International Journal of

Operations Research

International Journal of Operations Research Vo. 4, No. 2, 63-79 (2007)

Productive Performance Evaluation of the Banking Sector in India

Using Data Envelopment Analysis
Biresh K. Sahoo1, *, Jati K. Sengupta2, and Anandadeep Mandal3
1Amrita School of Business, Amrita Vishwa Vidyapeetham, Ettimadai Post, Coimbatore 641 105, India
2Department of Economics and Operations Research, University of California, Santa Barbara, CA 93106, USA
3Industrial Development Bank of India (IDBI), Kochi-35, India

Received April 2006; Revised October 2006; Accepted October 2006

Abstract¾This paper attempts to examine, using data envelopment analysis, the productivity performance trends of the
Indian commercial banks for the period: 1997-98 – 2004-05. Our broad empirical findings are indicative in many ways. First,
the increasing average annual trends in technical efficiency for all ownership groups indicate an affirmative gesture about the
effect of the reform process on the performance of the Indian banking sector. Second, the higher cost efficiency accrual of
private banks over nationalized banks indicate that nationalized banks, though old, do not reflect their learning experience in
their cost minimizing behavior due to X-inefficiency factors arising from government ownership. This finding also
highlights the possible stronger disciplining role played by the capital market indicating a strong link between market for
corporate control and efficiency of private enterprise assumed by property right hypothesis. And, finally, concerning the
scale elasticity behavior, the technology and market-based results differ significantly supporting the empirical distinction
between returns to scale and economies of scale, often used interchangeably in the literature.
Keywords¾Banks, Efficiency, Scale elasticity, Data envelopment analysis

1. INTRODUCTION the heightened competitive pressure in the banking

industry. These changes came in the form of greater use of
There is a widely held general belief that competition, a automatic teller machines and internet banking, huge
driving force behind numerous important policy changes, increase in housing and consumer credit, stronger and
exerts downward pressure on costs, reduces slacks, more transparent balance sheets and product
provides incentives for the efficient organization of diversification. A significant intent of these policies is to
production, and even drives innovation forward. However, have a radical transformation in the operating landscape of
the empirical evidence in its favor is mix. We in this paper the Indian banks. In this scenario we believe that banks in
analyze productivity performance of banking sector in India are in the pursuit of enlarging their size using
India, looking particularly at the impact of competition on available scale economies in order to enhance their asset
both the level and growth of productivity performance. base and profit so as to meet with the global standard.
Productivity performance is measured here through Also important is to examine the issues of efficiency
efficiency and scale economies, as these two are perceived and scale elasticity behavior of banking sector across the
to be the two most important key issues in the banking entire spectrum of ownership groups that might yield
literature. valuable information concerning productivity differentials
For a growing economy like India the faster growth of across ownership groups. This will enable us to verifying
the industrial sector critically depends on an efficient and the issue of economic linkage of ownership vis-à-vis
liberal market-oriented financial policies, and hence the performance in the light of property right hypothesis
Indian banking sector is selected for empirical illustration. (Alchian (1965) and de Alessi (1980)) and public choice
The Indian financial sector, which had been operating in a theory (Nickskamen (1971) and Levy (1987)). As per
closed and regulated environment, underwent a radical property right hypothesis, private enterprises should
change during the nineties. To induce efficiency and perform more efficiently and more profitably than public
competition into the system, Reserve Bank of India (RBI) enterprises, i.e., there is a strong link between the market
initiated in 1992 a number of reforms such as entry for corporate control and efficiency of private enterprise,
deregulation, branch delicensing, deregulation of interest which precisely holds for developed countries where capital
rates, and allowing public sector banks to raise up to 49% market functions well. However, in the absence of
of their equity in the capital market, which all gave rise to well-functioning capital markets in developing country, the

*
Corresponding author’s email: [email protected]
1813-713X Copyright © 2007 ORSTW
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 64
IJOR Vol. 4, No. 2, 63-79 (2007)

Indian banking industry could provide a test for which Bhattacharyya et al. (1997) had used in their study.
performance differential across the entire ownership However, the research efforts in these studies have failed
groups arising from studying efficiency and scale to throw light on one important issue, i.e., scale elasticity,
economies behavior so as to examine whether reforms and its linkage with cost efficiency. There is no such study,
process are working. Against this backdrop, it is of interest to our knowledge, that employed DEA in evaluating the
to examine the efficiency and scale elasticity behavior of productive performance in terms of technical efficiency,
commercial banks having a minimum level of retail cost efficiency and scale elasticity for the banking sector in
presence in India with respect to ownership in the light of India. This present study, using the panel data, utilizes the
financial sector reforms. different variants of DEA models to evaluate the same for
We have used the nonparametric approach called data the period: 1997-98 – 2004-05.
envelopment analysis (DEA) for the simple reason that it The rest of the paper proceeds as follows: Section 2 not
does not require specification of arbitrary functional forms, only discusses the current state of the Indian banking
and it has the natural advantage of eliminating the effects sector describing how it has altered in the recent past, but
of all productive and scale inefficiencies prior to also puts the main arguments highlighting especially the
calculating scale economies. Applications of DEA in question of efficiency. Section 3 deals with the DEA
banking sector in developed countries are enormous, but, methodology wherein the detailed computational
in developing countries, are modest1. Several authors2 have procedures for efficiency, returns to scale, and economies
compared the efficiency of the Indian banks with that of of scale are elaborated. The data concerning the selection
those in other countries. Previously though, Tyagarajan of inputs and outputs, and their sources are provided in
(1975), Subramanyam (1993) and Rangarajan and Section 4. Section 5 presents the results of the empirical
Mampillly (1972) have examined various issues relating to analysis, and Section 6 concludes.
the performance of banks in India, none of these studies
have examined the efficiency of banks’ service provision. 2. AN OVERVIEW OF INDIAN BANKING
Bhattacharyya et al. (1997) used DEA methodology to SYSTEM
study the consequences of liberalization on the
performance of the banking sector in India. This study The banking sector in India is broadly divided into two
considered 70 banks for the period 1986 to 1991. Hence, groups: commercial banks and co-operative banks. On the
no estimates were available after the reform initiation basis of ownership mold, commercial banks are grouped
period. Moreover, during that period the Indian private into three categories - state owned or public sector banks
sector banks were yet to establish themselves entirely (PSBs), private banks under Indian ownership and foreign
whereas the public sector banks were well recognized. banks. There are 27 PSBs, which all account for 80 per
Hence, it is expected that the public sector banks were cent of commercial banking asset.
outperforming the rest during that regime. At the top of the banking system is the Reserve Bank of
Rammohan and Ray (2004) used DEA technique to India (RBI). RBI is the central bank of the country
compare the revenue efficiencies of the public, private and entrusted with monetary stability, management of currency
foreign sector banks in India during the period 1992-2000. and supervision of financial as well as payments system.
Using operational data on inputs (deposits and operating Established in 1935, its functions and focus have evolved
cost) and outputs (loans, investments and other income), in response to the changing economic environment. Its
they found significant difference in performance between history is intrinsically interwoven with the economic and
the public and private sector, whereas between the public financial history of the country.
sector and foreign banks, the results were comparable. The financial structure in India in 1950s was
Sathey (2003) analyzed using DEA the efficiency of Indian convincingly liberal. It had restricted control on interest
banking sector in the year 1997-98 using both the financial rates and had low statutory preemption on funds. The
All–India Rural Survey Committee observed that out of
and operational models of the intermediation approach.
total borrowings of Rs.750 billion for cultivators in
Das et al. (2004) used the operational side of the
1951-52, both agriculturalist and professional money
intermediation approach to examine the effects of lenders accounted for 24.9% and 44.8%, respectively.
liberalization on the efficiency of banks for the period: Hence, it shows unequal distribution of bank credit, which
1997-98 – 2002-03. The problem with their approach is ascertains the lack of ability of the market to efficiently
that they considered constructing annual DEA frontiers to allocate resources. In retort, to ensure adequate flow of
analyze various efficiency changes over time where the credit into indisputably productive activities in
information on trends in performance would not be conformance with the planned priorities, the Government
available, since the benchmarks would likely change from tightened its control over the credit allocation process.
year to year. This problem would have been overcome, had
they used the panel data to construct a ‘grand frontier’,

