Professional Documents
Culture Documents
Access to this document was granted through an Emerald subscription provided by emerald-srm:534301 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.
SEF
29,1 On the cross-methodological
validation of bank
efficiency assessments
26
Shrimal Perera and Michael Skully
Department of Accounting and Finance, Monash University,
Caulfield East, Australia
Abstract
Purpose Since there is no agreement on the consistency of their estimates, the purpose of this paper
is to investigate whether parametric stochastic frontier analysis (SFA) and nonparametric data
Downloaded by University of Ghana At 15:50 27 June 2016 (PT)
1. Introduction
Academic research on banking efficiency has predominantly focused on two main
streams of alternative frontier efficiency methodologies: parametric and nonparametric.
The main difference is in their assumptions in terms of:
.
the functional form of the best practice frontier;
Studies in Economics and Finance .
allowance/non-allowance of random error which may temporarily give some
Vol. 29 No. 1, 2012
pp. 26-42 units high or low outputs, inputs, costs, or profits; and
q Emerald Group Publishing Limited
1086-7376
DOI 10.1108/10867371211203837 The authors would like to thank an anonymous referee for his/her insightful comments.
.
if random error is allowed, the distributional assumptions imposed on it to Bank efficiency
separate the inefficiencies and random disturbance (Berger and Humphrey, 1997). assessments
The parametric approach employs econometric techniques to estimate efficiency scores.
They allow for random error and, therefore, are less prone to classify measurement errors
or momentary differences in costs as inefficiency (Bauer et al., 1998). They also impose
structure on the frontier by using a functional form for technology. The parametric 27
approach, however, may suffer from bias due to the distributional assumptions it
imposes to separate random error from inefficiency as neither of them is directly
observed (Bauer et al., 1998, p. 93)
In contrast, the nonparametric approach employs mathematical programming
techniques to obtain relative efficiency scores. It does not make restrictive assumptions
about the functional form of the technology and so impose relatively little structure on
the best practice frontier. Its key limitation, however, is the general assumption that
Downloaded by University of Ghana At 15:50 27 June 2016 (PT)
Our paper extends the existing literature by checking jointly the statistical consistency
of both DEA technical efficiency scores and SFA cost efficiency scores. The prior
studies focus either on technical efficiency or cost efficiency, but not both. Moreover, as
far as we are aware, this is the first methodological cross-examination study to focus
on bank efficiency in the context of a developing country banking market.
(1996), we also investigate whether frontier efficiency estimates (derived from DEA and
SFA specifications) are consistent with non-frontier, financial ratio-based efficiency
measures. Our formal hypotheses are as follows:
H1. Efficiency scores generated by different specifications rank the Indian banks
approximately in the same order.
H2. Different efficiency measurement specifications identify mostly the same
Indian banks as best-practice and as worst-practice.
H3. Efficiency scores generated by different specifications for Indian banks are
consistent with standard non-frontier performance measures.
The first two hypotheses assess the degree to which different approaches are mutually
consistent with each other while the last one helps determine whether the efficiencies
generated by different approaches are consistent with reality or are believable
(Bauer et al., 1998, p. 87).
in the logarithms of input prices and outputs. The formal Translog model used
to estimate equations (1) and (2) is as follows:
X
n X
t
ln TC it a0 am ln P m;it bs ln Qs;it r1 T
m s
" #
1 Xn X
m t X
X s
2
am;n ln P m;it ln P n;it bs;t ln Qs;it ln Qt;it lT
2 m n s t
n X
X t X
t X
n
fm;s ln P m;it ln Qs;it cs T ln Qs;it um T ln P m;it 1;
m s s m
3
where, subscripts i denote banks, and t time horizon and:
ln TC natural log of total costs;
ln Pm natural log of input prices;
ln Qs natural log of output values;
T time trend variable; and
1 composite error term as defined by Battese and Coelli (1995); and a, b, t, r,
f, c, and u are parameters to be estimated.
In line with previous studies, the following restrictions are used to impose linear
homogeneity on the input prices and total cost variable as required by the duality
theorem (Fries and Taci, 2005). Specifically, these restrictions are:
X
n X
n X
n X
n
am 1; am;n 0; fm;s 0; and um 0: 4
m m m m
Linear homogeneity is achieved by normalizing TC and Pm (in equation (3)) by the price
of physical capital (ratio of other operating expenses to fixed assets) before the log
transformations are undertaken (Bos and Kool, 2006). Such normalizing also helps
to account for heteroskedasticity (Berger and Mester, 1997). Furthermore, the symmetry
SEF of second-order parameters is achieved by setting:
29,1 am;n an;m and bs;t bt;s : 5
Moreover, factor share equations embodying Hotellings Lemma and Shephards
Lemma are excluded since they assume away possible allocative inefficiencies in the
sample (Berger and Mester, 1997)[4].
