9-v54 2 16 PDF
9-v54 2 16 PDF
9-v54 2 16 PDF
I. INTRODUCTION
In today’s world the globe consists of the borderless economy and each
and every entity should be prepared to accept the challenges of this change if
they want to play a major role in businesses and remain competitive. An
*The authors are, respectively, Associate Research Fellow at Punjab Economic Research
Institute / Ph.D. Candidate and Assistant Professor/Chairman (Incharge) at Department of
Economics, University of the Punjab, Lahore (Pakistan).
Corresponding author e-mail: [email protected]
364 Pakistan Economic and Social Review
labor force. The productivity and performance of LSMI sector has much
importance for sustained growth and development of the country as LSMI
comprises more than fifty percent of the industrial sector.
However, it is not sufficient to measure the technical efficiency only
without determining its sources. This study measures efficiency scores of
LSMI and also assesses its determinants as Pakistan is ranked 133 among
148 in global competitiveness index (the Global Competitiveness Report
2013-14; Schwab (2014)). There is no study in Pakistan that has estimated
bias-corrected efficiency scores of manufacturing sector and assessed its
sources. This will be the first study to evaluate the technical efficiency and
its determinants by applying the DEA double bootstrap. In this study
industries are also regrouped for making them comparable.
The remaining of the study is ordered as follows: Review of literature
is given in section 2. Section 3 provides methodological framework and
describes sources of data. Empirical results of manufacturing industries are
interpreted in Section 4. Section 5 consists of conclusions and policy
recommendations.
optimum scale.
Baten et al. (2006) analyzed the technical efficiency of selected
manufacturing industries of Bangladesh by applying the stochastic frontier
production approach over the period from 1981/1982 to 1999/2000. They
covered the selected 3-digit census factories. They included three variables
(value added, capital and labor) in their research. They applied two
alternative distributions to model the: the truncated normal distribution and
the half normal distribution. They found that estimated technical efficiency
for selected industries was 40.22% of potential output under the truncated
normal distribution while it was 55.57% of potential output for the half
normal distribution. They also examined that the time-varying inefficiency
parameter was positive which indicated that the technical efficiency declined
over the mentioned period of time.
Duzakin and Duzakin (2007) examined the performance of 480
manufacturing firms related to 12 industries of Turkey for the year 2003.
Output oriented super slacks based model of data envelopment analysis was
applied under constant returns to scale (CRS) assumption. They used two
inputs (net assets and average number of employees) and three outputs
(profit before taxes, export revenues and gross value added). They found that
standard deviation average scores deviated from 0.178 to 0.989 and they
found that 278 firms remained below average results, and only 65 firms were
identified as efficient. It was examined that the major factor of the
inefficiency in leading firms of Turkey was the lack of insufficient level of
exports.
Watanabe and Tanaka (2007) examined the efficiency of Chinese
industries over the 1994 to 2002 period at province level. They chose the
directional output distance function for estimating the two efficiency
measures of Chinese industries, one was traditional that considered only
desirable output while the other considered desirable and undesirable
outputs. They used capital, labor and materials (coal) as inputs while
industrial products as desirable output and sulfur dioxide as undesirable
output. They found that if the only desirable output is included then the
efficiency level was biased. They concluded that neglecting the undesirable
output tends to lead to an overestimate of industrial efficiency levels in
Shandong, Sichuan, and Hubei. They also found that a province’s industrial
structure has significant effects on its efficiency levels.
Ahmadi and Ahmadi (2012) examined the technical efficiency level of
manufacturing industries in Iran during 2005 to 2007. They included 23
industries from (2-digit ISIC groups) industries and country's provinces,
368 Pakistan Economic and Social Review
III. METHODOLOGY
Measuring the efficiency of producing units was initiated by Farrell (1957).
There is much literature with respect to the Farrell’s (1957) classic definition
of technical efficiency. Basically, there are two common techniques for
measuring efficiency: parametric and non-parametric. Aigner et al. (1977)
and Meeusen and Broeck (1977) developed the parametric approach (SFA)
first. The SFA requires the specification of functional form and estimates the
cost frontier. The incorporation of the stochastic error in model specification
is the main feature of the SFA. So, this approach allows the testing of
hypothesis due to the existence of stochastic error. This approach suffers
from two drawbacks: It requires specification of functional form and
assumptions regarding distribution of the error term.