1Leightner and Lovell (1998), Shyu (1998), Gilbert and Wilson (1998) and Hao et al. (2001) are the few authors who have carried out
studies related to the Asian banking sector.
2For example, see Bhattacharyya et al. (1997), Chatterjee (1997) and Saha and Ravishankar (2000).
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 65
IJOR Vol. 4, No. 2, 63-79 (2007)

Some further measures, viz., control of lending rates, and borrowing rates. In the competitive money and capital
raising liquidity requirements and establishment of a markets, the inability to offer competitive market rates adds
system of development banks for serving the assorted to the disadvantage of marketing and building new
segments of industry and agriculture, were also introduced. business. In the face of the deregulated banking industry,
This process ended up in nationalization of 14 largest an ideal competitive working will be reached when the
commercial banks in 1969 and subsequently six more in banks are able to earn adequate amount of non-interest
1980, and in 1993 it leads to the merger of two banks. income to cover their entire operating expenses. In this
Reforms in India in mid-1991 heralded a new era in the case, the spread factor i.e. the difference between the gross
economic history of our country. Certain reforms in the interest income and interest cost will constitute its
financial sector were put through even before the onset of operating profits. Theoretically, even if the bank keeps 0%
economic reforms in 1991. These measures were taken in spread, it will still break even in terms of operating profit
1985 based on the report of Sukhmoy Chakravarthi, an and not return on operating loss. The net profit is the
expert committee constituted by RBI. amount of the operating profit minus the amount of
This Committee initiated the process of financial sector provisions to be made including for taxation. On account
reforms in the country. This was followed by the report of of the burden of heavy NPA, many nationalized banks
the working group headed by Mr. N. Vaghul in 1985. This have little option, and they are unable to lower lending
is in the form of a follow-up report of the earlier Sukhmoy rates competitively, as a wider spread is necessitated to
Chakravarthi Committee Report. After the introduction of cover cost of NPA in the face of lower income from off
economic reforms in 1991, the recommendations of the balance sheet business yielding non-interest income.
Narasimhan Committee in 1997 provided the impetus for NPA affects the liquidity of nationalized banks even
further initiatives. A second report submitted by Mr. though they (except Indian Bank) are able to meet norms
Narasimhan in 1987 signaled the need for the 2nd phase of of capital adequacy. The fact that their net NPA on the
financial and banking sector reforms. average is as much as 7%, is a potential threat for them.
The main objectives of the financial sector reforms were RBI has indicated the ideal position as Zero percent Net
to widen, deepen and integrate the different segments of NPA. Even granting 3% net NPA within limits of
financial sector, viz., the money market, debt market and tolerance the nationalized banks are holding an
foreign exchange market. The Sukkmoy Chakravarthi uncomfortable burden at 7.1% as on March 2001. They
Committee made several recommendations for the have not been able to build additional capital needed for
development of money and government securities markets. business expansion through internal generations or by
The noteworthy developments in the financial system tapping the equity market, but have resorted to II-Tier
include, among others, restricting the call, notice, term capital in the debt market or looking forward towards
money market as a pure inter-bank market with additional capital to be granted by Government of India. Banks in
access only to primary dealers, phasing out non-bank the process of financial intermediation are confronted with
participants by granting permission to operate in repo various kinds of financial and non-financial risks. These
market, smoothening the maturity structure of debt, risks are highly interdependent, and events that affect one
improving the liquidity of government securities, phased area of risk can have ramifications for a range of other risk
reduction in SLR requirements from an effective 37.4 per categories. It therefore becomes very essential for the top
cent in March 1992 to a little over 28 per cent in March management to attach considerable importance to improve
1996, activating the Repo market by allowing repos/reverse the ability to identify, measure, monitor and control the
repos transactions in all government securities. overall level of risks undertaken.
Non performing assets (NPA) has not only affected the Decades before the reforms, the exercise of risk
profitability, liquidity and competitive functioning of PSBs, assessment and risk management were never seriously
but also the psychology of bankers in respect of their considered, as banks were operating in a captive economy.
disposition towards credit delivery and credit expansion. Since 1998 RBI has been giving serious attention towards
Between 01.04.93 to 31.03.2001 Commercial banks evolving suitable and comprehensive models for
incurred a total amount of Rs.31251 crores towards risk-management, and to integrate this new discipline in
provisioning NPA. This has brought Net NPA to 6.2% of the working systems of banks to enhance operational
net advances. To this extent the problem is contained, but efficiency. In this circumstance, efficiency assumes decisive
at a cost. This costly remedy is made at the sacrifice of importance aligned with setting of competitive viability
building healthy reserves for future capital adequacy. NPA and improved performance in future.
is not merely non-remunerative but also cost absorbing and
profit eroding. 3. METHODOLOGY
In the context of intense competition, the weak banks
are at disadvantage for leveraging the rate of interest in the We deal with n banks; each bank uses m inputs to
deregulated market and securing remunerative business produce s outputs. For each bank o (o = 1, 2, …, n), we
growth. This is the margin between cost of resources denote, respectively, the input and output vectors by xo Î
employed and return there from. When interest rates were Rm and yo Î Rs. The corresponding input/output matrices
directed by RBI, there was no option for banks. But, today, are defined by X = (x1, x2, …, xn) Î Rmxn and Y = (y1,
in the deregulated market, banks decide on their lending y2, …, yn) Î Rsxn. Let the input and output price vectors be,
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 66
IJOR Vol. 4, No. 2, 63-79 (2007)

respectively, wo Î Rm and po Î Rs., and the corresponding discussions.

input and output price matrices are defined, respectively, In the case of single input and output technology, rp is
by C = (w1, w2, …, wn) Î Rmxn and P = (p1, p2, …, pn) Î simply expressed as the ratio of marginal product (MP) to
Rsxn. average product (AP), i.e., r p ( x , y ) º MP AP =
The technology (T) is defined as the set of all feasible ( dy dx ) ( y x )
input-output vectors, i.e., T º {(x, y): x can produce y}. The
Färe et al. (1988) redefine y ( x , y ) in terms of Di(x, y)
standard neoclassical characterization of technology
involving multiple inputs and outputs is the transformation as y (x, y) = Di(x, y) - 1 = 0, which yields the following
function y ( x , y ), which exhibits the following properties: measure of SE (Färe et al. (1986)):

s
¶Di ( x , y )
y ( x , y ) = 0,
¶y ( x , y )
< 0 ( "r ) and
¶y ( x , y )
> 0 ( "i )
r p ( i ) ( x , y ) = Di ( x , y ) å ¶y r
yr (2)
¶y r ¶x i r =1

Alternatively, T can be described by its input set L(y) or Analogously, they define scale elasticity in terms of
output set P(x), defined, respectively, as L(y) º {x: (x, y) Î output distance function as
T} for all y, and P(x) º {y: (x, y) Î T} for all x.
T can also be represented by the input/output distance
functions (which are due to Shephard (1970)), defined,
m
¶Do ( x , y )
r p( o ) ( x , y ) = å x i Do ( x , y ) (3)
respectively, as Di(x, y) º sup {q: qx Î L(y), q > 0} and i =1 ¶x i
Do(x, y) º inf{d: y/d Î P(x), d > 0}.
(Local) RTS is increasing, constant and decreasing if r(x, y)
3.1 Efficiency > 1, r(x, y) = 1 and r(x, y) < 1 respectively (we omit
subscripts here to mean any of these three scale elasticity).
The input technical efficiency (TEi), defined as the ratio
of minimum input qx to actual input x, is nothing but 3.2.2 Scale elasticity in dual form
Di(x, y) itself. Similarly, the output technical efficiency (TEo),
defined as the ratio of actual output y to potential output Following Panzar and Willig (1977), the dual measure of
y/d, is nothing but Do(x, y) itself. The corresponding price SE in cost, rc(w, y) is defined as
measures of efficiency are, respectively, the cost and
revenue efficiency. Cost efficiency (CE) is defined as the s
¶C ( y ; w )
ratio of actual cost (w.x) to C(y, w) where C(y, w) = rc ( y ; w ) = C ( y ; w ) åy
r =1
r
¶y r
(4)
minx{w.x: x Î L(y)} is the minimum cost of producing y,
given the input price vector w. Revenue Efficiency (RE)
can be analogously defined as the ratio of actual revenue In the case of single output, rc can be expressed as the
(p.y) to R(x; p) where R(x; p) = maxy{p.y: y Î P(x)} is the ratio of average cost (AC) to marginal cost (MC).
maximum revenue of selling y, given the output price Economies of scale exist if rc > 1, diseconomies of scale
vector p. For various concepts of efficiency and their best exist if rc < 1, and there is neither economies nor
treatments in economics, see among others, Färe et al. diseconomies if rc = 1. However, the duality relationship
(1985), Sengupta (2000, 2003) and Ray (2004). between C(y, w) and Di(x, y) suggests that SE in both
production and cost environments are the same, i.e.,
3.2 Scale elasticity
s
¶C ( y ; w )
3.2.1 Scale elasticity in primal form rc ( y ; w ) = C ( y ; w ) åy
r =1
r
¶y r
(5)
The returns to scale (RTS)/scale elasticity (SE) in
s
¶Di ( x , y )
production is defined as the ratio of the maximum
= r p ( i ) ( x , y ) = Di ( x , y ) å
r =1 ¶yr
yr