32 Fourier cost efficiency specification. The Fourier functional form imposes a
second-order polynomial in the dependant variables together with a combination of
trigonometric (sine and cosine) terms. This specification has the Translog form nested
within it as a special form and tends to closely approximate any well-behaved function
(Chung et al., 2001). Moreover, due to its trigonometric series, the Fourier functional
form has the ability to represent any function exactly and, therefore, can potentially
approximate any function beyond its local domain (Gallant, 1982). Moreover, Fourier
form is argued to be superior when firms in the sample show wide size disparities
Downloaded by University of Ghana At 15:50 27 June 2016 (PT)
is a scaler and l is a (N*1) vector of constants. The value of u obtained is the efficiency score
for the ith firm. It satisfies the condition, u # 1, with a value of 1 indicating a point
on the efficient frontier and hence a technically efficient firm according to Farrel (1957)
definition. With regard to the scale assumptions, equation (8) represents the DEA variable
scale to returns specification (DEA-VRS) while DEA constant returns to scale specification
(DEA-CRS) is given by equation (8) minus the convexity constraint (N 1l 1).
In equation (8), the input orientation is selected as opposed to an output one following
previous studies (Lozano-Vivas et al., 2001). Under an input orientation, technical
efficiency is measured as a proportional reduction in input usage, with output levels
held constant (Coelli et al., 1998, p. 158). After all, the approach selected will have no
significant impact on the efficiency scores. For example, under DEA-CRS models, both
input and output orientations generate identical efficiency estimates. When VRS
assumption is used, both orientations identify the same set of banks as lying on the
efficient frontier (Coelli et al., 1998).
In order to maintain the consistency with the SFA-Translog (equation (3)) and
SFA-Fourier (equation (6)) specifications explained earlier, the same two outputs
(net loans and other earning assets) are used in the two DEA models (DEA-CRS and
DEA-VRS). However, instead of input prices, nonparametric DEA method requires
input quantities to compute technical efficiency scores. Thus, two inputs are used:
(1) deposits; and
(2) operating expenses (personnel expenses other administration
expenses other operating expenses).
The use of deposits as an input (as opposed to interest expenses) is in line with Tripe
(2005) who argues that interest expenses might reflect any risk premiums included in
interest rates[6].
For ease of reference, a summary of the four different efficiency measurement
specifications used in the cross-examination exercise and their estimation procedures
are presented in Table II. The maximum likelihood estimates of the SFA-Translog
(equation (3)) and SFA-Fourier (equation (6)) specifications are obtained using
Coelli (1996) Frontier econometric software. Alternatively, Zhus (2003) DEA Excel Bank efficiency
Solver algorithm is utilized to compute DEA technical efficiency scores. assessments
3.3 Procedures for cross-examination of efficiency
The efficiency scores from these four different measurement models (summarized in
Table II) are used in the cross-examination exercise. The aim is to test whether the
different efficiency measurement specifications: 35
(1) rank the Indian banks approximately in the same order (H1);
(2) identify mostly the same Indian banks as best-practice and as
worst-practice (H2); and
(3) are consistent with standard non-frontier performance measures (H3).
H1 is tested using Spearmans rank-order correlation coefficients for the efficiency scores
generated by four different assessment specifications. The nonparametric Spearmans
Downloaded by University of Ghana At 15:50 27 June 2016 (PT)
Model Description
SFA-Translog As given in equation (3); maximum likelihood estimates are obtained using Coelli
(1996)s frontier econometric procedure
SFA-Fourier As given in equation (6); maximum likelihood estimates are obtained using Coelli
(1996)s frontier econometric procedure Table II.
DEA-CRS Equation (8) minus the convexity constraint, N 10 l 1; estimated using Zhu (2003)s Alternative efficiency
DEA Excel Solver algorithm measurement
DEA-VRS As given in equation (8); estimated using Zhu (2003)s DEA Excel Solver algorithm specifications
SEF efficiency assessment results are widely used in respective decision-making processes
29,1 by regulators, managers and industry consultants, etc.
To that end, Spearmans rank-order correlations between the frontier efficiencies
and standard non-frontier measures are calculated. The corresponding t-statistics are
then used to test H3. Following Bauer et al. (1998), three financial statement based
non-frontier ratios are used in this regard: return on assets (ROA), the negative of the
36 total cost per dollar of total assets (2 TC/TA) and the negative of total cost per dollar of
revenue (2 TC/TR).