The analysis of production efficiency is widely based upon the linear
programming models of Charnes et al. (1978) and Fare et al. (1985). The
ground is provided for these methods by Farrell (1957). In that literature,
those techniques which adopted the convexity assumption are famous with
the name of data envelopment analysis (DEA). DEA is a non-parametric
approach and it does not require a priori assumption of functional
specification relating inputs to outputs and does not demand for the
distributional assumption of the error term. DEA creates an efficient frontier
for every observation. The maximum output can be gained empirically from
a given set of inputs.
In DEA technique, it is assumed that decision making units (DMUs)
have full control over the inputs which can be suggested as discretionary
variables. But it is not true in reality as Ouellette and Vierstraete (2004) and
many others have justified that there are non-discretionary inputs in every
370 Pakistan Economic and Social Review
In equation (1) x and y are observed inputs and outputs and i=1….,n is
the specific industry. The is the efficient level of inputs, is the scalar
and is the non-negative vector of constants defining the optimal weights
of inputs to outputs. The obtained value of is the technical efficiency
estimate for the ith industry. In case of input-oriented, inputs should be
decreased for getting the higher technical efficiency where =1 means
that the industry is considered fully efficient while >1 means that the
industry is inefficient and it needs to reduce the inputs for reducing the
MUJADDAD and AHMAD: Measuring Efficiency of Manufacturing Industries 371
inefficiencies.
The bias-corrected efficiencies are estimated in the first stage and are
left truncated by 1. The second stage uses a bootstrapped truncated
regression estimated by maximum likelihood method to estimate
determinants of TE as below:
i= b + z i β + εi (2)
In Eq. (2), b is the constant term, εi is statistical noise, and zi is a vector
of specific variables (these are known as environmental variables) for
industry i that is expected to be related to the industry’s efficiency score.
WHY DOUBLE BOOTSTRAP?
Usually very few results are available for the sampling distribution of
interest. It is for this reason that bootstrap techniques are adopted by Simar
and Wilson (2000, 2007). The concept behind the bootstrapping is very
simple i.e. Simulate the sampling distribution of any specific object by
mimicking the data generating process (DGP). The DGP that gives the logic
for Simar and Wilson’s (2007) double bootstrap is the DEA model
represented by eq. (1) and the second step truncated regression described by
Eq. (2).
To apply the bootstrap procedure, it is assumed that the original sample
data is produced by the DGP and that we can simulate the DGP by using the
‘new’ or pseudo data set that is derived from the actual data set. Then DEA
model is re-estimated by incorporating this new data set. It is possible to
derive an empirical distribution of bootstrapped values by doing this process
again and again which provides a Monte Carlo approximation of the
sampling distribution and also helps out in inference measures. The
efficiency of the bootstrapped methodology and the consistency of the
statistical inference significantly depend on how well it specifies the true
DGP and on the exact re-sampling simulation to copy the DGP.
The Simar and Wilson’s (2007) algorithm 2 of bootstrap procedure is
employed in this study that provides inference about coefficients and consists
of seven steps which are defined in different studies like Barros and Barrio
(2011) and Barros and Assaf (2009) briefly.
1
According to the definition of CMI, “Large Scale manufacturing covers the establishments
registered under factories Act 1934 or qualifying for registration having 10 or more
employees”. We have used data on LSMI according to this definition.
MUJADDAD and AHMAD: Measuring Efficiency of Manufacturing Industries 373
2
Firstly we used the value of fixed assets as the proxy for the size of the industry but now
we used the employment as the proxy for the size of the industry as suggested by reviewer
and it is found that this model gives better results than previous model, so it is included in
main body of the article and results of previous model presented in table A-5 of Appendix.
) For robustness check, we repeated the analysis by using the data of 1995-6 and 2000-01
as per suggested by other reviewer. It has been found that, still there is no efficient
industry in case of bias corrected efficiency scores and the results of second stage of this
model are presented in table A-4 of Appendix. It could be seen that signs and significance
level of this model is same as was the previous model.
374 Pakistan Economic and Social Review
confidence intervals while DEA scores do not lie within the interval because
it underestimates the frontier and it is assumed to touch the frontier before
reaching to the actual one.