proportional (b) expansion of outputs to a given

proportional (a) expansion of inputs. So, on differentiation Similarly, the duality relationship between R(x; p) and
of y (a x , b y ) = 0 with respect to scaling factor a, and Do(x, y) says that SE in both revenue and production
then equating it with zero yields the following local SE environment are the same, i.e.,
measure rp:
m
¶R( x ; p )
m
¶y s
¶y
rr ( x ; p ) = å x i R( x ; p ) = r p( o ) ( x , y )
r p (x , y ) º - å xi ¶x i
i =1 ¶xi
å
r =1
yr
¶y r
(1) i =1

m
¶Do ( x , y )
=å x i Do ( x , y ) (6)
i =1 ¶x i
See Hanoch (1970), Starrett (1977), Panzar and Willig
(1977) and Baumol et al. (1982) for the detailed
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 67
IJOR Vol. 4, No. 2, 63-79 (2007)

See Färe and Primont (1995) for proof of these two s

duality theorems. C ( w ; y ) = max å u r y ro + wo

r =1
m s
3.3 DEA models to evaluating efficiency3 and SE s .t . - å v i x ij + å u r y rj + wo £ 0, ( "j ), (11)
i =1 r =1
3.3.1 Scale elasticity in primal measure v i £ w i , ( "i ), u r , v i ³ 0, ( "r , i ), wo : free

The dual of the input-oriented Banker, Charnes and

to compute SE of bank ‘o’ in cost environment as follows:
Cooper (BCC) model (Banker et al, 1984), which is based
If the bank ‘o’ is efficient, then it holds that
on the assumption of variable returns to scale (VRS), is
used for obtaining SE in production of bank ‘o’ as follows: s
C ( y; w ) = åu
r =1
*
r yro + wo* (12)
m
Di ( x o , yo ) = max å u r y ro + u o
i =1
Then, using (4) SE of bank ‘o’ can be obtained as
s m
s .t . å ur yrj - å v i x ij + uo £ 0, ( "j ), (7)
s

åu
r =1 i =1 *
s rc ( y ; w ) = C ( y ; w ) r y ro = 1 éë1 - (wo* / C ( y ; w ))ùû (13)
åv x
r =1
i io = 1, u r , v i ³ 0, and u o : free r =1

Bank ‘o’ exhibits economies of scale/no economies/

If the bank ‘o’ is efficient, then it holds that diseconomies of scale when wo* is greater than/equal
to/less than zero, respectively. See Sueyoshi (1997, 1999)
s
Di ( x o , yo ) = å u r* y ro + u o* = 1 (8) for the case of multiple optimal solutions.
r =1 The following dual of the VRS [Revenue] DEA model:

m
On differentiation of Di(xo, yo) with respect to yro yields
R( x o ; po ) = min å v i x io + d o
i =1
¶Di ( x o , y o ) m s
= u r*
¶y ro
(9) s .t . å v x -å u
i =1
i ij
r =1
r y rj + d o ³ 0, ( "j ), (14)

u r ³ pr , ( "r ), u r , v i ³ 0, ( "r , i ), d o : free

Now, using (2) SE4 of bank ‘o’ in production, ri(p) can be
obtained as:
to compute SE of bank ‘o’ in revenue environment as
follows:
s
¶Di ( x o , y o )
r p ( i ) = Di ( x o , y o ) år =1 ¶y ro
y ro The objective function of (14) can be expressed as

s
(10) m
= Di ( x o , y o ) å u yro = 1 (1 - u )
*
r
*
o R( x o ; po ) = åv x
i =1
*
i io + d o* (15)
r =1

However, if bank ‘o’ is inefficient, then rp(i) equals Then, using (6) SE for bank ‘o’ in revenue environment can
( Di ( x o , y o )/( Di ( x o , y o ) - u o* )). RTS is increasing, be obtained as
constant and decreasing if u o* > 0, u o* = 0 and u o* < 0, m
respectively. However, when model (7) suffers from the rr ( x o ; po ) = å v i* x io R ( x o ; po ) =
problem of multiple optima, one can then compute lower i =1 (16)
and upper bounds of rp(i) to arrive at average SE. See ( R ( x o ; po ) - d o* ) R ( x o ; po ) = 1 - (d o* R ( x o ; po ))
Sueyoshi (1997, 1999) for the details.
3.4 Alternative measures of scale elasticity
3.3.2 Scale elasticity in dual measures: Cost and
revenue DEA models The above [Cost] and [Revenue] DEA models suffer
from two problems: 1) SE in production (rp(i)/rp(o)) does
The following dual of the VRS [Cost] DEA model: not differ from its dual counterpart, i.e., cost/revenue

3Due to lack of space, we are excluding here the discussion on standard DEA models, viz., BCC and Cost efficiency models for
technical and cost efficiency measurements. Refer to Cooper et al. (2000) for their detailed discussion.
4Several authors (Färe et al. (1988), Førsund (1996), Fukuyama (2001), Sueyoshi (1997, 1999)) have derived this same scale elasticity

formula in (10) in different ways. Tone and Sahoo (2004) have employed this approach to measure SE in production in the presence of
congestion.
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 68
IJOR Vol. 4, No. 2, 63-79 (2007)

elasticity (rc/rr), thus giving the illusion that RTS and programming (LP):
economies of scale are the one and the same5; and 2) this
cost/revenue model declares a cost/revenue inefficient q * = max q
DMU as efficient. n n
Note that in [Cost] DEA model where input prices are s .t . åyl j =1
rj j - s r+ = q y ro , ( "r ), åc
j =1
j l j + s - = co , (17)
held constant, the cost structure is based on its underlying
n
technology L(y) where increasing returns to scale implies
economies of scale. Similarly, in [Revenue] DEA model ål
j =1
j = 1, l j ³ 0
where output prices are held constant, the revenue
structure is entirely determined from its underlying
If the bank ‘o’ is efficient, then q * = 1, then the
technology P(x) where both production and revenue
elasticities are the same. However, since the input/output minimum cost ( c o* ) for producing output vector y o , is
market is typically imperfect, these two concepts can no obtained from the following LP:
longer be the same6. See Sahoo et al. (1999) and Tone and
n
Sahoo (2003) for the historical distinction between these
two concepts, and Tone and Sahoo (2005, 2006) for
c o* = min åc j =1
j lj
empirical measurements.. n n
(18)
Concerning the evaluation of CE, Tone (2002) points s .t . åyl rj j ³ yro , ( "r ), ål j = 1, l j ³ 0
out that if any firms, A and B have the same amount of j =1 j =1

