118 PRAGUE ECONOMIC PAPERS, 2, 2010

BANK EFFICIENCY AND NON-PERFORMING LOANS:

EVIDENCE FROM MALAYSIA AND SINGAPORE
Mohd Zaini Abd Karim, Sok-Gee Chan, Sallahudin Hassan*
Abstract:
The objective of this paper is to investigate the relationship between non-performing loans and
bank efciency in Malaysia and Singapore. To achieve the objective, cost efciency was estimated
using the stochastic cost frontier approach assuming normal-gamma efciency distribution
model proposed by Greene (1990). The cost efciency scores were then used in the second
stage Tobit simultaneous equation regression to determine the effect of non-performing loans
on bank efciency. The results indicate that there is no signicant difference in cost efciency
between banks in Singapore and Malaysia although banks in Singapore exhibit a higher average
cost efciency score. The Tobit simultaneous equation regression results clearly indicate that
higher non-performing loan reduces cost efciency. Likewise, lower cost efciency increases
non-performing loans. The result also support the hypothesis of bad management proposed
by Berger and DeYoung (1992) that poor management in the banking institutions results in bad
quality loans, and therefore, escalates the level of non-performing loans.
Keywords: bank efciency, problem loans, stochastic cost frontier, tobit regression
JEL Classication: G21
1. Introduction
The banking sector is still the primary form of nancial intermediation in the Asian
and Pacic region, and as such is the largest conduit for the mobilization of domestic
savings, the main source of external capital to rms and the key player in the payment
system. Consequently, the development of an efcient banking sector is crucial for the
growth of the economies in the region. As liberalization of the banking sector around
the world continues apace, banking markets in different countries are now becoming
increasingly integrated. Moreover, foreign banks are allowed to set up branches in
other countries, subject to the regulations of the home country. Furthermore, banking
competition in the ASEAN region is expected to be more intense in the future with
* Mohd Zaini Abd Karim, College of Art and Sciences, Universiti Utara Malaysia (zaini500@uum.
edu.my); Sok-Gee Chan, Institute of China Studies, University of Malaya ([email protected]);
Sallahudin Hassan, College of Art and Sciences, Universiti Utara Malaysia ([email protected])
PRAGUE ECONOMIC PAPERS, 2, 2010 119
the realization of ASEAN Free Trade Area (AFTA).
1
In Malaysia and Singapore, both
governments have been promoting consolidation of domestic banks to prepare them
for the anticipated arrival of new foreign competition.
As foreign banks take a greater role in the market, the strength of the local banks
will be put to the test. How these increased competitive pressures will affect banks
depend in part on their ability to adapt and operate efciently in the new environment.
Banks that fail to do so will be driven out of the market by the more efcient ones. That
is, the most efcient banks will have a competitive advantage. Therefore, information
on banking efciency when compared across nations is important, as this will enable
policy makers to formulate appropriate and sound policies to direct their banking
industry. Despite these facts, cross-country comparisons of banking efciency in
developing countries are lacking in the literature.
2

At the same time, due to the Asian nancial crisis, growth and innovation in both
Malaysia and Singapore are constrained by banks whose capital has been eroded
by the accumulation of non-performing loans (NPL). Although the non-performing
loan ratios in Malaysia have fallen recently, the reduction in non-performing loan
ratios were largely brought about by the transfer of NPLs from banks to public asset
management companies. The question here is how non-performing loans affect the
cost efciency of banks. It is argued that it will have a detrimental effect since such
banks will exert additional managerial effort and give additional expense dealing
with these problem loans. These extra operating costs include, but are not limited to,
a) additional monitoring of the delinquent borrowers and the value of their collateral,
b) the expense of analyzing and negotiating possible workout arrangements, c) the cost
of seizing, maintaining, and eventually disposing of collateral if default later occurs, and
d) the diversion of senior management attention away from solving other operational
problems. Faced with an exogenous increase in non-performing loans, even the most
cost-efcient banks have to purchase the additional inputs necessary to administer
these problem credits. By estimating the relationship between non-performing loans
and bank efciency, we can determine whether an increase in problem loans have
a negative impact on bank efciency.
The rest of the paper is organized as follows. Section 2 reviews the literature.
Section 3 provides an overview of problem loans in Malaysia and Singapore. The
hypothesized relationship between non-performing loans and cost efciency is
discussed in Section 4. Section 5 discusses the methodology and the data used. Section
6 presents the empirical results. Finally, Section 7 concludes.
1 ASEAN member countries are working towards the elimination of import duties on all products to
achieve the eventual objective of a free trade area. The AFTA Council agreed that the target dates
to achieve these objectives will be the year 2015 for the six original member countries (Brunei,
Indonesia, Malaysia, the Philippines, Singapore, and Thailand) and the year 2018 for the new
members of ASEAN.
2 To our knowledge, the only study on cross-country comparisons of banking efciency in developing
countries is Karim (2001). However; there are quite a number of bank efciency studies on
individual developing countries. Among others Bhattacharya, Lovell, and Sahay (1997) on Indias
commercial banks, Chaffai (1997) on Tunisias banking industry, and Oral and Yolalan (1990) on the
Turkish banking industry.
