ROE (eq. (12)) CONC (eq. (13)) MS (eq. (14))
Constant 0.3382 0.0242 0.1431(0.0744) (0.0016) (0.0248)
Concentration rate (CONC) 1.2954(2.4342)
Market share (MS) 1.1115(0.2809)
Relative cost efficiency (Efficiency) 0.1305 0.0022
0.0607(0.0591) (0.0011) (0.0164)
Scale economy efficiency (SEFFE) 0.3086 0.0055 0.0790(0.0969)
(0.0019) (0.0283)
Real growth of deposit market (MGTH) 1.1094 0.2740
0.1553(0.5429) (0.0049) (0.0754)
Adjusted R-squared 0.5348 0.9603 0.6508Approximate standard
error in parentheses.Significantly different from zero at 10%
level.Significantly different from zero at 5% level.Significantly
different from zero at 1% level.
market share Eq. (14). We conclude that the ESX hypothesis
cannot determine part of theprofit-structure relationship.
On the other hand, the scale efficiency coefficient (SEFFE) is
negative and significant atthe 1% critical level in the major Eq.
(14). We may conclude that ESS hypothesis finds nosupport in the
Taiwanese banking market after financial reform. Moreover, the
coefficientof the real growth of deposit market (MGTH) is
positively related to ROE and significantat 5% critical level in
the major Eq. (12). This means that the real growth rate of
thedeposit market can bring about a positive effect on the
profitability of Taiwanese banks.The results of our analysis of the
market structureprofit relationship are summarized inTable 10. It
is striking that none of the four alternative explanations offers a
superior approachto explaining profitability in the Taiwanese
banking sector before financial reforms wereenacted. This is very
similar to conclusions drawn by Berger (1995) for the US
bankingindustry using essentially the same empirical model. The
latter author finds that the RMPand ESX are weakly supported by
single year cross-section analyses for the period between19801989.
However, as the median R2 of the regressions are below 10%, he
concludesthat it is impossible to distinguish whether the profit
structure relationship reflected superior
Table 10. Summary of results on market structureProfit
relationship
19861991 19931997
Market powerStructural conduct performance (SCP) Rejected
RejectedRelative market power (RMP) Insignificant Accepted
Efficient structureX-Efficiency (ESX) Rejected RejectedScale
efficiency (ESS) Rejected Rejected
-
404 YU AND LUU
management (as according to the ESX hypothesis) or greater
market power (as postulatedby the RMP theory). For the period after
financial reform, however, our own results forTaiwan are more
clear-cut. While the SCP, ESX and ESS are still rejected even after
theliberalization of financial markets, the hitherto insignificant
coefficient of market sharebecomes significantly positive, offering
firm support to the RMP thesis. The liberalization offinancial
markets seems to have favoured banks with well differentiated
products, enablingthem to gain a high market share. Those banks
have been able to exploit their market power tobecome more
profitable than their smaller competitors. Unlike the SCP
hypothesis, however,the acceptance of the RMP hypothesis is not a
strong argument against banks becomingbigger. Consumer welfare may
even benefit from banks pursuing product innovation
anddifferentiation actively. Under the RMP hypothesis, market power
is not derived from a highdegree of concentration, as the rejection
of SCP shows, but from the provision of productsmore tailored to
the needs of customers, as suggested by the higher market share of
thesebanks.
Conclusion
In the early 1990s, when the Taiwanese government wanted to
enhance local banks com-petitiveness, financial markets were
liberalized and the government allowed sixteen privatecommercial
banks to be established. To address the impact of financial
liberalization on theTaiwanese banking industry, we investigate the
profit-structure relationship, relative costefficiency, and examine
whether the current wave of mergers observed elsewhere might
besuitable for Taiwanese banking. For all sample banks and the
whole Taiwanese bankingindustry, the values of OES are
statistically significant at the 1% level and smaller than one.The
implications of the results are that, if Taiwanese banks want to
benefit further from OES,they should produce more output. Moreover,
since our positive values of expansion pathsubadditivity indicate
that breaking up large banks into smaller ones can lead to higher
costs,the Taiwanese banking market is characterized by expansion
path subadditivity. The latteralso implies that, if Taiwanese banks
want to obtain the benefit from OES, they shouldchoose to merge
with other banks rather than to expand their network by opening
morebranches. The results for expansion path subadditivity suggest
that, from a cost perspective,the Taiwanese banking industry would
be better off with fewer banks.