inputs and outputs, i.e., xA = xB and yA = yB, and the unit

input price of firm A is twice that of firm B for each input, Let the optimal solutions be c o* and l *j . The dual of
i.e., wA = 2wB, then both firms exhibit the same CE. (18) can be expressed as
Similarly, concerning RE evaluation, if any two firms, A
and B, have the same amount of inputs and outputs, i.e., s s
xA = xB and yA = yB, and the unit output price of firm A is max åu
r =1
r y ro + s o s .t . åu
r =1
r y rj + s o £ c j , ( "j ),
(19)
twice that of bank B for each output, i.e., pA = 2pB, then
both firms exhibit the same RE. u r ³ 0 ( "r ) and s o : free
It is thus imperative to recognize that other than
technological factors, banks’ cost/revenue structures are Let the optimal solution of (19) be u r* ("r) and s o* . At
influenced by pecuniary factors as well, thus aiming the optimum, the objective function values of (18) and (19)
atreducing/increasing cost/revenue without which scale are same. That is,
elasticity information based on [Cost]/[Revenue] DEA
models are potentially misleading. In the light of imperfect s
input/output markets where non-linear scale line is c o* = å u r* y ro + s o* (20)
r =1
increasingly common, we therefore argue to consider
alternative models to estimate scale elasticity, which can
provide important insights concerning scale economies Using formula (4), SE of bank ‘o’ can be obtained as
behavior not only to managers when making operational
s s
decisions, but also to policy makers debating on regulatory
issues.
rc ( N ) = c o* å y ( ¶c
r =1
r
*
o / ¶y ro ) = c o* åu
r =1
*
r yro
(21)
* * * * *
= c ( c - s ) = 1 (1 - (s / c ))
o o o o o
3.4.1 Scale elasticity in cost environment

We discuss the empirical evaluation of SE of bank ‘o’ Bank ‘o’ exhibits economies of scale/no economies/
whose input and output considered are, respectively, total diseconomies of scale if s o* is greater than/equal to/
cost (co) and yo = (y1o, y2o, ..., y5o). Here, c o = å i =1 w ij x ij .
m
less than zero, respectively.
One needs first to run the following output-oriented linear

5RTS and economies of scale are the same under two conditions: 1) input/output prices are exogenous, and 2) technology structure is
homothetic.
6When input market is imperfect where input demand is inversely related to its price, the production and cost elasticities are different,

dy / y dy / y ew
i.e., rc = = = r p where ew is input price elasticity of demand. Similarly, when output
(1 - e w )( dw / w ) (1 - (1/ e w ))( dx / x ) (1 - e w )
market is imperfect where output demand is inversely related to its price, production and revenue elasticities are different, i.e.,
dr / r d ( py )/ py æ 1 ö
rr = = = ç1 - ÷ r p , where e p is the output price elasticity of demand.
dx / x dx / x ç e p ÷ø
è
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 69
IJOR Vol. 4, No. 2, 63-79 (2007)

However, if bank ‘o’ is inefficient (i.e., q * < 1 in (17)), associated with one input, i.e., (co) and one output i.e., (ro)
where c o = å i =1 w ij x ij and ro = å r =1 pro yro
m s
then one first needs to find out the following projections: for all o.
The maximum revenue of bank ‘o’ can be obtained from
y ro ® q * y ro + s r+ * ( "r ), and c o ® c o - s - * (22) the following LP:

Using these projected data, one needs to run the model n

(19) to compute rc(N). However, in case of multiple ro* = max å r j l j

j =1
optima, one can compute lower and upper bounds of rc(N) n n
(27)
to arrive at average SE. For details on this cost DEA model,
see Tone and Sahoo (2005, 2006).
s .t . å c j l j £ co ,
j =1
å l j = 1, l j ³ 0
j =1

3.4.2 Scale elasticity in revenue environment The scale elasticity in cost-revenue environment can be
obtained from the dual of (34), which is given below:
Each bank ‘o’ (o = 1, 2, …, n) is associated here with
revenue (ro) with input vector xo = (x1o, x2o, ..., xmo) where min vc o + xo
ro = å r =1 pro yro The maximum revenue of bank ‘o’ can be
s
(28)
s .t . vc j + xo ³ r j , ( "j ), and xo : free
obtained from the following LP:

n
Let the optimal solution to (28) be v* and xo* . At the
ro* = max å r j l j optimum, the objective function values of models (27) and
j =1
(23) (28) are same. That is,
n n
s .t . å x ij l j £ x io , ( "i ),
j =1
å l j = 1, l j ³ 0
j =1 ro* = v *c o + xo* (29)

Let the optimal solutions be ro* and l *j . The dual of So, from (29) SE of bank ‘o’ can be obtained as:
(23) can be expressed as:
rcr = ( dro* / dc o ) ( ro* / c o ) = v *c o ro*
(30)
m
= ( ro* - x o* ) ro* = 1 - (xo* ro* )
min åv x
i =1
i io +y o
(24)
m Bank ‘o’ exhibits economies of scale/no
s .t . å v i x ij +y o ³ r j , ( "j ), ( "i ) and y o : free
i =1
economies/diseconomies of scale if xo* is less
than/equal to/greater than zero, respectively. In case of
which can be used to calculate SE. Let the optimal solution multiple optima, in the spirit of Tone and Sahoo (2005,
to model (24) be v i* and y o* . At the optimum, the 2006), one can compute the lower and upper bounds of
rcr(N) so as to arrive at average SE.
objective function values of (23) and (24) are same. That is,
Note that in case of single input and single output, SE
m
in cost-revenue vis-à-vis production environment can be
ro* = å v i* x io + y o* (25) related as:
i =1

rcr = ( dr r ) ( dc c )

( ( )
Using formula (6), SE of bank ‘o’ can be obtained as (31)
ë û )
= é1 - 1/ e p ù éë1 - (1/ e w ) ùû r p
m
¶ro* m
rr ( N ) = å x i ro* = å v i* x io ro*
i =1 ¶x i i =1 (26) and, in case of multiple input and output, the above
*
= ( r -y ) r = 1 - (y * * * *
r ) relationship can be expressed as
o o o o o

æ s ö s
Bank ‘o’ exhibits economies of scale/no economies/ d ç å pr y r ÷ / å pr y r
dr r
diseconomies of scale if y o* is less than/equal to/greater rcr = = è r =m1 ø r =1
dc c æ ö m
than zero, respectively. However, in case of multiple d ç å wi x i ÷ / å wi x i
optima, one can, in the spirit of Tone and Sahoo (2005, è i =1 ø i =1 (32)
2006), compute the lower and upper bounds of rr(N) to s

arrive at average SE. å éë1 - (1/ e pr )ùû( dyr / yr )s r

= r =1
m
3.4.3 Scale elasticity in cost-revenue environment å éë1 - (1/ e
i =1
wi )ùû( dx i / x i )s i

One can use a DEA model in which each bank ‘o’ is

Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 70
IJOR Vol. 4, No. 2, 63-79 (2007)

where sr and si are, respectively, the rth output’s revenue loan assets’ and ‘non-interest income’ are, respectively, the
share and ith input’s cost share, i.e., s r = pr y r / å r =1 pr y r ‘average interest earned on per rupee unit of investment’,
s