120 PRAGUE ECONOMIC PAPERS, 2, 2010
2. Review of Literature
Issues of non-performing loans and cost efciency are related in several important ways.
First, a number of researchers have found that failing banks tend to be located far from
the best practice frontier
3
(e.g. Berger and Humphrey, 1992; Wheelock and Wilson,
1994). Thus, in addition to having high ratios of problem loans, banks approaching
failure also tend to have lower cost efciency. A number of other studies have found
negative relationships between efciency and problem loans even among banks that
do not fail (Kwan and Eisenbeis, 1995). A positive relationship between asset quality
and cost efciency (DeYoung, 1997) suggests that the negative relationship between
problem loans and cost efciency holds for the population of banks as well as for the
subset of failing banks.
Tsai and Huang (1999), by utilizing a translog cost function, examined the
relationship between management quality and cost efciency within Taiwans banking
industry. They discovered that asset quality and cost efciency are related; the
non-value-added activities of bad assets incur a negative consequence on the operating
performance.
In recent years, studies on bank efciency have taken into account asset quality,
specically non-performing loans. The omission of such a variable might lead to an
erroneous bank efciency measure (Mester, 1996). This is particularly true since
a large proportion of non-performing loans may signal that banks use fewer resources
than usual in their credit evaluation and loans monitoring process. In addition,
non-performing loans lead to inefciency in the banking sector as found by Altunbas
et al. (2000), Fan and Shaffer (2004) and Girardone et al. (2004). This is because
efcient banks are better at managing their credit risk as highlighted by Berger and
DeYoung (1997).
By taking into account risk and quality factors into the estimation of banks cost
efciency in the Japanese commercial banks for the period 1993 to 1996, Altunbas
et al. (2000) nds that the level of non-performing loans are positively related to bank
inefciency. Furthermore, banks tend to experience a decrease in their scale efciency
level after controlling for risk factors. This result is also consistent with the study of
bank efciency levels in the U.S. by Hughes and Mester (1993) and on the evaluation
of cost efciency in the Italian banks by Girardone et al. (2004).
On the other hand, Fan and Shaffer (2004) analyzed prot efciency of large
commercial banks in the U.S. by accounting for non-performing loans. They nd that,
although non-performing loans are negatively related to banks prot efciency, it is
not statistically signicant.
3. Overview of Problem Loans in Malaysia and Singapore
Non-performing loans have been a hindrance to economic stability and growth of
economies. In Malaysia and Singapore, non-performing loans continued to improve,
3 Berger and Humphrey (1992) dened that the best practice frontier is where the rms are considered
to be fully efcient or achieve 100% efciency level. The best practice frontier unit is achieved
when there is no other decision making unit that can alter the combination of its inputs for a given
output levels or vice versa.
PRAGUE ECONOMIC PAPERS, 2, 2010 121
underpinned by higher reclassication of non-performing loans to performing
status and recoveries, as well as efforts to achieve healthier balance sheets via loan
write-offs.
4
As a result, net non-performing loan ratios in the Malaysian banking
system, since the Asian nancial crisis, has gradually been in decline from a high of
13.6% (3-month classication) in December 1998 to 3.2% in 2007 (Table 1).
Table 1
Three-month Non-performing Loans in the Malaysian Banking System
Year
Gross non-performing
loans (RMbillion)
As a percentage of
total advances
1998 76,953 13.6
1999 65,540 11.0
2000 64,256 9.7
2001 76,976 11.5
2002 71,693 10.2
2003 65,774 8.9
2004 60,380 7.5
2005 53,570 5.8
2006 50,391 4.8
2007 41,763 3.2
Source: Bank Negara Malaysia (BNM) monthly statistical bulletin.
The bulk of the banking systems non-performing loans is in the business sector
(56% of total non-performing loans) followed by the household sector (41%). In terms
of banking system loans, the household sector accounts for the majority of total loans
at 55%, while the business sector accounts for 40%. In Singapore, non-performing
loans peaked for most banks in 1999 following the 1997 Asian nancial crisis. The
non-performing loan ratio for all banks (including multi-nationals with branches
in Singapore) was 4.3% in 2005, down from 8.3% in 2001 (Monetary Authority of
Singapore, 2006). This gure is much lower than in Malaysia.
Table 2 presents the loan disbursement and non-performing loans by sector in
Malaysia. The ratio of non-performing loans to the total loans disbursed by sector
clearly indicate that the education and health sector exhibit the highest ratio among
other sectors with reported ratios of 16.83% in year 2006 and 14.20% in year 2007.
This is followed by nance, insurance and business activities with 12.02% in year 2006
and 7.26% in year 2006. These sectors are mostly dominated by individual households
borrowing for educational, health, and investment purposes. This is consistent with
the BNM report; a large part of the non-performing loans comes from the household
sector.
4 To resolve the non-performing loans problem in the banking sector, the Malaysian government
established Danaharta, an asset management company, in 1998 to purchase non-performing loans
from banking institutions. As of 31 March 2001, Danaharta had acquired loans amounting to around
RM 48 billion from the nancial system.