Turning to the cost efficiency results, we observe that after
financial reform, differentbank groups have similar relative cost
efficiency, but their rankings do not depend on whichbank group
they belong to. However, for some banks, comparing with the
pre-analysis,the results obtained after financial reform are
reversed and the rankings of relative costefficiency are changed
dramatically.
Finally, we study four alternative explanations of the observed
profitability of banks usinga reduced-form model suggested by
Berger (1995). Before financial reform, none of the fourhypotheses
has significant explanantory power for variations in profitability.
After financialreform, however, it is striking that the strength of
empirical support for the RMP hypothesisincreased. Although the SCP
and RMP hypotheses are related in the sense that they linkmarket
power with profitability, there is a fine but crucial difference
between them. Whilemarket power in the RMP hypothesis is said to
stem from innovative and well-differentiated
-
BANKING MERGERS 405
products (hence the attribute relative), the SCP hypothesis
states that high concentrationitself is the source of market power.
If SCP was accepted and the three other hypothesesrejected for
Taiwan, it would have substantially weakened the normative case for
furtherconsolidation. It would mean that banks already use their
oligopolistic power to chargeunfavourable prices to consumers,
while other sources of higher profitability, such as
scale,technical efficiency or high relative market share are ruled
out. The insignificance of RMPbefore financial reform and
acceptance thereafter suggests that government regulations inplace
before 1992 weakened the positive correlation of market share and
profitability. IfTaiwanese banks want to achieve higher profits in
the liberalized environment, they shouldaim at developing highly
differentiated products to increase their market share. Since SCP
isin fact rejected in our study, they cannot rely on oligopolistic
pricing power to enhance theirprofitability. This also suggests
that banks strive for higher market shares is not
necessarilydetrimental to consumer welfare. The rejection of ESS
and ESX implies that neither scalefactors nor technical or
managerial efficiency played a significant role in determining
bankprofitability. Our analysis of scale economies and expansion
path subadditivity indicatesthat overall efficiency gains may be
reaped through a trend towards bigger and fewer banks.It should be
pointed out, however, that consolidation should proceed cautiously
so as notupset the balance between efficiency considerations on the
one hand and the interests ofTaiwanese bank customers on the other.
After emerging from years of protected regulation,Taiwans efforts
to fully liberalize its financial sector will, in our view, provide
importantlessons for China and South East Asia.
Appendix 1: Definition of variables
Definition of input variables
Since we assume that Taiwanese domestic banks are in a
competitive market, we can con-sider input prices as exogenous
variables.
1. p1, the average price of interest rate: p1is the average
interest cost per dollar of interest-bearing total deposits and
total borrowed funds.
Interest cost = p1 Rwhere: R = (total deposits23) + (borrowed
funds)
But if banks hold government deposits, they do not have to pay
interest expense. Accord-ing to this reason, government deposits
must be eliminated from total deposits, then wecan get more
accurate interest cost.