[(p) I], ‘average interest earned on per rupee unit of

and s i = w i x i / å i =1 w i x i .
m
performing loan assets’, [(p) PLA], and ‘non-interest
fee-based income on per rupee of working funds, [(p)
4. THE INDIAN COMMERCIAL BANK DATA NonII]. The input and output data as well as their prices
have been taken from the various sections of ‘Statistical
There has been debate concerning what banks produce Tables Relating to Banks in India’, Reserve bank of India
(outputs) and what resources (inputs) banks consume in and from Indian Banking Association publications. The
that process. In the literature, there are two approaches: relevant data are downloaded from http://www.rbi.org.in/
production approach and intermediation approach, to rbi-sourcefiles/annualdata/bs_annualdata.aspx.
measure bank efficiency. In the former, Ferrier and Lovell Our study covers eight years commencing from the
(1990) use capital, labor and other non-financial inputs to financial year 1997-98. This is the year in which
provide deposits and advances. In the latter, however, a competition intensifies in the banking industry with a total
bank is treated as a producer of intermediation services - of around 100 banks, a shift from around 80 banks in the
by transforming risk and maturity profile of funds received preceding years7. The Regional Rural Banks have their
from depositors to investment or loan portfolio of operations limited to a few contiguous districts and mostly
different risk and maturity profile. Banks also provide serve credit to local farmers and a few small-scale
services for which specific charges are levied, money value enterprises. Because these banks operate for some special
of non-interest income is considered another output purpose, and provide service to a small target group, they
variable. To sum up, banks in general are considered to have been excluded from our study to avoid inconsistencies.
have three outputs: Investments (I), performing loan assets Since data are not available for all the banks for all the
(PLA) and non-interest income (NonII), and three inputs: years, we have considered a balanced panel data on 81
borrowed funds (BF), labor (L) and fixed assets (FA). See banks (26 nationalized banks (NB), 29 Indian private banks,
Berger and Mester (1997) for a comprehensive discussion PB(I) and 26 foreign banks, PB(F)) over a period of eight
of these two approaches. years: 1997-98 – 2004-05. In the spirit of Bhattacharyya et
Besides being profit driven, banks are also forced to take al. (1997), each bank’s annual performance is treated here a
up economic and social responsibilities like safety of distinct bank. So, we have in total 648 (= 81 * 8)
customers, financing much needed public sector commercial banks in our sample period to construct a
expenditure in various social and economic services, and single ‘grand frontier,’ which provides a benchmark against
this study, therefore, adopted the intermediation approach. which to calculate the efficiency of each bank in each
More importantly, the essence of taking PLA, as an output year8.
measure is more realizable in Indian context, because only
earning asset contributes to revenue of bank and not total 5. RESULT AND DISCUSSION
loan. This approach is effective in analyzing management’s
success. Coates (1990) also provides a comprehensive The analysis of efficiency on the input-side9 is becoming
description of the objectives of the Indian banking system increasingly common in DEA applications for a variety of
for which production approach seems to be inappropriate. reasons. First, real world managers are never given a bundle
All the monetary values of inputs and outputs have been of inputs and told to produce the maximum output from it.
deflated using wholesale price index deflator with base Instead they are given output targets and told to produce it
1993-94. most efficiently. Second, profitability in any business hinges
Concerning the prices of inputs and outputs, the unit on the efficiency of operations. But if the business
prices of the inputs: ‘borrowed funds’, ‘labor’ and ‘fixed involves a commodity, then what depends on efficient
assets’ are, respectively, the ‘average interest paid per rupee operations is survival. When prices are beyond companies’
of borrowed funds’, [(w) BF], ‘average staff cost’, [(w) L], control, what remain are costs. This reflects the companies’
and ‘non-labor operational cost per rupee amount of fixed emphasis on the input dimensions of policies. On a
asset’, [(w) FA]; and outputs: ‘investments’, ‘performing tentative basis, it has been suggested in the literature that

7It would have been interesting to examine productivity performance variations of banks just after the financial liberalization was
introduced in 1991. However, the unavailability of data on unit prices of inputs of banks up to 1996, which are required to estimate
cost efficiency and scale elasticity, forced us to conduct this study starting with the year 1997-98.
8This approach has some advantage in terms of yielding information on trends in performance, which would not have been available,

had we used DEA to calculate annual frontiers (which Das et al. (2004) have used in their study), since the benchmarks would likely
change from year to year. However, a Malmquist productivity index is another approach to study productivity change, which can be
taken up as an extension of this study.
9
Bhattacharyya et al. (1997) have, however, considered in their study the output orientation where each bank seeks to maximize its
service provision (advances, investments and deposits), given the resources at its disposal.
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 71
IJOR Vol. 4, No. 2, 63-79 (2007)

costs (or inputs) are generally more predictable than annual trends in productivity performance behavior of
outputs, giving cost targets a greater credibility than those banks, first, in terms of efficiency, and then, in term of
for outputs. Sengupta (1987) has argued that: “…… data scale elasticity with respect to ownership groups.
variations may arise in practical situations …… when the
output measures have large and uncertain measurement 5.1 Efficiency
errors which are much more significant than in the input
measures (p.2, 290). For example in school efficiency 5.1.1 Technical efficiency
studies, the input costs, such as teachers’ salaries,
Three types efficiency are computed: technical efficiency
administrative expenses, etc., may have low measurement
(TE) from model (7), cost efficiency from model (11) and
errors whereas the performance test scores of students
new cost efficiency (NCE) from the input-oriented version
may contain large errors of measurement of true student
of model (17). The annual average TE trends with respect
quality”. This argument is most compelling where
to ownership are exhibited in Figure 1.
measurement errors are large relative to true random
It is seen that deregulation is yielding a remarkable
fluctuations in the production process. Note that the
increase in annual average trend performance for all the
assumption of variable returns to scale is maintained since
ownership groups. PB(F) group consistently remained high
our DEA results do not support the assumption of
and above NB and PB(I) groups over the entire sample
constant returns to scale.
period. However, on comparison of NB with PB(I) reveals
Following Sengupta (1988), we have used the step-wise
that the former outperforms the latter11. It is noteworthy
DEA approach where aggregated metrics in the first step
that that TE performance trends of both NB and PB(F)
were disaggregated into productivity-significant
remained above the grand average (shown in dotted line)
determinant factors to give a robust DEA productivity
whereas PB(I) remained below it. Also seen is that the
metric in step 2. Because inputs and outputs used in DEA
variation in trend performance of each ownership group
should satisfy the condition that greater quantities of the
remains more or less the same, reflecting their similar
selected inputs provide increased output, an isotonicity test
familiarity with the regulatory system in terms of
between inputs and outputs at step 1 was conducted. If
dependence on wholesale or corporate resources,
positive intercorrelations between inputs and outputs were
inter-bank market borrowings, refinance of assets, etc.
found (Pearson correlations, a = 0.05), the isotonicity test
On seeing the banks on the ‘grand frontier’ reveals that a
was passed. In step 1 BCC efficiency scores obtained from
total of 70 banks, approximately 11% of the sample, are
composite (aggregated) output (y) and input (x)10 were
rated being technically efficient. See Table 2. Of these 70
correlated with disaggregated inputs/outputs, correlations
best banks, 45 (2 + 3 + 18 + 22) come from the last four
between BCC scores and I (p = 0.382, a = 0.000), PLA (p years and 25 (8 + 6 + 6 + 5) come from first four years of
= 0.516, a = 0.000), NonII (p = 0.364, a = 0.000), BF (p = the sample period, which justifies for their increasing
-0.289, a = 0.000), FA (p = -0.383, a = 0.000), and L (p = performance trends. On the distribution of 70 best banks
-0.285, a = 0.000) revealed that all the outputs and inputs with respect to ownership, we find 39 foreign banks, 19
can significantly enhance and determine the productivity nationalized banks and 12 private banks, which clearly
levels. Thus, in constructing DEA model at step 2, all the indicate that with the deregulation of the banking sector in
three outputs and three inputs were used, and this model India, foreign banks are not only found playing an active
can be considered a robust productivity metric. role in Indian financial market but also setting performance
Table 1 exhibits the descriptive statistics of all the three standards.
output and three input variables, and their respective unit Adopting and practicing latest technologies may be one
prices where average NonII and FA values are more less of the reasons for each bank group exhibiting increasing
constant over all the eight years. All the variables are TE trend. The improvement in TE for all the ownership
measured in crores of rupees. (1 crore = 10 million). Both groups may be due to the intense competition. Leibenstein
composite output and input as well as their significant (1966) maintains that exposure to competition will generate
constituents have grown fairly steadily over years improvement in efficiency (i.e., X-efficiency or technical
(excepting for the year 1998-99). This trend holds true for efficiency). He argues that enterprises exposed to
cost and revenue figures as well. Also evident is the steadily competition respond by eliminating internal inefficiency,
increasing variations in output and input variables, as and seek out opportunities for innovation. To Stigler
reflected in their SD scores, being more than their means.
We first present our discussion concerning average

Here y = å r =1 s r y r and x = å i = 1 s i y i where sr and si are, respectively, the rth output’s revenue share and ith input’s cost
10 3 3

share, i.e., s r = pr y r / å r = 1 pr y r and s i = w i x i / å i = 1 w i x i .

s m

11The differences in TE performance among all possible ownership groups are statistically tested with the help of rank-sum-test
developed by Wilcoxon-Mann-Whitney. We find the t-values significant at 1% level between NB and PB(I) (t = 10.349), between NB
and PB(F) (t = 11.171) and PB(I) and PB(F) (t = 31.846). However, TE difference between nationalized banks, and private and foreign
banks as a whole turns out to be insignificant (t = -0.002).
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 72
IJOR Vol. 4, No. 2, 63-79 (2007)