122 PRAGUE ECONOMIC PAPERS, 2, 2010
Table 2
Loans Disbursed and Non-performing Loans by Sector in Malaysia
Sector
Loans disbursed
1
(in RM million)
Non-performing loans
2
(in RM million)
Ratio of
non-performing loans
to loans disbursed
2006 2007 2006 2007 2006 2007
Primary Agriculture 13080.70 13593.30 516.50 462.30 3.95 3.40
Mining and Quarrying 1327.30 1559.80 55.40 41.00 4.17 2.63
Manufacturing 65239.30 73743.30 6181.30 5992.90 9.47 8.13
Wholesale and Retail
Trade
52144.50 55890.70 3961.60 3528.40 7.60 6.31
Transport, Storage and
Communication
11128.80 14367.80 526.90 428.40 4.73 2.98
Finance, Insurance and
Business Activities
33142.80 38752.90 3984.30 2815.10 12.02 7.26
Education and Health 5923.00 6707.50 996.80 952.80 16.83 14.20
Household Sector 332264.50 357381.70 23596.80 19041.90 7.10 5.33
Construction 29654.30 30911.90 3104.10 2089.80 10.47 6.76
1
Loans disbursed by the banking system including Islamic banks.
2
Loans that were in default for three months. Malaysia had, until the end of December 1997, dened non-performing
loans as those loans up to six months overdue, and then revised the criterion to those loans up to three months
overdue on J anuary 1, 1998.
Source: BNM monthly statistical bulletin.
4. Theoretical Relationship between Non-performing Loans and Bank Ef ciency
From the point of view of management accounting, bank asset quality and operating
performance are positively related. If a banks asset quality is inadequate (e.g. the loan
amount becomes the amount to be collected), the bank will have to increase its bad debt
losses as well as spend more resources on the collection of non-performing loans. This
increase in non-performing loans in the banking industry can be due to external events,
such as adverse situation in economic activities (Berger and DeYoung, 1997, refers to
it as bad luck hypothesis).When banks list the loan amount for collection, banks will
incur extra operating costs from non-value-added activities to handle and supervise
the collection process. These non-value-added activities consist of constantly tracking
the debtors nancial status, being cautious of the collateral value, discussing the
amortization plan, paying expenses for contract negotiation, calculating the costs to
withhold, deposit and dispose of collateral at the time the loans become non-payable.
The costs include winning the trust from management and the public, preserving the
banks from being rated poor as a consequence of external affairs, declining deposits
because of a loss in credibility, and extra costs to monitor loan quality. Furthermore,
higher future costs are generated by the ignorance of the problems from other
operations when the loan quality issues grab the attention of the senior management.
This escalation in cost, in turn, deteriorates bank efciency.
PRAGUE ECONOMIC PAPERS, 2, 2010 123
On the other hand, Berger and DeYoung (1997) also suggest that efciency of
the banking rms might affect the non-performing loans in the banking industry. The
bad management hypothesis was developed to explain this relationship. Berger and
DeYoung (1997) argue that bad management of the banking rms will result in banks
inefciency and affects the process of granting loans. The banks management might
not thoroughly evaluate their customers credit application due to their poor evaluation
skills. In addition, the problem of asymmetric information between lenders and
borrowers further complicates the matter. Besides that, the management might not be
efcient in managing loan portfolios. Consequently, this leads to lower credit ratings for
the approved loans and high probability of default resulting in higher non-performing
loans. Therefore, banks inefciencies might lead to higher non-performing loans.
5. Methodology and Data
To compare the cost efciency of banks in Malaysia and Singapore, a cost efciency
score index is calculated by estimating a stochastic cost frontier function. We then
regressed these scores against non-performing loans and other control variables to
determine the relationship between non-performing loans and bank efciency.
Estimation of Cost Ef ciency
We used a cost function to measure efciency instead of the production function
because we want to determine how cost efcient a bank is as a nancial intermediary
in channeling funds from depositors to borrowers.
The SFA model proposed by Aigner et al. (1977), Meeusen and van de Broeck
(1977), and Battese and Corra (1977) is used. The SFA cost function consists of
variable costs, such as prices of the variable inputs, quantities of variable outputs and
any xed inputs or outputs needed for the production of goods and services in the
banks as well as environmental factors that inuence the costs of the banks and the
random error (Berger et al., 1997). The general cost function is given by Equation 1.
y
it
= + X'
it
+ Z'
it
+
it
where
it
= v
it
+
it
(1)
where y
it
is total cost in logarithm form of bank i at time t, X'
it
is the matrix of outputs
and input prices in logarithm form, Z'
it
is dened as bank-specic or environmental
variables for bank i at time t. Next, the random error is dened as
it
= v
it
+
it
where

it
is the unconditional mean given
it
and takes the value between 0 and 1.
The translog cost function is then obtained based on SFA. We assume that the
translog cost function denes the input use for all banks in the sample. The translog
function is the most frequently selected function to measure bank efciency (Greene
1980).