As in this study input-oriented DEA Bootstrapped model was used in
the first stage, the efficiency score 1 represents the technically fully efficient
industry while estimated efficiency score of greater than 1 shows the
inefficient or less efficient industry. In case of input-oriented model, fixed
output is obtained by utilizing the different set of inputs. So, for minimizing
the inefficiencies, use of inputs should be reduced for getting the same level
of output. So, it was found that there was not even a single industry fully
efficient for the period of 1995-96, 2000-01 and 2005-06 in case of bias-
corrected TE. The group-wise mean efficiency analysis is shown in table A-2
of appendix. The efficiency analysis of industrial groups shows broad
description of the industrial sector. It can be noticed from this table that
almost every industrial group improved its efficiency. It can also be observed
from table A-3 of appendix that overall TE of LSMI has increased over the
period of time.
After estimating the bias-corrected efficiencies of three cross sections, these
bias-corrected efficiencies of three different time periods were pooled in one
equation as the truncated regression form showed in Eq. (2) and maximum
likelihood method was employed for truncated regression as described in the
second step of the Simar and Wilson’s double bootstrap procedure in the
previous section. Results of determinants of the technical efficiency scores,
standard error, t-statistics and bootstrap confidence interval at 95% i.e.,
lower bound (LB) and upper bound (UB) are presented, respectively in
columns 2 to 6 of table 1 in the text. One should remain careful while
interpreting the coefficients as it can be seen in the study of Keramidou et al.
(2011). Since the efficiency scores are based on the assumption of input-
orientation, therefore, signs of coefficients must be reversed during
interpretation for clarity in interpretation3. It means a positive coefficient
shows the negative impact on efficiency scores and vice versa because we
33
In this study input-oriented approach is utilized which indicates that firms or industries
having a score of 1 are fully efficient while those having efficiency score of greater than 1
are less efficient or inefficient. A coefficient having a positive sign means that the relevant
determinant is inversely related to efficiency. While a coefficient having a negative sign
means that the relevant determinant leads to enhance the efficiency. In order to have
straightforward interpretation we reversed the signs of our regression coefficients during
interpretation.
MUJADDAD and AHMAD: Measuring Efficiency of Manufacturing Industries 375
In the results of second step, where coefficients are bootstrapped 1000 times,
average wage, which represents the employee’s skill, is a highly significant
and possessing a negative sign which shows that it is a favorable source of
technical efficiency. It shows that there is more competence, well-educated
and post related staffing which enhances the ability of labor to perform and it
is participating positively in the efficiency of large scale manufacturing
industries. The coefficient of size of the industry (proxy for economies of
scale) is significant at 5% level and possesses a positive sign which means it
has a negative influence on the efficiency scores. It indicates that it does not
help to promote the efficiency of the manufacturing sector and there is no
evidence of economies of scale in selected industries and not well utilization
of the production capacity. The third and the last coefficient is market size
which possesses the negative sign but that is an insignificant variable means
it is not participating in affecting the efficiency positively more. These
results are found to have the same signs as Keramidou et al. (2011) found in
their study.
VI. CONCLUSION
The purpose of this study was to measure the efficiency of the large
scale manufacturing industries. Performance analysis is one of the main
objectives of the managers of establishments because they want to know that
376 Pakistan Economic and Social Review
how well are their companies working under the given resources.
Performance analysis also helps them to see how well are their past decisions
and how they can bring their establishment to the top position. For
measuring the efficiency, there are many techniques but in this study, DEA
double bootstrap approach was applied because it is more appropriate
technique as compared to the existing approaches. DEA double bootstrap
consists of two steps. In first step, it measures the bias-corrected efficiency
scores while DEA measures the biased efficiencies and it exaggerates the
efficiency scores. It can be seen from this study that DEA efficiency scores
do not lie in the confidence interval and these scores are beyond the interval
due to the bias which exists in DEA scores while bootstrapped efficiency
scores lie within the confidence interval and these are bootstrapped by 2500
iterations.