Table 1. Summary statistics for Indian banks: 1997-98 - 2004-05

1997-98 1998-99 1999-00 2000-01 2001-02 2002-03 2003-04 2004-05
(O) II Mean 20.615 23.891 28.404 32.756 37.102 43.173 53.414 53.884
SD 45.647 51.241 61.410 72.643 89.516 111.576 142.663 143.917
(O)PLA Mean 66.392 149.415 75.348 83.679 89.630 97.734 109.363 110.324
SD 149.105 701.636 185.142 210.525 219.883 230.052 243.612 245.753
(O) Nonll Mean 0.891 1.060 1.059 1.241 1.268 1.336 1.458 1.471
SD 2.371 2.464 2.683 2.770 2.954 3.163 3.440 3.470
Y Mean 43.446 86.554 51.820 58.133 63.287 70.371 81.290 82.005
SD 96.587 357.338 122.928 141.219 154.441 170.258 191.446 193.128

(I) BF Mean 2.067 2.184 3.268 3.522 4.077 5.657 9.770 9.856
SD 6.211 6.969 7.874 7.687 9.099 12.704 23.339 23.544
(I) FA Mean 0.595 1.070 1.181 1.198 1.204 1.255 1.419 1.432
SD 1.531 1.753 2.099 2.164 2.133 2.146 2.425 2.446
(I) L Mean 11530 11359 11301 11419 10754 10501 11046 11143
SD 29688 29417 28929 28843 27627 26668 27451 27692
X Mean 2465.603 2348.595 2346.146 2305.229 2425.808 2628.666 3063.996 3090.951
SD 6939.791 6829.633 6467.732 6130.776 6513.525 6972.524 8053.851 8124.636

(P) I Mean 0.392 0.327 0.302 0.280 0.263 0.262 0.275 0.277
SD 0.203 0.124 0.135 0.113 0.089 0.105 0.145 0.146
(P) PLA Mean 0.101 0.105 0.106 0.104 0.102 0.101 0.103 0.104
SD 0.024 0.023 0.016 0.015 0.013 0.018 0.033 0.033
(P) Nonll Mean 0.018 0.023 0.017 0.022 0.019 0.019 0.021 0.021
SD 0.014 0.019 0.018 0.025 0.023 0.027 0.041 0.041

(C) BF Mean 4.939 5.573 8.606 8.682 6.348 9.872 23.957 24.168
SD 7.390 8.502 24.753 33.110 10.721 20.483 72.793 73.433
(C) FA Mean 1.420 1.522 1.889 1.629 1.436 1.529 1.813 1.829
SD 3.155 3.803 5.255 4.174 2.614 2.847 3.621 3.653
(C) L Mean 0.0001 0.0007 0.0005 0.0005 0.0008 0.0042 0.0041 0.0046
SD 0.0002 0.0037 0.0027 0.0016 0.0030 0.0228 0.0182 0.0212

REVENUE Mean 12.202 13.008 14.659 15.779 17.367 19.285 22.091 22.481
SD 28.137 28.807 32.674 35.236 380847 42.878 48.416 49.271
COST Mean 5.733 7.041 7.162 7.568 8.365 9.801 11.160 11.427
SD 13.393 15.921 15.962 17.138 19.371 22.250 25.792 26.295

Figure 1. TE w.r.t. ownership over time.

(1976), this X-efficiency gain is nothing but an increase in the period), confirming the above-mentioned claim of
the intensity of labor or, equivalently, a reduction in increased work intensity.
on-the-job leisure. Ganley and Grahl (1988) pointed out
that, where labor productivity has increased due to such 5.1.2 Cost/New cost efficiency
competition, there is evidence of increased work intensity.
A closer look at our data set reveals that labor productivity The average annual trends in CE behavior with respect to
shows an increasing trend (NB: from 0.004 to 0.008, PB(I): each ownership group are all exhibited in Figure 2. Here,
from 0.014 to 0.089, and PB(F): from 0.122 to 0.241 over banks on each ownership group show increasing cost
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 73
IJOR Vol. 4, No. 2, 63-79 (2007)

Table 2. TE Frontier banks by ownership from and by year

Banks Ownership Years
1997-98 1998-99 1999-00 2000-01 2001-02 2002-03 2003-04 2004-05
State Banks of Indian Public C C C C C
State Banks of Bikaner & Public C
Jaipur
State Banks of Hyderabad Public C C
State Banks of Mysore Public C
State Banks of Patiala Public C C
Bank of Baroda Public C
Canar of Bank Public C
Indian Overseas Bank Public C
Punjab National Bank Public C C C
Andhra Bank Public C
Oriental Bank of Public C
Commerce
Indusind Bank Private C
Benares State Bank Private C
Karur Vysya Bank Private C C
Global Trust Bank Private C C
Lord Krishna Bank Private C C
Jammu & Kashmir Bank Private C
Ratnakar Bank Private C C
Nainital Bank Private C
Chase Manhattan Bank Foreign C C C
Chinatrust Commercial Foreign C C
Bank
ABN Amro Bank Foreign C C
Bank of Tokyo Foreign C C C
DBS Bank Foreign C C
Citibank Foreign C C C
Commerzbank Foreign C C
Abu-Dhabi Commercial Foreign C C
Bank
Credit Lyonnais Foreign C C C C
Bank of Ceylon Foreign C C
Bank of International Foreign C C
Indonesia
State Bank of Mauritius Foreign C C C C C C
Standard Chartered Bank Foreign C C C C
HongKong & Shanghai Foreign C C
Bank

efficiency accrual, and the performances of nationalized frontier banks (see Table 3) where out of 19 cost efficient
banks are outstanding followed by foreign banks and banks (approximately 3% of the sample), nine are PB(F),
private banks. The differences in CE among all ownership eight are NB and only two are PB(I). On further statistical
groups are also confirmed from significant Wilcoxon-Mann-Whitney test reveals that t = 14.502
Wilcoxon-Mann-Whitney t-test values found between NB between NB and PB(I), t = 7.253 between NB and PB(F)
and PB(I) (t = 3.971), NB and PB(F) (t = 21.641) and PB(I) and t = 31.301 between PB(I) and PB(F). However,
and PB(F) (t = 28.584). However, CE difference between differences in cost performance between NB, and private
NB, and PB(I + F) was found insignificant (t = -0.003). and foreign banks as a group, are not statistically significant
Note that these results may not be reliable for, as (t = -0.0004).
discussed earlier, the cost DEA model (11) suffers from The main reason for the foreign banks found superior
some fundamental shortcomings, and therefore, we present over the rest is the use of the relatively fewer number of
the average annual trends in new cost efficiency in Figure employees implying higher work intensity. However, in
3. spite of the facts that nationalized banks are the oldest
Foreign banks are found the best, and private banks the banks with strong asset base, their cost performances are at
worst. This is also seen from the distribution of cost stake. This might be due to the possibility that with the
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 74
IJOR Vol. 4, No. 2, 63-79 (2007)

emergence of new private and foreign banks, the older particular group, NB (say) is evaluated with respect to all
banks have devised a market responsive product-mix the banks in the other group, PB(I)/PB(F)/PB(I + F)
concerning saving and invest plans offering attractive where PB(I + F) represents all the private and foreign
returns, and they are going through the process of banks. For example, to measure TE of banks ‘o’ (o Î NB)
overhauling with significant decentralization in the in PB(I), we formulate the following LP:
management and organizational structure, causing huge
loss in allocative efficiency. min q
Note that we have used a mixed evaluation of banks in
three ownership groups where each bank is evaluated on
s .t . å
j ÎPB ( I )
x j l j £ q xo , å
j ÎPB ( I )
y j l j ³ yo ,
both inter and within group bases. However, following
Cooper et al. (2000), a sharper discrimination between any å
j ÎPB ( I )
l j = 1, l j ³ 0 ( "j Î PB( I ))
two groups can be obtained when each bank from a

Figure 2. CE w.r.t. ownership over time.

Figure 3. New CE w.r.t. ownership over time.