5
Although the use of the translog cost function reduces the degrees of freedom
in econometric estimation, this function is usually selected because it is a exible
5 The function was developed by Kmenta (1967) as a means of approximating the CES production
function and was introduced formally in a series of papers by Berndt, Christensen, Jorgensen and
Lau, including Berndt and Christensen (1973) and Christensen et al. (1973). The literature has
produced something of a competition in the development of exotic functional form. However, the
translog function has remained the most popular.
124 PRAGUE ECONOMIC PAPERS, 2, 2010
functional form that places no a priori restrictions on substitution possibilities among
the factors of production and hence allows both economies and diseconomies of scale
at different output levels.
(2)
In equation (2), ln C
it
is the natural logarithm of total cost (discussion of what
constitutes a banks inputs and outputs is in the next section); ln y
it
= the natural
logarithm of the jth output (j = 1,2,.,n); ln w
kit
is the natural logarithm of the kth
input price (k = 1,2,,m). ; t is the year of observation; and are the coefcients to
be estimated. The v
it
s are random variables associated with measurement errors in the
input variable or the effect of unspecied explanatory variables in the model, and the
u
it
s are non-negative random variables, associated with inefciency of inputs used,
given the levels of outputs and the quasi-xed inputs.
The cost efciency of inputs used for the i-th bank in the t-th year of observation,
given the values of the outputs and inputs, is dened as the ratio of the stochastic
frontier input use to observed input use. The stochastic frontier input use is dened by
the value of input use if the cost inefciency effect, u
it
, is zero (i.e., the bank is fully
efcient in the use of input). If a translog stochastic frontier cost function is used, the
cost efciency for rm i at time t is dened by equation (3),

( , ; ) exp( )
( , ; ) exp( )
it it it it
it
it it it it it
C y w v
CE
C y w v u
|
|
=
+
= exp(-u
it
) (3)
where CE
it
s 1. The reciprocal of this value, exp (u
it
), can be interpreted as a measure
of the cost inefciency of input usage.
The estimation of cost efciency employs the normal-gamma model proposed by
Greene (1990). This model is more practical since it corrects the problems in stochastic
frontier analysis as a result of the one-sided disturbances in the half-normal distribution
model. The half-normal distribution model proposed by Aigner, Lovell, and Schmidt
(1977) suffers from great inexibility, because it uses a single parameter distribution
with the assumption that the density of the disturbances mostly concentrates near zero.
Hence, any deviation in the dependent variable will be extremely damaging to the
analysis. In addition, a single outlier or unexpected observation from the sample can
result in over specication of the efciency scores irrespective of sample size used.
Greene (1990) argued that this model has an added advantage, since it does not
require the assumption that the rm-specic inefciency measures be predominantly
near zero. Furthermore, since the range of random variable is no longer restricted, the
distribution of inefciency can take different shapes.
To determine the relationship between non-performing loans and bank efciency
we employ the Tobit model since the efciency scores is bounded between zero and
one. The Tobit model developed by Tobin (1958) is also known as truncated or censored
0
1 1
1
In ln ln ln
2
n m n m
it j jit k kit jl j l
j k j l
C y w y y | | | |
= =
= + + +

1 1
ln ln ln ln
2 2
m m n m
kp kit pit jk jit kit q it it
k p j k
w w y w t u v | | | + + + + +

PRAGUE ECONOMIC PAPERS, 2, 2010 125
regression models where expected errors are not equal zero. Hence, estimation with
Ordinary Least Squares (OLS) would lead to bias, since OLS assumes a normal
distribution of the error term. More formally, the standard Tobit model can be dened
as follows:
EFF
i
*
= X +
i
,
i
~ N(0,
2
) if 0 < EFF
i
*
< 1
EFF
i
= 0 if EFF
i
*
= 0 (4)
EFF
i
= 1 otherwise
where EFF
i
*
is the cost efciency scores from the stochastic cost frontier estimation,
represents a vector of parameters to be estimated, X is a vector of explanatory
variables, and
i
is a normally distributed error term. The explanatory variables for this
study are NPL, STATE, FOREIGN, ASSET, and AGE.
NPL is the ratio of non-performing loans to total loans. Variables to control for
other factors that affect bank efciency are as follows; STATE is a dummy variable
that takes a value of 1 if the bank is state-owned and 0 if the bank is privately-owned;
FOREIGN is the dummy variable that takes a value of 1 if the bank is foreign bank and
0 if it is local; ASSET is the natural logarithm of the value of total assets. It measures
the size effect of rm; AGE is the age of rm to control for bank experience. Banks
that were established earlier are expected to be more efcient. However, after the bank
reaches some age, the incremental effect of experience will be negligible.
It is believed that the relationship between non-performing loans and efciency are
bi-directional instead of unidirectional. Low efciency reects poor daily operations
and loan portfolio management practices. Poor management skills in terms of credit
scoring will lead to negative returns on a high proportion of loans and thus higher
non-performing loans. Hence, efciency also affects non-performing loans (Berger
and DeYoung, 1997). Therefore, a Tobit simultaneous equation regression model
(Equation 4 and 5) is used to estimate the relationship between non-performing loans
and efciency to control for the simultaneity effect.