It was found in this study that none of the industries was technically
fully efficient while industries showed performance over the period of 1995-
96, 2000-01 and 2005-06. It can be seen that the industry which was the most
inefficient in 1995-96 was not the highly inefficient in 2005-06 which means
that there is learning behavior in industries. In this study, firstly efficiency
score of individual and every industry was presented, secondly group-wise
efficiency analysis was presented and thirdly overall TE scores were shown.
In the second step of this technique, the bias-corrected efficiency
scores are used as the dependent variable with left truncation, the
bootstrapped truncated regression model is used because the common
standard regression models are inappropriate. In this study, coefficients are
bootstrapped with 1000 iterations because further iterations did not improve
the results, so 1000 iterations were considered enough in this stage. It was
found that there is no evidence of economies of scale in the manufacturing
sector and the production capacity is not well utilized. The market size does
not have any impact on the efficiencies and the average wage has a negative
sign which implies the positive impact on the efficiency scores and it shows
that labor is fully able to perform in the favor of technical efficiency due to
the well knowledge, more competent and skilled staff.
On the basis of our results, firstly it can be suggested that there is
intense need to establish latest technical universities and institutions for the
guidance of the labor and to equip them with the modern techniques required
by the industry. Secondly industries need to reduce their size as there is
evidence of diseconomies of scale.
MUJADDAD and AHMAD: Measuring Efficiency of Manufacturing Industries 377
REFERENCES
Ahmadi, V., & Ahmadi, A. (2012). Application of data envelopment analysis
in manufacturing industries of Iran. Interdisciplinary Journal of
Contemporary Research in Business, 4(8), 534-544.
Aigner, D., Lovell, C. A. A., & Schmidt, P. (1977). Formulation and
estimation of stochastic frontier production function models. Journal
of Econometrics, 6(1), 21-37.
Banker, R. D., & Morey, R. C. (1986). Efficiency analysis for exogenously
fixed inputs and outputs. Operations Research, 34(4), 513-521.
Barros, C. P., & Assaf, A. (2009). Bootstrapped efficiency measures of oil
blocks in Angola. Energy Policy, 37 (10), 4098-4103.
Barros, C. P., & Garcia-del-Barrio, P. (2011). Productivity drivers and
market dynamics in the Spanish first division football league. Journal
of Productivity Analysis, 35(1), 5-13.
Baten, M. A., Rana, M., Das, S., & Khaleque, M. A. (2006). Technical
efficiency of some selected manufacturing industries in Bangladesh: a
stochastic frontier analysis. Lahore Journal of Economics, 11(2), 23-
41.
Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency
of decision making units. European Journal of Operational
Research, 2(6), 429-444.
CMI (various issues), Government of Pakistan, Census of Manufacturing
Industries, Pakistan Bureau of Statistics, Islamabad.
Din, Musleh ud, Ghani, E., and Mahmood, T. (2007). Technical Efficiency
of Pakistan's Manufacturing Sector: A Stochastic Frontier and Data
Envelopment Analysis. The Pakistan Development Review, 46(1), 1-
18.
Duzakın, E., & Duzakın, H. (2007). Measuring the performance of
manufacturing firms with super slacks based model of data
envelopment analysis: An application of 500 major industrial
enterprises in Turkey. European Journal of Operational Research,
182(3), 1412-1432.
Fare, R., Grosskopf, S., & Lovell, C. K. (1985). The measurement of
efficiency of production (Vol. 6). Springer Science & Business
Media.