Table 3. NCE Frontier banks by ownership from and by year

Banks Ownership Years
1997-98 1998-99 1999-00 2000-01 2001-02 2002-03 2003-04 2004-05
State Banks of Indian Public C C C C C C C
Andhra Bank Public C
Lord Krishna Bank Private C C
Chase Manhattan Bank Foreign C
Bank of Tokyo Foreign C C
Citibank Foreign C C
Abu-Dhabi Commercial Bank Foreign C C
Standard Chartered Bank Foreign C
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 75
IJOR Vol. 4, No. 2, 63-79 (2007)

Table 4. Bilateral comparisons among ownership groups

Technical Efficiency New Cost Efficiency

Rank Sum NB PB(I) Test Stat. (t) NB PB(I) Test Stat. (t)
44829 52191 -0.7772 55666 41354 7.3606
NB PB(F) Test Stat. (t) NB PB(F) Test Stat. (t)
Rank Sum
55737 30999 10.0877 63667 23069 16.5552
NB PB(I + F) Test Stat. (t) NB PB(I + F) Test Stat. (t)
Rank Sum
92028 118248 11.0266 111686 98590 19.8625
NB PB(F) Test Stat. (t) NB PB(F) Test Stat. (t)
Rank Sum
64216 32804 9.8071 72253 24767 15.8422

We test against the null hypothesis that the two The lower and upper bounds of SE estimates of banks
ownership groups have the same distribution of efficiency over the years are computed using BCC (eqn.10), COST
scores. The results of these bilateral comparisons (eqn.21), REVENUE (eqn.26) and COST-REVENUE
concerning TE and CE across all ownership groups are all (eqn.30) DEA models, and the distribution of returns to
exhibited in Figure 4(a)-4(d) and Figure 5(a)-5(d), scale exhibited below in Table 5 is uneven indicating that
respectively. policy prescription concerning whether to expand or
Concerning the TE comparison, excepting the case contract business operations depends upon the very
between nationalized and private banks, the picture is quite objective of the bank. Also evident is the distinction
clear-cut. Foreign banks are seen outperforming over both between RTS (in production environment) and economies
the nationalized banks and private banks, and private and of scale (in price environment) from the number of banks
foreign banks as a group outperforming over the operating in IRS column.
nationalized banks. And, as regards the new cost efficiency The annual average SE estimates for all the banks with
comparisons, we find private banks scoring over the respect to ownership over time in all environments are all
nationalized banks, private and foreign banks as a group exhibited in Figures 6, 7, 8 and 9.
outperforming over the nationalized banks, and foreign In all the models, one could clearly see that SE estimates
banks scoring over both nationalized and private banks. of foreign banks are higher, followed by those of private
These visual findings are further confirmed through and nationalized banks, respectively, and foreign banks
statistical Wilcoxon-Mann-Whitney test statistics exhibited mostly exhibit IRS/economies of scale whereas
in Table 4. Excepting the case between NB and PB(I), we nationalized banks exhibit DRS/diseconomies of scale as
reject throughout the null hypothesis that TE/NCE scores their average SE scores are mostly above unity and less
of any possible two groups belongs to the same than unity respectively. This result is not at all surprising
distribution at the 1% level of significance. because of the RBI’s branching policy. As Bhattacharyya et
The finding of higher cost efficiency of private banks al. (1997) pointed out, under the RBI’s branching policy,
over nationalized banks clearly highlights the possible Indian banks are required to open branches but are not
disciplining role increasingly played by the capital market in allowed to close unprofitable branches, and this policy
improving the weak relationship between market for prevents banks from optimizing their resources across the
corporate control and efficiency of private enterprise branch network because banks have neither control over
assumed by the property right hypothesis. Even though the location of branches nor the ability to close
institutional conditions are favorable, the lesser loss-making branches. However, foreign banks are mostly
cost-efficiency growth can be understandable because of young, and tend to have smaller branch networks, since
X-inefficiency factors arising from government they have not yet fully expanded their business and have
ownership12, which might be argued to be leading to not been forced by regulators to expand branch networks
diminishing return to income, reduction in interest spread, beyond their optimal size. However, the SE nature for
and the presence of scale economies due to fixed cost. private banks is mix, supporting both economies and
Now let us turn to discuss the scale elasticity issue across diseconomies of scale. This might be due that Indian
ownership groups. private banks of two types: old and new, the former
follows the tradition of NB whereas the latter with PB(F).
5.2 Scale elasticity

12Government officials are in general more inclined to pursue their own interests, or interest of pressure group, rather than interests of
public. Frequently changing objectives of nationalized banks arising from government’s attempts to accommodate diverse interest
groups creates hindrances in their growth.
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 76
IJOR Vol. 4, No. 2, 63-79 (2007)

Figure 4(a). NB vis-a-vis PB(I). Figure 4(b). NB vis-a-vis PB(F).

Figure 4(c). NB vis-a-vis PB(I + F). Figure 4(d). PB(I) vis-a-vis PB(F).

Figure 5(a). NB vis-a-vis PB(I). Figure 5(b). NB vis-a-vis PB(F).

Figure 5(c). NB vis-a-vis PB(I + F). Figure 5(d). PB(I) vis-a-vis PB(F).

Figure 6. SE in production environment.

Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 77
IJOR Vol. 4, No. 2, 63-79 (2007)

Table 5. Distribution of RTS

Models IRS/Economies of scale CRS/No economies DRS/ Diseconomies of scale
BCC 267(IRS) 31(CRS) 350(DRS)
NB 6 6 196
PB(I) 117 8 107
PB(F) 144 17 47

COST 395(economies of scale) 3(no economies) 250(diseconomies of scale)

NB 21 1 146
PB(I) 214 0 18
PB(F) 160 2 46

REVENUE 290(economies of scale) 28(no economies) 330(diseconomies of scale)

NB 19 10 179
PB(I) 118 3 111
PB(F) 153 15 40

COST-REVENUE 303(economies of scale) 1(no economies) 344(diseconomies of scale)

NB 2 0 206
PB(I) 114 0 88
PB(F) 157 1 50

Figure 7. SE in Cost environment.

Figure 8. SE in revenue environment.

Note that the conflicting signals concerning scale model), or doing both, i.e., minimizing cost and
economies behavior of banks obtained from both maximizing revenue (COST-REVENUE model).
production and price-based DEA models are simply due to
the very nature of their underlying objectives pertaining 6. CONCLUDING REMARKS
whether to minimize inputs (BCC model), or to minimize
cost (COST model), or to maximize revenue (REVENUE This paper empirically estimates the productivity
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 78
IJOR Vol. 4, No. 2, 63-79 (2007)

Figure 9. SE in Cost-revenue environment.