The average price of interest rate can be calculated by the
equation below:
p1 = interest cost(total deposits + borrowed funds)2. p2, the
average price of labor:
p2 = personnel expense for yearAverage employee number per
branch branch number
-
406 YU AND LUU
Subsequently, it is difficult to differentiate between labor
expense and capital expense, be-cause these two specific items are
included in the big item:the selling and administrativeexpense.24
We can not gather more details for labor expense and capital
expense. To over-come this difficulty, in this study labor expense
and capital expense are measured in thefollowing way:
Personnel expense include wage, overtime pay, reward, pensions,
bonus and so on. Weuse the personnel expense from Financial
Statistics Abstract published by Bureau ofMonetary Affairs from
1994 to 1997. By observing that ratios of personnel expense
dividedby selling and administrative expenses are almost constant
by years, we can calculate theaverage ratio for each bank by the
available data. Since we use the average ratio timesselling and
administrative expenses, the personnel expense for each bank in
every year canbe inferred. Finally, according to the equation, the
average price of labor can be obtained.3. p3, the average price of
capital: The average price of capital is calculated by the
followingequation:
p3 = summing the capital expensenet fixed assets
= summing the capital expense(fixed assets accumulated
depreciation)Many studies on the structure of costs in banking
define capital equipment as the sum ofconcepts like rent,
depreciation, furniture and equipment (Mester, 1987; Murray and
White,1983). In this study, we assume capital expenses include four
specific items:
(1) the depreciation for fixed assets and all equipment(2)
rental expense(3) the expenses for maintenance and repair(4)
insurance cost
Because of the same difficulty, we can not obtain the capital
expense directly from balancesheets and income statements. The data
we only can obtain are selling and administrativeexpenses. By using
the relationship described below,
Capital expense = (selling and administrative expenses personnel
expense)
The capital expense can be inferred for each bank in every year.
Since selling and adminis-trative expenses includes not only
personnel expense, capital expense but also the expensesfor water
supply, electricity, and advertisement, the only disadvantage is
capital expense isoverestimated slightly.
Definitions of two categories of outputs
The empirical approach to output definition in this study is
supported theoreticallyby Molyneux, Altunbas and Gardeners (1997)
model of scale and scope economies in
-
BANKING MERGERS 407
European banking markets. The definitions of outputs in this
study are similar with thedefinition in Kolari and Zardkoohi (1987)
and most other European studies.
In this study, we define two categories of outputs as total
investments and total loans.
1. Q1, total loans: In our models, total loans comprise:
Q1 = (discounts) + (bills purchased net) + (overdrafts) +
(short-term loan)+ (middle-long term loan) + (other loans) (reserve
for loan loss)
2. Q2, total investments: Total investments include:
Q2 = investments in government bonds and securities+ other
investments allowance for unrealized loss
Definitions of other variables
1. Total cost (TC): Total cost as the dependent variable
comprises interest expense, laborexpense and capital expense. Their
relationship can be explained as follows:
T C = p1 R + p2 L + p3 K= (interest cost) + (ive selling and
administrative expenses)
where
p1 R : interest expensep2 L : labor expensep3 K : capital
expense
2. ROE: The value of ROE is the ratio of net before-tax income
to equity.3. CONC: We choose to measure the degree of concentration
in the Taiwanese banking
industry by using banking deposits and the Herfindahl Index.4.
MS: MS is defined as the banks share of deposits market.5. Relative
cost efficiency: We define that Relative cost efficiency is
comprised by X-
efficiency and allocation efficiency.In this study, we apply the
distribution-free method to estimate Relative cost effi-ciency. It
is the ratio of the smallest n-year average multiplicative cost
function residualof banks to the banks n-year average residual
(current years data excluded).
6. SEFFE, scale economy efficiency: We will use the value
obtained from the estimationof OES. If a bank locates on the left
hand side of the bottom of the average cost (AC)curve for the whole
banking industry at that year, SEFFE will equal the value of
OES;equal one otherwise.
7. SEFFD: On the other hand, if a bank locates on the right hand
side of the bottom ofthe average cost (AC) curve for the whole
banking industry at that year, SEFFD willequal the value of overall
diseconomies of scale; equal one otherwise.
-
408 YU AND LUU
8. MGTH, market growth: MGTH is estimated by the real growth of
the Taiwanese depositmarket.
10. sdroa, indicator of the portfolio risk: The portfolio risk
is defined as the stand error ofthe ROA (return of asset). For
example, sdroa for the kth period is obtained from thestandard
error of ROA for k, k 1, and k 2 period.