0 1 1 2 3 4 it it it it it it
EFF NPL STATE FOREIGN ASSET AGE o o o o o o c = + + + + + +
(5)
0 1 1 2 3 4 it it it it it it
NPL EFF STATE FOREIGN ASSET AGE | o | | | | u = + + + + + + (6)
Choice of Banks Input and Output
In general, the literature on banking efciency has considered two approaches in
considering what constitute a banks output and costs.
6
For this study, the intermediation
approach is employed for two reasons. First, we are concerned with how cost
efcient the bank is as a nancial intermediary in channeling funds from depositors
to borrowers. Second, the number of accounts for each output category in the bank
data is unavailable. Hence, our banks total cost will include the sum of expenses on
6 Little agreement exists as to what a bank produces or how to measure output. In general, though,
two approaches are used to examine the banking industry; the production and intermediation
approach is discussed in Humphrey (1985).
126 PRAGUE ECONOMIC PAPERS, 2, 2010
wages and salaries, land, buildings, and equipment and interest on deposits, while the
outputs are dollar amounts of total loans, total deposits, and investments. All the output
variables take the value of USD million. The input price will include expenses on
wages and salaries per employee (unit price of labor), expenses on land, buildings, and
equipment per dollar of assets (unit price of physical capital), and expenses on interest
per dollar of deposits (unit price of nancial capital). The price of labor is computed
using the total personnel expenses to total assets, the price of physical capital takes the
ratio of other operating expenses to total xed asset, and the price of nancial capital
is computed by dividing the total interest expenses with total deposits.
Data
The data for estimating the cost frontier function for ASEAN will be drawn from IBCA
BANKSCOPE and the Banks Annual Report. Table 3 presents a summary statistics
of all the banks in the sample. All dollar amounts are in USD million. They are in real
2000 terms and have been converted using individual country GDP deators.
The average asset size of the banks in the whole sample (both local and foreign
banks) over the period 1995 to 2000 is USD 6,665.44 million. The average asset size
for local banks in Singapore is the biggest in our sample. Their average asset size
is USD 14,534.03 million with a maximum asset size USD 31,521.94 million. The
average asset size of Malaysias local banks is USD 4,161.80 million with a maximum
asset size of USD 29,608.89 million.
Table 3
Summary Statistics for Sample Banks

Assets
(USD
million)
Loans
(USD
million)
Deposits
(USD
million)
Invest-
ments
(USD
million)
Non-
performing
Loans
(USD
million)
Price of
Labor
Price of
Capital
Price of
Funds
Malaysia
Mean 4161.80 8059.72 10613.99 4422.02 764.73 90.39 35.83 540.66
Standard
Deviation
5331.31 10349.77 13271.11 6003.48 1093.33 114.39 50.45 712.75
Minimum 39.64 146.30 311.40 100.30 1.10 4.00 0.72 14.20
Maximum 29608.89 64509.60 76111.40 32419.70 6988.70 654.50 247.10 5656.60
Singapore
Mean 14534.03 8490.23 12444.31 5592.94 636.42 65.31 61.75 500.81
Standard
Deviation
13222.08 10335.04 16435.36 8518.82 916.37 85.89 76.43 653.02
Minimum 418.88 522.20 503.10 0.60 10.90 1.50 2.50 15.00
Maximum 31521.94 27740.70 46718.10 26820.10 3035.10 239.30 218.80 1984.90
PRAGUE ECONOMIC PAPERS, 2, 2010 127
Table 4
Maximum-Likelihood Estimates of the Translog Stochastic Frontier Cost Function
Coefcient Standard Error t-ratio P-value
Constant 1.4389 2.0439 0.7040 0.4814
Ln(r
1
) -0.1181 0.4523 -0.2610 0.7940
Ln(r
2
) 0.4330 0.2395 1.8080* 0.0706
Ln(r
3
) 2.3371 0.6769 3.4530*** 0.0006
Ln(y
1
) -0.3578 0.7443 -0.4810 0.6308
Ln(y
2
) 2.2266 1.7147 1.2990 0.1941
Ln(y
3
) -0.1705 2.3249 -0.0730 0.9416
Ln(r
1
)
2
0.0397 0.0434 0.9160 0.3599
Ln(r
2
)
2
0.0109 0.0112 0.9770 0.3288
Ln(r
3
)
2
0.4141 0.1016 4.0780*** 0.0000
Ln(y
1
)
2
0.0048 0.0276 0.1740 0.8618
Ln(y
2
)
2
0.1299 0.2461 0.5280 0.5974
Ln(y
3
)
2
-0.6017 0.5809 -1.0360 0.3004
Ln(r
1
)Ln(r
2
) 0.1565 0.0396 3.9490*** 0.0001
Ln(r
1
)Ln(r
3
) -0.1816 0.0809 -2.2450** 0.0248
Ln(r
2
)Ln(r
3
) -0.1085 0.0558 -1.9440* 0.0519
Ln(y
1
)Ln(y
2
) -0.3842 0.3187 -1.2050 0.2280
Ln(y
1
)Ln(y
3
) 0.4836 0.3361 1.4390 0.1502
Ln(y
2
)Ln(y
3
) 0.3574 0.8121 0.4400 0.6598
Ln(r
1
)Ln(y
1
) 0.1477 0.1159 1.2740 0.2027
Ln(r
1
)Ln(y
2
) 0.2077 0.3284 0.6320 0.5271
Ln(r
1
)Ln(y
3
) -0.2577 0.4073 -0.6330 0.5268
Ln(r
2
)Ln(y
1