4. Results and discussion
The section starts with a summary of the efficiency scores obtained from the four
different specifications. Then, it presents and discusses the empirical results for the
three hypotheses.
The summary of efficiency scores obtained from four models employed is presented in
Table III, Panel A[7]. The mean efficiency scores for the Indian banks vary from 0.7946
for DEA-CRS model to 0.9148 for the SFA-Fourier model. These values closely follow
the median efficiency scores for the four models. The minimum efficiency scores range
from 0.4712 for the DEA-CRS model to 0.6996 for the SFA-Fourier model. The two
parametric models (SFA-Translog and SFA-Fourier) display less score variations then
the two nonparametric models (DEA-CRS and DEA-VRS). Our observations are
compatible with the argument that nonparametric efficiency scores should be lower on
average (showing a greater dispersion) since random error is not eliminated in the
estimation process[8]. These descriptive statistics are broadly in line with those of
Bauer et al. (1998) and Ferrier and Lovell (1990).
Table III, Panel A also shows that the average efficiencies (mean and median) for
nonparametric methods are lower than those for parametric SFA-Translog and
SFA-Fourier methods. This preliminary observation reveals that nonparametric
methods generate relatively low efficiencies for most Indian banks. Similarly, the
parametric methods seem to be mutually consistent with each other yielding relatively
higher efficiency scores for most banks.
in similar studies. Bauer et al. (1998, p. 104) and Berger and Mester (1997, p. 924), for
example, report Spearmans rank-order correlations of 0.98 (or higher) and 0.70,
respectively, among efficiency scores obtained from SFA Translog and Fourier
specifications. This anomaly may be driven by the nature of the sample. For example,
the two studies mentioned above used balanced panel datasets as opposed to our
unbalanced data.
5. Conclusion
This paper contributes to the bank efficiency literature by jointly investigating the
statistical consistency of both DEA technical efficiency scores and SFA cost efficiency
scores. In particular, it extends the literature by simultaneously comparing:
.
DEA technical efficiency scores computed using different scale to returns
assumptions (DEA-CRS vis-a-vis DEA-VRS); and
.
SFA cost efficiency scores obtained employing different functional specifications
(SFA-Translog and SFA-Fourier).
are homogenous. Although this is a less severe issue in the present study (given the one
banking market) than in a cross-market analysis, it can still affect the computed
efficiency estimates if the sample banks exhibit short-term differences in objectives and
strategy due to competitive pressures. Moreover, it is possible that the results may be
specific to Indian banks and may not extend beyond the sample[10]. Our results,
however, are broadly consistent and comparable to findings by Bauer et al. (1998) and
Leong et al. (2003) who focus on developed country banking markets.
With regard to possible extensions to our work, it would be interesting to investigate
whether efficiency scores computed using different methods and specifications exhibit
similar stability over time. Such analyses involve calculating correlations between
n-year-apart efficiencies for each bank for each period covered by the sample.
Regrettably, our unbalanced sample does not allow such scrutiny.
Notes
1. In the empirical literature, SFA and DEA are the most widely used parametric and
nonparametric approaches, respectively, (Brown and Skully, 2003). Bauer et al. (1998)
observe that that nonparametric models (such as DEA) typically account for technical
efficiency while parametric approaches (such as SFA) measure cost efficiency.
2. Excellent overviews are presented in Berger and Humphrey (1997) and Brown and Skully
(2003).
3. See Das and Ghosh (2006) for an excellent review of the Indian bank efficiency literature.
4. An excellent review of the technical details about factor share equations consistent with
Hotellings Lemma and Shephards Lemma can be found in Coelli et al. (1998).
5. Even though ratio of other operating expense to physical capital is commonly used in the
literature as the price of physical capital, it suffers from some limitations. For example,
the amount of reported physical capital depends on accounting choices and whether they are
owned or leased. In addition, banks use relatively small investments in physical capital
compared to other inputs to their production process. We thank an anonymous referee for
this insightful comment.
6. As pointed out by an anonymous referee, the inclusion of risk premiums (by use of interest
expense, in place of deposits) might be desirable to bank efficiency in an overall sense. This
measure, however, is distorted by broader monetary policy decisions and their changes
SEF (in the form of announced base interest rate changes by the Reserve Bank of India) over the
sample period. Hence, the bank-specific deposit base was chosen as an input in the bank
29,1 production process. This approach is consistent with the intermediation approach to bank
production.