Appendix 2: The constitution of sample
Table A.2.1. The constitution of sample
Group Bank Name
(1) Government-owned banks A Chiao Tung Bank Co., Ltd.B The
Farmers Bank of ChinaC Bank of TaiwanD TAIPEIBANK CO., LTD.E Land
Bank of TaiwanF Taiwan Cooperative BankG First Commercial BankH Hua
Nan Commercial Bank, Ltd.I Chang Hwa Commercial Bank, Ltd.N Taiwan
Business BankYQ Chinatrust Commercial Bank
(2) Local banks O Taipei Business BankP Taichung Business BankQ
Hsinchu BankR Tainan Business BankS Kaoshang Business BankT Hwalain
Business BankU Taidon Business Bank
(3) Old-private banks J The International Commercial Bank of
ChinaK United World Chinese Commercial BankL The Shanghai
Commercial & Savings Bank., Ltd.M Overseas Chinese Commercial
Banking Corporation
(4) New-private banks YA Grand Commercial BankYB Dah An
Commercial BankYC Union Bank of TaiwanYD The Chinese BankYE Far
Eastern International BankYF Asia Pacific BankYG Bank SinoPaoYH E.
Sun Commercial Bank., Ltd.YI Cosmos Bank, TaiwanYJ Pan Asia BankYK
Chung Shing Commercial BankYL Taishin International BankYM Fubon
Commercial BankYN Ta Chong Bank Ltd.YO BaoDao Commercial Bank
Ltd.YP Entie Pacific Bank
-
BANKING MERGERS 409
Appendix 3: The cost efficiency analyses of different groups
Table A.3.1. Relative cost efficiency for government-
ownedspecialised banks (GOSB)
19851991 19931997
Bank A 0.7039 0.7579Bank B 0.7404 0.6928Bank E 0.7807
0.7815Average 0.7416
(0.0384)0.7441(0.0459)
Table A.3.2. Relative cost efficiency for
Government-ownedcommercial banks (GOCB)
19851991 19931997
Bank C 0.8378 0.7586Bank D 1 0.7624Bank F 0.6307 0.6286Bank G
0.8069 0.8724Bank H 0.7745 0.8400Bank I 0.8430 0.8418Bank N 0.8504
0.8759Average 0.8205
(0.1097)0.7971(0.0883)
Table A.3.3. Relative cost efficiency for local banks (LB)
19851991 19941997
Bank O 0.7803 0.7680Bank P 0.7833 0.7972Bank Q 0.7186 0.7405Bank
R 0.7507 0.7105Bank S 0.7336 0.7549Bank T 0.6612 0.8709Bank U
0.7999 0.8822Average 0.7468
(0.0476)0.7892(0.0653)
Table A.3.4. Relative cost efficiency for old privatebanks
(OPB)
19851991 19931997
Bank J 0.7549 0.8782Bank K 0.7807 0.8019Bank L 0.7983 0.7560Bank
M 0.7381 0.7530Average 0.7680
(0.0267)0.7983(0.0575)
-
410 YU AND LUU
Table A.3.5. Relative cost efficiency fornew-private banks
(NPB)
19931997
Bank YA 0.7083Bank YB 0.8448Bank YC 0.7138Bank YD 0.7768Bank YE
0.8542Bank YF 0.7556Bank YG 0.7375Bank YH 0.6696Bank YI 0.6127Bank
YJ 0.6591Bank YK 0.6468Bank YL 0.6708Bank YM 0.7239Bank YN
0.6671Bank YO 0.5955Bank YP 0.9970Bank YQ 0.5661Average 0.7176
(0.1069)
Appendix 4: Describe statistics of all variables in the modelof
profit-structure relationship
Table A.4.1. Describe statistics of all variables from
19861991
Relative CostEfficiency CONC MS ROA ROE MGTH
19861991 Data StatisticsMean 0.7747 0.0992 0.0476 0.0099 0.2046
0.2102Median 0.7704 0.0962 0.0192 0.0087 0.1951 0.2204Maximum
1.0000 0.1117 0.1886 0.0536 0.6427 0.2635Minimum 0.5990 0.0910
0.0006 0.007543 0.0586 0.0859Std. Dev. 0.0794 0.0082 0.0511 0.0077
0.1199 0.0573Skewness 0.6618 0.4695 1.0491 2.4364 0.8150
1.4319Kurtosis 4.3845 1.5248 2.8132 12.5403 4.4761
3.7444Jarque-Bera 22.4726 16.0553 23.2943 602.4961 25.3878
45.9684Probability 0.0000 0.0003 0.0000 0.0000 0.0000
0.0000Observations 126 126 126 126 126 126
Source: Calculated from the data which are collected by the
Central Bank of China (CBC).