) -0.0526 0.0411 -1.2810 0.2002
Ln(r
2
)Ln(y
2
) -0.1680 0.1325 -1.2680 0.2048
Ln(r
2
)Ln(y
3
) 0.2106 0.1496 1.4070 0.1593
Ln(r
3
)Ln(y
1
) -0.1829 0.1819 -1.0050 0.3147
Ln(r
3
)Ln(y
2
) 0.7707 0.2984 2.5830*** 0.0098
Ln(r
3
)Ln(y
3
) -0.6101 0.4391 -1.3890 0.1647
Variance parameters for compound error
Theta 5.2758 1.5108 3.4920*** 0.0005
P 0.5308 0.0743 7.1470*** 0.0000
Sigmav 0.1535 0.0188 8.1470*** 0.0000
Log likelihood function 39.7503
Variances:
Sigma-squared (v)
Sigma-squared (u)
Sigma (v)
Sigma (u)
0.0220
0.0196
0.1485
0.1399
Notes:
*
signicant at 10% level,
**
signicant at 5% level.
***
signicant at 1% level.
128 PRAGUE ECONOMIC PAPERS, 2, 2010
The price of labor and price of funds in Malaysia is relatively higher as compared to
Singapore. Besides that, Singapore banks seems to be relatively superior in their output
production, since the total loans and deposits generated from the traditional banking
activities are relatively higher compared to the commercial banks in Malaysia. In addition,
the amounts of non-performing loans of the commercial banks in Singapore are relatively
lower compared to Malaysia, with the reported non-performing loans of USD 636.42
million whereas in Malaysia the reported gure of non-performing loans is USD 764.73.
6. Results and Discussion
This section reports the results obtained from estimating the stochastic cost frontier,
including parameter estimates and hypothesis testing. Following Karim (2001),
ordinary least squares (OLS) was rst used to check whether OLS assumptions were
not violated. The Box-Pierce statistics were insignicant indicating that the error terms
were non-autocorrelated, while the Breusch-Pagan statistics were also insignicant
indicating that the null hypothesis of homoskedasticity cannot be rejected. We then
proceed to estimate the stochastic cost frontier using the maximum likelihood method.
The maximum-likelihood estimates for the twenty-seven parameters in the translog
stochastic cost frontier for input use (dened by Equation 1) are presented in Table 4.
The inputs employed in this model are the price of labor (r
1
), price of capital (r
2
), and
the price of funds (r
3
), while the outputs used in this study are total investment (y
1
),
total loans (y
2
), and total deposits (y
3
).
From the estimation obtained based on standard translog specication, it is found
that the cost of the banks in the sample study are positively related to the price of capital
and the price of funds, and they are statistically signicant at 10% and 1% signicance
level respectively. This is consistent with the theory in which an increase in the price
of factors of production will lead to an increase in the cost of the commercial banks.
The descriptive statistics for the average efciency scores obtained from the
translog cost function for both local and foreign commercial banks in Singapore and
Malaysia are presented Table 5.
Table 5
Descriptive Statistics on Average Cost Efciency Scores for the Commercial Banks in
Singapore and Malaysia from Year 1995 to 2000
Mean Standard Deviation Minimum Maximum
Full Sample 0.8768 0.0809 0.5236 0.9654
Singapore
Full 0.9253 0.0275 0.8574 0.9654
Local 0.9277 0.0227 0.8785 0.9611
Foreign 0.9225 0.0329 0.8574 0.9654
Malaysia
Full 0.8688 0.0840 0.5236 0.9612
Local 0.8809 0.0677 0.5236 0.9612
Foreign 0.8492 0.1028 0.5418 0.9561
PRAGUE ECONOMIC PAPERS, 2, 2010 129
The average cost efciency score of the full sample is 87.68% which means that
the banks are wasting 12.32% of their inputs mix. Hence, banks may further reduce
their input mix by 12.32% in order to enhance their efciency level. From Table 5,
it is clearly indicated that the commercial banks in Singapore are relatively cost
efcient as compared to the commercial banks in Malaysia for both local and foreign
commercial banks. This may be due to the reason that Singapore, as a whole, is the
most economically efcient country in the region. Apart from that, a relatively more
efcient commercial banking industry in Singapore might also due to the openness of
the country in terms of foreign entrance to its banking sector as compared to Malaysia.
As highlighted by Karim (2001), tighter regulatory framework in one country may
also affect the efciency level of the commercial banks, and this is consistent with
the results obtained in this study. However, the difference in both mean and variance
efciency score are not statistically signicant (Table 6).