performance in terms of technical efficiency, cost REFERENCES

efficiency and scale elasticity of the banking sector for the
period: 1997-98 – 2004-05. Our broad empirical results are 1. Alchian, A.A. (1965). Some economics of property
indicative in many ways: first, the average annual trends in rights. II Politico, 30: 816-829.
TE for all ownership groups have improved, indicating an 2. Banker, R.D., Charnes, A., and Cooper, W.W. (1984).
affirmative gesture about the effect of the reform process Some models for estimating technical and scale
on the performance of the Indian banking sector, second, inefficiencies in data envelopment analysis. Management
foreign banks have a leading edge over both the Science, 30: 1078-1092.
nationalized banks in both operational and price measures 3. Baumol, W.J., Panzar, J.C., and Willig, R.D. (Eds.)
of performance, which indicates that foreign banks are (1982). Contestable Markets and The Theory of Industrial
mostly strongly exposed to international markets, and are Structure, Harcourt Brace Jovanovich, New York.
more sensitive to competitive pressures from outside the 4. Berger, A.N. and Mester, J.L. (1997). Beyond the black
country, third, the higher cost efficiency accrual of private box: What explains differences in the efficiencies of
banks over the nationalized banks indicate that the financial institutions? Journal of Bank and Finance, 21:
nationalized banks though old, do not reflect their learning 87-98.
experience in their cost minimizing behavior due to 5. Bhattacharyya, A., Lovell, C.A.K., and Sahay, P. (1997).
X-inefficiency factors arising from government ownership. The impact of liberalization on the productive
This finding also highlights the possible stronger efficiency of Indian commercial banks. European
disciplining role played by the capital market indicating a Journal of Operational Research, 98: 332-345.
strong link between market for corporate control and 6. Chatterjee, G. (1997). Scale economies in banking:
efficiency of private enterprise assumed by property right Indian experience in deregulated era. RBI Occasional
hypothesis. And, finally, concerning the scale elasticity Paper, 189: 25-59.
behavior, the technology- and market-based results differ 7. Coates, W. Jr. (1990). Lessons of financial liberalization
significantly supporting the empirical distinction between for India. Economic and Political Weekly, 5-12: 1043-1046.
returns to scale and economies of scale, which are most 8. Cooper, W.W., Seiford, L.M., and Tone, K. (Eds.)
often used interchangeably in the literature. (2000). Data Envelopment Analysis: A Comprehensive Text
This study points to avenues for future research in two with Models, Applications, References and DEA-Solver
ways: first, one can use both quadratic and dynamic cost Software. Kluwer Academic Publishers, Boston.
frontier DEA models proposed by Sengupta (2003) to 9. Das, A., Nag, A., and Ray, S.C. (2004). Liberalization,
study efficiency and technical changes over time, and ownership and efficiency in Indian banking: A
second, one can use Malmquist productivity index nonparametric approach. Department of economics
approach to measure and analyze productivity change and working paper series, University of Connecticut.
its components such as pure efficiency change, scale 10. De Alessi, L. (1980). The economics of property rights:
efficiency change and technical change over time. A review of the evidence. In: Richard O.Z. (Ed.),
Research in Law and Economics: A Research Annual, vol. 2,
ACKNOWLEDGEMENTS Jai Press, Greenwich, C.T. 1980., pp. 1-47.
11. Färe, R., Grosskopf, S., and Lovell, C.A.K. (1985). The
The authors are grateful to Dr. Chin-Tsai Lin, the Measurement of Efficiency of Production, Kluwer-Nijhoff
Editor-in-Chief and three anonymous referees for their Publishing, Boston.
helpful comments and suggestions on the previous draft 12. Färe, R., Grosskopf, S., and Lovell, C.A.K. (1986).
of this paper. Scale economies and duality. Journal of Economics, 46:
175-182.
13. Färe, R., Grosskopf, S., and Lovell, C.A.K. (1988).
Sahoo, Sengupta, and Mandal: Productive Performance Evaluation of the Banking Sector in India Using Data Envelopment Analysis 79
IJOR Vol. 4, No. 2, 63-79 (2007)

Scale elasticity and scale efficiency. Journal of economy: The case of India. European Journal of
Institutional and Theoretical Economics, 144: 721-729. Operational Research, 148: 662-671.
14. Färe, R. and Primont, D. (1995). Multi-output Production 33. Sengupta, J.K. (1987). Efficiency measurement in
and Duality: Theory and Applications, Kluwer Academic non-market systems through data envelopment
Publishers, Boston. analysis. International Journal of Systems Science, 18:
15. Ferrier, G. and Lovell, C.A.K. (1990). Measuring cost 2279-2304.
efficiency in banking: Econometric and linear 34. Sengupta, J.K. (1988). A robust approach to the
programming evidence. Journal of Econometrics, 46: measurement of Farrell efficiency. Applied Economics,
229-245. 20: 273-283
16. Førsund, F.R. (1996). On the calculation of scale 35. Sengupta, J.K. (2000). Dynamic and Stochastic Efficiency
elasticity in DEA models. Journal of Productivity Analysis, Analysis: Economics of Data Envelopment Analysis, World
7: 283-302. Scientific, New Jersey.
17. Fukuyama, H. (2001). Returns to scale and scale 36. Sengupta, J.K. (2003). New Efficiency Theory: with
elasticity in Farrell, Russell, and Additive models. Application of Data Envelopment Analysis, Springer, New
Journal of Productivity Analysis, 16: 225-239. York.
18. Ganley, J. and Grahl, J. (1988). Competitive tendering 37. Shephard, R.W. (1970). Theory of Cost and Production
and efficiency in refuse collection: A critical comment. Function, Princeton University Press, Princeton.
Fiscal Studies, 9: 80-85. 38. Shyu, J. (1998). Deregulation and bank operating
19. Gilbert, R.A. and Wilson, P.W. (1998). Effects of efficiency: An empirical study of Taiwan banks. Journal
deregulation on the productivity of Korean banks. of Emerging Markets, 3: 27-46.
Journal of Economics and Business, 50: 133-155. 39. Starrett, D.A. (1977). Measuring returns to scale in the
20. Hanoch, G. (1970). Homotheticity in joint production. aggregate, and the scale effect of public goods.
Journal of Economic Theory, 2: 423-426. Econometrica: Journal of the Econometric Society, 45:
21. Hao, J., Hunter, C.W., and Yang, W.K. (2001). 1439-1455.
Deregulation and efficiency: The case of private 40. Stigler, G.J. (1976). The existence of x-efficiency.
Korean banks. Journal of Economics and Business, 53: American Economic Review, 66: 213-216.
237-254. 41. Subrahmanyam, G. (1993). Productivity growth in
22. Leibenstein, H. (1966). Allocative efficiency vs. India’s public sector banks: 1979-89. Journal of
x-efficiency. American Economic Review, 56: 392-45. Quantitative Economics, 9: 209-223.
23. Leightner, E.J. and Lovell, C.A.K. (1998). The impact 42. Sueyoshi, T. (1997). Measuring efficiencies and returns
of financial liberalization on the performance of Thai to scale of Nippon telegraph & telephone in
banks. Journal of Economics and Business, 50: 115-132. production and cost analyses. Management Science, 43:
24. Levy, B. (1987). A theory of public enterprise behavior. 779-796.
Journal of Economic Behavior and Organization, 8: 75-96. 43. Sueyoshi, T. (1999). DEA duality on returns to scale
25. Nickskamen, W. (1971). Bureaucrats and politicians. (RTS) in production and cost analyses: An occurrence
Journal of Law and Economics, 18: 617-643. of multiple solutions and differences between
26. Panzar, W. and Willig, R.D. (1977). Economies of scale production-based and cost-based RTS estimates.
in multi-output production. Quarterly Journal of Management Science, 45: 1593-1608.
Economics, 91: 481-493. 44. Tone, K. (2002). A strange case of the cost and
27. Rammohan, T.T. and Ray, S.C. (2004). Comparing allocative efficiencies in DEA. Journal of the Operational
performance of public and private sector banks: A Research Society, 53: 1225-1231.
revenue maximization efficiency approach. Economic 45. Tone, K. and Sahoo, B.K. (2003). Scale, indivisibilities
and Political Weekly, 39: 1271-1276. and production function in data envelopment analysis.
28. Rangarajan, C. and Mampilly, P. (1972). Economics of International Journal of Production Economics, 84: 165-192.
scale in Indian banking. Technical studies for banking 46. Tone, K. and Sahoo, B.K. (2004). Degree of scale
commission report, Reserve Bank of India, Mumbai, economies and congestion: A unified DEA approach.
244-268. European Journal of Operational Research, 158: 755-772.
29. Ray, S.C. (2004). Data Envelopment Analysis: Theory and 47. Tone, K. and Sahoo, B.K. (2005). Cost efficiency and
Techniques for Economics and Operations Research, returns to scale in life insurance Corporation of India
Cambridge University Press, Cambridge, UK. using data Envelopment analysis. Socio-Economic
30. Saha, A. and Ravishankar, T.S. (2000). Rating of Indian Planning Sciences, 39: 261-285.
commercial banks: A DEA approach. European Journal 48. Tone, K. and Sahoo, B.K. (2006). Re-examining scale
of Operational Research, 124: 187-203. Eeasticity in DEA. Annals of Operations Research,
31. Sahoo, B.K., Mohapatra, P.K.J., and Trivedi, M.L. published online.
(1999). A comparative application of data 49. Tyagarajan, M. (1975). Expansion of commercial
envelopment analysis and frontier translog production banking: An assessment. Economic and Political Weekly,
function for estimating returns to scale and efficiencies. 10: 1819-1824.
International Journal of Systems Science, 30: 379-394.
32. Sathey, M. (2003). Efficiency of banks in a developing