-
BANKING MERGERS 411
Table A.4.2. Describe statistics of all variables from
19931997
Relative CostEfficiency CONC MS ROA ROE MGTH
19931997 Data StatisticsMean 0.7560 0.0616 0.0263 0.0082 0.1042
0.1171Median 0.7515 0.0600 0.0122 0.0081 0.0904 0.1086Maximum
1.0000 0.0710 0.1538 0.0161 0.2963 0.1494Minimum 0.4673 0.0534
0.0023 0.0083 0.1363 0.0935Std. Dev. 0.1100 0.0059 0.0306 0.0035
0.0663 0.0207Skewness25 0.1754 0.2598 1.8893 0.9410 0.2919
0.4536Kurtosis26 2.7971 2.0277 6.1676 6.8319 3.5418
1.6676Jarque-Bera27 1.2997 9.6215 192.4728 144.2873 5.0222
20.5702Probability28 0.5221 0.0081 0.0000 0.0000 0.08112
0.0000Observations 190 190 190 190 190 190
Source: Calculated from the data which are collected by the
Central Bank of China (CBC).
Notes
1. Measure adopted by the Taiwanese government include:
deregulation of interest rates and foreign exchangerates
restrictions, liberalization of establishment of new banks and
foreign entry, enlargement of the businessscope of financial
institutions, and internationalisation of financial market
operations.
2. In the past, stock brokering and investment banking had
largely been domain of Taiwans two hundred andsix local stock
brokering firms.
3. The same as Berger (1995), we apply the distribution-free
method on the same banking data rather thanimposing predetermined
distributions on the X-efficiencies and random error.
4. Although accounting ratios in banking are typically used to
obtain a partial measure of banking productivity,these measures are
problematic. For example, if we want to increase labor productivity
by replacing peoplewith machines, then labor productivity will
rise, but actually the cost of the machines is not included in
themeasure.
5. Prior studies did not measure the total cost of banking
operations. For example, demand deposits are separatedfrom
commercial loans.
6. For example, if we want to increase labor productivity by
replacing people with machines, then labor produc-tivity will rise,
but actually the cost of the machines is not included in the
measure.
7. The conventional translog cost function system was developed
by Christen, Jorgenson and Lau (1973), as asecond-order Taylor
expansion series in output quantities, input prices and control
variables.
8. In the study of Mitchell and Onvural (1996), total assets of
their sample banks range from $0.5 billion to $100billion.
9. This limitation can be avoided by using the hybrid translog
cost function proposed by Caves, Christensenand Tretheway (1980).
Subsequently, Kolari and Zardkoohi (1987), Mester (1990),
Rodriguez, Alvarez andGomez (1993), and the others have employed
the hybrid translog cost function.
10. The cost-minimising input vector is just given by the vector
of derivatives of the cost function with respect tothe prices.
11. Zellners (1962) iterative SURE technique will only be
practicable by dropping one of the cost share equations.12. People
sometimes have the tendency to use the terms X-efficiency and
technical efficiency interchangeably.
However, there are important distinctions in the economic
theories underlying X-efficiency and technicalefficiency
(Leibenstein, 1980, p.27). Leibenstein also suggests that
X-inefficiency was often much larger thanallocative
inefficiency.
13. The concept of T.E (Technical efficiency) suggests that the
problem is a technical one and has to do with thetechniques of an
input called management. Under X-efficiency, the basic problem is
viewed as one that is
-
412 YU AND LUU
intrinsic to the nature of human organization, both rganization
whithin the firm and organization outside ofthe firm (Leibenstein,
1977, p. 312).
14. Deyoung (1997) demonstrate the diagnostic test in DFA cost
efficiency model that uses eleven years datafrom U.S. commercial
banks. His results suggest that 6 years of data is adequate to be
reasonably sure thatestimated X-efficiency contains only small
amounts of random error and that using 8 or more years of datamay
violate the central DFA assumption that banklevel inefficiency
remains constant over time.