Table 6
Test for Variance and Mean Difference
Singapore Malaysia
Mean cost efciency 0.8768 0.8688
Variance cost efciency 0.0066 0.0071
F test: Two-sample for Variance
F-statistic 0.9278
P(F<=f) one-tail 0.3054
t-test: Two-Sample Assuming Equal Variances
t-statistic 0.9283
P(T<=t) two-tail 0.3539
Notes:
*
signicant at 10% level,
**
signicant at 5% level,
***
signicant at 1% level.
Hence, there is no sufcient evidence to conclude that there are differences in
the cost efciency level between commercial banks in Singapore and Malaysia even
though the commercial banks in Singapore exhibit a higher average cost efciency
score than the commercial banks in Malaysia.
To study the effect of non-performing loans on the commercial banks performance
and vice-versa, a Tobit simultaneous equation regression model (Equation 4 and 5)
is used to estimate the relationship between non-performing loans and efciency to
control for the simultaneity effect. The Tobit regression model was used in this study
as the cost efciency scores take on values between 0 and 1. The estimation results are
presented in Table 7.
130 PRAGUE ECONOMIC PAPERS, 2, 2010
Table 7
Tobit Simultaneous Equation Model Estimates
Dependent: EFF
Coefcient Standard Error t-ratio
Constant 0.598 0.225 2.659***
NPL -1.869 0.277 -6.741***
STATE 0.032 0.056 0.568
FOREIGN -0.026 0.066 -0.400
ASSET 0.057 0.034 1.665*
AGE 0.000 0.001 -0.009
Dependent: NPL
Coefcient Standard Error t-ratio
Constant 0.320 0.149 2.144**
EFF -0.535 0.079 -6.741***
STATE 0.017 0.031 0.559
FOREIGN -0.014 0.035 -0.403
ASSET 0.030 0.016 1.874*
AGE 0.000 0.000 -0.009
Variance parameters
s
12
/s
22
1.869 0.277 6.741***
s[e
1
,e
2
] 0.000 0.000 9.008***
Notes:
*
signicant at 10% level,
**
signicant at 5% level,
***
signicant at 1% level.
The results obtained from the Tobit simultaneous equation model show that the
coefcient of NPL in the equation where cost efciency is the dependent variable
is negative and is statistically signicant at the 1% level. This indicates that
non-performing loan have a negative effect on cost efciency. The result is consistent
with the studies by Altunbas et al. (2000), Fan and Shaffer (2004), and Girardone et
al. (2004). As pointed out by Berger and DeYoung (1997), the negative relationship
exists since banks will incur extra operating costs from non-value-added activities,
such as handling and supervising the collection process of the non-performing loans.
The efcient banks are better at managing their credit risk and hence it leads to lower
non-performing loans.
Likewise, the results show that the coefcient of EFF in the equation where NPL
is the dependent variable is negative and statistically signicant at the 1% level. The
result indicates that an increase in bank efciency decreases non-performing loans.
This supports the bad management hypothesis proposed by Berger and DeYoung
(1997) which suggests that poor management in banking institutions will result in
bad quality loans. The results also found that the banks total assets, which are used to
control for scale of operation, are positively related to cost efciency. This shows that
banks enjoy economies of scale, which is consistent with theory.
PRAGUE ECONOMIC PAPERS, 2, 2010 131
7. Conclusion
The objectives of this paper are twofold. First, the paper examines whether there are
signicant differences in banking efciency between Malaysian and Singaporean
banks. Second, the paper seeks to explain the relationship between non-performing
loans and bank efciency. Using actual data of both Malaysian and Singaporean banks
from 1995 to 2000, we estimate banks cost efciency by the stochastic cost frontier
method. The efciency scores are then used in the second stage Tobit analysis to
investigate the relationship between non-performing loans and banking efciency.
The cost efciency estimation results indicate an average cost efciency score
of 87.68% for the full sample. This suggests that banks are wasting 12.32% of their
inputs. The results also indicate that there is no signicant differences in the cost
efciency level between commercial banks in Singapore and Malaysia even though
the commercial banks in Singapore exhibit a higher average cost efciency score than
the commercial banks in Malaysia. The Tobit regression results clearly indicate that
higher non-performing loan reduces cost efciency. Likewise, lower cost efciency
increases non-performing loans. The results are consistent with the studies by Altunbas
et al. (2000), Fan and Shaffer (2004), and Girardone et al. (2004) that found that
non-performing loans lead to inefciency in the banking sector. In addition, the results
also support the hypothesis of bad management proposed by Berger and DeYoung
(1997), which suggests that poor management in the banking institutions results in bad
quality loans, and therefore, escalates the level of non-performing loans.
References
Aigner, D., Lovell, C., Schmidt, P. (1977). Formulation and Estimation of Stochastic Frontier
Production Function Models. Journal of Econometrics, 6, pp. 21-37.
Allen, L., Rai, A. (1996), Operational Efciency in Banking: An international Comparison. Journal of
Banking and Finance, Vol. 20, pp. 655-672.