378 Pakistan Economic and Social Review
APPENDIX
TABLE A-1
: Comparable Set of Industries:
1995-96 and 2000-01 2005-06
1 Manufacturing of Textiles
1 Cotton spinning Spinning of textiles
2 Cotton weaving Cotton fabrics
3 Woolen textiles Fabrics other than cotton
4 Narrow fabrics Narrow woven fabrics & embroidery
5 Made-up textile goods Made-up textile articles, not apparel
6 Knitting mills Knitted & crocheted fabrics
7 Cordage, rope and twine Cordage, rope, twine & netting
8 Spooling and thread ball making Text. yarn & thread of MM staple fibers
2 Food Manufacturing
9 Dairy products &Ice cream & Ice Dairy Product
10 Canning of fruits & vegetables Fruit and vegetable juices
11 Canning of fish & sea food Fish & fish products
12 Vegetable Ghee & Cotton seed and inedible Refined oils & fats
animal oils
13 Rice milling Rice Husking & Rice milling
14 Wheat & grain milling & Grain milled products Cereal & vegetable flour milling & Cereal grain
and other grain milling products
15 Bread & bakery products Bread, fresh pastry & cake
16 Biscuits Rusks, biscuits
17 Refined sugar Sugar
18 Confectionery, not sweetmeats & Desi” Cocoa, chocolate & sugar confectionery
sweetmeats and confectionery
19 Blending of tea Processing & blending of tea
20 Feeds for animals & Feeds for fowls Animal feeds
21 Starch Starches & starch products
3 Industrial Chemicals
22 Alkalies & Acids, salts & intermediates & Inorganic acids & compound
Sulphuric acid
23 Dyes, colours & pigments Dyes & pigments
24 Compressed gases, etc. Industrial gases
MUJADDAD and AHMAD: Measuring Efficiency of Manufacturing Industries 381
TABLE A-2
Mean Efficiencies of the 22 Industrial Groups
t Group wise mean 1995-96 2000-01 2005-06
No of group DEA dhat.bc DEA dhat.bc DEA dhat.bc
1 Manufacturing of Textiles 6.21 8.66 3.44 4.23 2.08 2.60
2 Food Manufacturing 6.04 7.98 2.97 3.59 2.67 3.32
3 Industrial Chemicals 3.47 4.67 1.81 2.30 2.03 2.5
4 Other Non-metallic Mineral 2.93 4.10 2.41 3.04 1.72 2.18
5 Tobacco Manufacturing 1 1.67 1 1.44 1 1.43
6 Iron and Steel 2.34 3.49 1 1.44 1 1.40
7 Medicines 6.31 9.54 1 1.45 1 1.25
8 Electrical Machinary 2.98 4.16 2.20 2.63 2.51 3.08
9 Transport Equipment 4.99 6.83 3.11 3.91 2.00 2.44
10 Other Chemical Products 2.40 3.05 1.55 1.77 2.36 2.85
11 Non-electrical Machinery 3.59 4.89 1.38 1.73 1.25 1.71
12 Printing and Publishing 3.70 5.02 1.42 1.74 2.37 2.91
13 Petroleum Refining 1 1.51 1 1.41 1 1.44
14 Paper and Paper Products 4.41 6.01 2.62 3.15 1.98 2.43
15 Wearing Apparel 11.20 16.5 4.08 4.89 1.24 1.55
16 Leather and Leather Products 6.65 8.90 1.87 2.20 3.03 3.64
17 Ginning and Baling of Fibre 4.25 5.92 1.05 1.34 1 1.33
18 Rubber Products 3.80 5.19 1.30 1.51 1.49 1.88
19 Glass and Glass Products 5.22 7.30 2.11 2.65 3.42 4.28
20 Non-ferrous Metal Industries 1.85 2.66 1 1.37 2.17 2.58
21 Surgical Instruments 12.76 15.95 5.57 6.96 3.06 3.76
22 Sports and Athletic Goods 15.16 20.73 1.43 1.59 1.65 1.97
384 Pakistan Economic and Social Review
TABLE A-3
Overall T. E. level
Year DEA Bias-Corrected
efficiencies
1995-96 4.827 6.583
2000-01 2.333 2.871
2005-06 2.132 2.650
For robustness check, by using the data of 1995-6 and 2000-01
TABLE A-4
Determinants of VRS TE Scores, Using a Bootstrapped Truncated
Regression
Regressor Bhats S.E t-statistics
constant 9.1716212 8.894823 1.03111898
Average wage -12.898164** 4.215235 -3.0598917
Size 5.5086915** 1.617621 3.40542863
Market size -0.6163304 1.14989 -0.5359907
** indicate significance at 5% level
TABLE A-5
Determinants of VRS TE Scores, Using a Bootstrapped Truncated
Regression
bootstrapped confidence
Regressor Bhats S.E t-statistic interval at 95%
LB UB
constant 24.47** 11.90 2.06 23.60 33.61
Average wage -18.68** 6.37 -2.93 -23.53 -16.66
Size 3.31* 1.71 1.93 2.49 3.87
Market size -1.91 1.67 -1.14 -2.96 -1.49
** and * indicate significance at 5% and 10% levels, respectively