15. Comparing with the assumptions for a single equation model,
this specific error components model only relaxthe assumption that
the covariance of residuals between equations is zero.
16. Accounting data were available from important businesses of
Taiwanese financial institutions from 1985 to1998 made by the
Central Bank of China.
17. The data about branch number for Taiwanese domestic banks
was gathered from Financial StatisticsMonthly Taiwan District the
Republic of China from 1985 to 1998 published by Economic Research
De-partment, the Central Bank of China.
18. Personnel expense was obtained from Financial Statistics
Abstract from 1994 to 1998 published byBureau of Monetary
Affairs.
19. Suitable construction cost index in Taiwan area was not
available because before 1991. Based on thisreason, we choose
general index of consumer price in Taiwan area to replace
construction cost index.General index of consumer price in Taiwan
area was available from Commodity-Price Statistics Monthlyin Taiwan
Area of the Republic of China published by Directorate-General of
Budget, Accounting andStatistics, Executive Yuan, Republic of
China.
20. They are Chiao Tung Bank, The Farmers Bank of China, Taiwan
Cooperative Bank, First Commercial Bank,Hua Nan Commercial Bank and
Chang Hwa Commercial Bank.
21. Cf. Bureau of Monetary Affairs (2000)22. In analyzing the
market share, foreign banks accounted for 3.4% of total bank
deposits, 3.65% of total
bank loans, 30.09% of total foreign exchange trading , 28.57% of
total guarantee business, and 89% of totalderivatives trading at
the end of June 1998. See Bureau of Monetary Affairs (2002).
23. Total deposits include: (1) due to the Central Bank of China
and other banks (2) checking deposits (3) demanddeposits (4) time
deposits (5) saving deposits (6) foreign deposits (7) (government
deposits)
24. According the income statements form the CBC, we just can
obtain the big item, the selling and admistrativeexpense, which
include two specific items, labor expense and capital expense.
25. Skewness is a measure of asymmetry of the distribution of
the series around its mean. Positive skewnessmeans that the
distribution has a long right tail and negative skewness implies
that the distribution has a longleft tail.
26. Kurtosis measures the peakedness or flatness of the
distribution of the series. The kurtosis of the normaldistribution
is 3. If the kurtosis exceeds 3, the distribution is peaked
relative to the normal; if the kurtosis isless than 3, the
distribution is flat relative to the normal.
27. Jarque-Bera is a test statistic for testing whether the
series is normally distributed. Under the null hypothesisof a
normal distribution, the Jarque-Bera statistic is distributed as x2
with 2 degrees of freedom.
28. The reported probability is the probability that a
Jarque-Bera statistic exceeds the observed value under thenullsmall
probability value leads to the rejection of the null hypothesis of
a normal distribution.
References
Altunbas, Y., E. P. M. Gardener, P. Molyneux and B. Moore,
Efficiency in European Banking. European EconomicReview 45,
19311955 (2001).
Avery, R. B., Error Components and Seemingly Unrelated
Regressions. Econometrica 45(1), 199209 (1977).Bauer, P., Recent
Developments in the Econometric Estimations of Frontiers. Journal
of Econometrics 46, 3956
(1990).Benston, G. J., A. N. Berger, G. Hanweck and D. Humphrey,
Operating Costs in Commercial Banking. Federal
Reserve Bank of Atlanta Economic Review 621 (1982).Berndt, E.
R., B. E. Hall, and J. A. Hausman, Estimation and Inference in
Nonlinear Structural Models. Annals
of Economic and Social Measurement 3, 653665 (1974).
-
BANKING MERGERS 413
Berger, A. N., Distribution-Free Estimates of Efficiency in the
US Banking Industry and Tests of the StandardsDistributional
Assumptions. Journal of Productivity Analysis 4, 261292 (1993).
Berger, A. N., The Profit-Structure Relationship in BankingTests
of Market-Power and Efficient-StructureHypotheses. Journal of
Money, Credit and Banking 27(2), 404431 (1995).
Berger, A. N., G. A. Hanweck and D. B. Humphrey Competitive
Viability in Banking: Scale, Scope and ProductMix Economies.