Altunbas, Y., Liu, M. H., Molyneux, P., Seth, R. (2000). Efciency and Risk in J apanese Banking.
Journal of Banking and Finance, Vol. 24, pp.1605-1628.
Bank Negara Malaysia (BNM), Monthly Statistical Bulletin, various issues.
Battese, G. E., Corra, G. S. (1977). Estimation of a Production Frontier Model: With Application to
the Pastoral Zone of Eastern Australia. Australian Journal of Agricultural Economics, Vol. 21,
pp. 169-179.
Berger, A. N., DeYoung, R. (1997). Problem Loans and Cost Efciency in Commercial Banks.
Journal of Banking and Finance, Vol. 21, pp. 849-870.
Berger, A. N., Humphrey, D. B. (1992), Measurement and Efciency Issues Commercial Banking, in
Griliches, Z., ed., Output Measurement in the Services Sectors. National Bureau of Economic
Research Studies in Income and Wealth, pp. 245-279, Chicago: University of Chicago Press.
Berndt, E. R. and L. R. Christensen (1973), The Translog Function and the Substitution of Equipment,
Structures, and Labor in U.S. Manufacturing, 1929-1968. Journal of Econometrics, 1(1),
pp. 81-113.
Bhattacharya, A., Lovell, C. A. K., Sahay, P. (1997), The Impact of Liberalization on the Productive
Efciency of Indian Commercial Banks. European Journal of Operational Research, 98,
pp. 332-345.
132 PRAGUE ECONOMIC PAPERS, 2, 2010
Chaffai, M. (1997), Estimating Input-Specic Technical Inefciency: The Case of the Tunisian Banking
Industry. European Journal of Operational Research, 98, pp. 314-331.
Christensen, L. R., Jorgensen, D. W., Lau, L. J. (1973), Transcendental Logarithmic Production
Frontiers, The Review of Economics and Statistics, 55(1), pp. 28-45.
DeYoung, R. (1997), X-Efciency and Management Quality in National Banks. Journal of Financial
Services Research, Vol. 15, No. 2.
Fan, L., Shaffer, S. (2004). Efciency versus Risk in Large Domestic US Banks. Managerial Finance,
Vol. 30, pp. 1-19.
Farrell, M. J. (1957). The Measurement of Productive Efciency. Journal of the Royal Statistical
Society. Series a (General), 120, pp. 253-290.
Girardone, C., Molyneux, P., Gardener, E. P. (2004). Analysing the Determinants of Bank Efciency:
The Case of Italian Banks. Applied Economics, 36, pp. 215-227.
Greene, W. (1980), Maximum Likelihood Estimation of Econometric Frontier Functions. Journal of
Econometrics, 13, pp. 27-56.
Greene, W. H. (1990). A Gamma-distributed Stochastic Frontier Model. Journal of Econometrics, 46,
pp. 141-163.
Hughes, J. P., Mester, L. J. (1993). A Quality and Risk-adjusted Cost Function for Banks: Evidence on
the Too-Big-to-Fail Doctrine. Journal of Productivity Analysis, 4, pp. 293-315.
Humphrey, D. B. (1985), Costs and Scale Economies in Bank Intermediation, in Aspinwall, R. C.,
Eisenbeis, R., eds., Handbook of Banking Strategy. New York: Wiley, pp. 745-783.
Hymer, S. (1976). The International Operations of National Firms: a Study of Direct Foreign Investment.
Cambridge, Massachusetts: MIT Press.
Karim, M. Z. A. (2001), Comparative Bank Efciency Across Select ASEAN Countries. ASEAN
Economic Bulletin, 18 (3), pp. 289-304.
Kmenta, J. (1967), On Estimation of the CES Production Function. International Economic Review,
8, pp. 180-189.
Kwan, S., Eisenbeis, R. (1995) An Analysis of Inefciencies in Banking. Journal of Banking and
Finance, 19(3-4), pp. 733-734.
Meeusen, W., van den Broeck, J. (1977), Efciency Estimation from Cobb-Douglas Production
Functions with Composed Error. International Economic Review, 18, pp. 435-44.
Mester, L. J. (1996). A Study of Bank Efciency Taking into Account Risk-Preferences. Journal of
Banking and Finance, Vol. 20, pp. 1025-1045.
Mian, A. (2006). The Limits of Foreign Lending in Poor Countries. The Journal of Finance, Vol. 61,
pp. 1465-1505.
Monetary Authority of Singapore (2006). Annual Report 2006. Singapore: MAS
Oral, M., Yolalan, R. (1990), An Empirical Study on Measuring Operating Efciency and Protability of
Bank Branches. European Journal of Operational Research, 46, pp. 282-294.
Tsai, D. H., Huang, F. W. (1999), Management Quality and Bank Efciency: Empirical Evidence for
Taiwanese Banks. Management Review, 18 (3), pp. 35-55.
Wheelock, D. C., Wilson, P. W. (1995), Explaining Bank Failures: Deposit Insurance, Regulation, and
Efciency. Review of Economics and Statistics, 77, pp. 689-700.