7. The issue of estimating a common frontier becomes less of an issue in this paper since we
estimate efficiency scores for each bank and for each year in the sample period. Moreover,
40 the effect of technological change (identified to be important by Tulkens and Eeckhaut
(1995)) is captured (partially, at least) in the way banks have combined their inputs to
generate observed output levels and mixes over the sample period. A similar efficiency
computation approach has been utilised by Perera et al. (2007) in their cross-country study of
120 South Asian banks. We thank an anonymous referee for this insightful comment.
8. The authors would like to thank an anonymous referee for this comment.
9. The authors would like to thank an anonymous referee for this comment.
10. The authors would like to thank an anonymous referee for raising this concern.
Downloaded by University of Ghana At 15:50 27 June 2016 (PT)
References
Altunbas, Y., Liu, M., Molyneux, P. and Seth, R. (2000), Efficiency and risk in Japanese
banking, Journal of Banking & Finance, Vol. 24 No. 10, pp. 1605-28.
Ashton, J.K. and Hardwick, P. (2000), Estimating inefficiencies in banking, Journal of
Interdisciplinary Economics, Vol. 11 No. 1, pp. 1-33.
Ataullah, A.T., Cockerill, H. and Le, H. (2004), Financial liberalization and bank efficiency:
a comparative analysis of India and Pakistan, Applied Economics, Vol. 36 No. 17,
pp. 1915-24.
Battese, G.E. and Coelli, T.J. (1995), Model for technical inefficiency effects in stochastic frontier
production function for panel data, Empirical Economics, Vol. 20 No. 2, pp. 325-32.
Bauer, P.W., Berger, A.N., Ferrier, G.D. and Humphrey, D.B. (1998), Consistency conditions for
regulatory analysis of financial institutions: a comparison of frontier efficiency methods,
Journal of Economics and Business, Vol. 50 No. 2, pp. 85-114.
Berger, A.N. and Humphrey, D.B. (1997), Efficiency of financial institutions: international
survey and directions for future research, European Journal of Operational Research,
Vol. 98 No. 2, pp. 175-212.
Berger, A.N. and Mester, L.J. (1997), Inside the blackbox: what explains the differences in the
efficiencies of financial institutions?, Journal of Banking & Finance, Vol. 21 No. 7,
pp. 895-947.
Bos, J.W.B. and Kool, C.J.M. (2006), Bank efficiency: the role of bank strategy and local market
conditions, Journal of Banking & Finance, Vol. 30 No. 7, pp. 1953-74.
Brown, K. and Skully, M. (2003), International studies in comparative banking: a survey of
recent developments, SSRN Working Paper ID No. 365920, available at: http://papers.
ssrn.com/sol3/papers.cfm?abstract_id365920 (accessed December 10, 2009).
Carvallo, O. and Kasman, A. (2005), Cost efficiency in the Latin American and Caribbean
banking markets, Journal of International Financial Markets, Institutions and Money,
Vol. 15 No. 1, pp. 55-72.
Cavallo, L. and Rossi, S.P.S. (2002), Do environmental variables affect the performance of
European banking systems? A parametric approach using the stochastic frontier
approach, The European Journal of Finance, Vol. 8 Nos 4/5, pp. 123-46.
Christensen, L.R., Jorgenson, D.W. and Lau, L.J. (1973), Transcendental logarithmic production Bank efficiency
frontiers, Review of Economics and Statistics, Vol. 55 No. 1, pp. 28-45.
assessments
Chung, U., Jung, J., Han, S. and Kim, G. (2001), Economies of scale and scope in Koreas banking
Industry: evidence from the Fourier flexible form, Journal of the Korean Economy, Vol. 2
No. 1, pp. 87-111.
Coelli, T. (1996), Frontier Computer Program A Maximum Likelihood Estimator for the
Parameters of Frontier Regression Models, Centre for Efficiency and Productivity 41
Analysis, University of New England, Armidale.
Coelli, T., Rao, D.S.P. and Battese, G.E. (1998), An Introduction to Efficiency and Productivity
Analysis, Kluwer Academic, Boston, MA.
Das, A. and Ghosh, S. (2006), Financial deregulation and efficiency: an empirical analysis of
Indian banks during the post reform period, Review of Financial Economics, Vol. 15 No. 3,
pp. 193-221.
Farrel, M.J. (1957), The measurement of productive efficiency, Journal of the Royal Statistical
Downloaded by University of Ghana At 15:50 27 June 2016 (PT)