Journal of Monetary Economies 20, 501520 (1987).
Berger, A. N., W. C. Hunter and S. G. Timme, The Efficiency of
Financial Institutions: A Review and Previewof Research Past,
Present and Future. Journal of Banking and Finance 17, 221249
(1993).
Box, G. and D. Cox, An Analysis of Transformation. Journal of
Royal Statistical Society, Series B, 211264(1964).
Bureau of Monetary Affairs, The Modernization of ROC Financial
System, Taipei: Taiwan Ministry of Finance,2000.
Bureau of Monetary Affairs, Annual Report 2001, Taipei: Taiwan
Ministry of Finance, 2002.Carbo, S., E. P. M. Gardener and J.
Williams, Efficiency in Banking: Empirical Evidence from the
Savings Banks
Sector. The Manchester School 70, 204228 (2002).Chen, T. Y. and
T. L. Yeh, Technical and Scale Efficiency in Taiwans Banks: A
Comparative Study. Asia Pacific
Journal of Finance 2, 191205 (1999).Christensen, L. R., D. E.
Jorgenson and L. J. Lau, Transcendental Logarithmic Production
Frontiers. Review of
Economics and Statistics 55, 2845 (1973).Deyoung, Robert, A
Diagnostic Test for the Distribution-Free Efficiency Estimator: An
Example Using U.S.
Commercial Bank Data. European Journal of Operational Research
98, 243249 (1997).Greene, W. H., Econometric Analysis, 3rd edition,
New Jersey: Prentice Hall, 1997.Humphrey, D. B., Cost and Scale
Economies in Bank Intermediation. In R. Aspinwall and R. Eisenbeis
(Eds.),
Handbook for Banking Strategy, New York: John Wiley, 1985, pp.
745785.Kolari, J. W. and A. Zardkoohi, Bank Costs, Structure and
Performance, Lexington, MA: D.C. Heath and Company,
1987.Lambson, Val E. Is the Concentration-Profit Correlation
Partly an Artifact of Lumpy Technology? American
Economic Review 77, September 73133 (1987).Leibenstein, H.,
Allocative Efficiency vs. X-Efficiency. American Economic Review
56, 392415 (1966).Leibenstein, H., X-Efficiency, Technical
Efficiency, and Incomplete Information Use: A Comment. Economic
Development and Cultural Change 25, 311316 (1977).Leibenstein,
H., Inflation, Income Distribution and X-Efficiency Theory, London:
Croom Helm, 1980.McAllister, P. H. and D. A. McManus, Resolving the
Scale Efficiency Puzzle in Banking. Journal of Banking
and Finance 17, 389405 (1993).Mester, L. J., A multiproduct Cost
Study of Savings and Loans. Journal of Finance 42, 423445
(1987).Mitchell, K. and N. M. Onvural, Economies of Scale and Scope
at Large Commercial Banks: Evidence from the
Fourier Flexible Functional Form. Journal of Money, Credit and
Banking 28(2), 178199 (1996).Molyneux P., Y. Altunbas and E.
Gardener, Efficiency in European Banking New York: John Wiley &
Sons, Inc.,
1997.Noulas, A. G., S. M. Miller and S. C. Ray, Regularity
Conditions and Scope Estimates: The Case of Large-Sized
U.S. Banks. Journal of Financial Services Research 7, 235248
(1993).Peltzman, S., The Gains and Losses from Industrial
Concentration. Journal of Law and Economics 20, 229263
(1977).Punt, L. W. and M. C. J. van Rooij, The Profit-Structure
Relationship, Efficiency and Mergers in the European
Banking Industry: An Empirical Assessment. Research Memorandum
WO&E no. 604 Dutch Central Bank,1999.
Pulley, L. and D. Humphrey, The Role of Fixed Costs and Cost
Complementarities in Determining ScopeEconomies and the Cost of
Narrow Banking Proposals. Journal of Business 66, 437462
(1993).
Shepherd, W. G., Economies of scale and Monopoly Profits. In J.
V. Craven (Ed.) Industrial Organization,Antitrust, and Public
Policy, Boston: Kluwer Nijhoff, 1982.