Effects of input composition on technical efficiencies of textile industries in Pakistan.
Mahmood, Tariq
This paper studies the technical efficiencies of the textile
manufacturing industries in Pakistan using 5-digit level industry data.
Technical efficiencies are computed by the Data Envelopment Analysis
technique assulning constant as well as variable returns to scale. The
efficiency scores thus obtained are analysed by the TOBIT regression
technique to determine how input composition influences these efficiency
scores. It is found that imported raw material and machinery exercises a
positive effect, whereas non-industrial costs affect technical
efficiencies in a negative way. Electricity does not play its due role
in affecting technical efficiencies.
JEL Classification: C24, D24, L6, O14
Keywords: Technical Efficiency, Data Envelopment Analysis, TOBIT
Analysis, Manufacturing Industries
I. INTRODUCTION
Pakistan is the fourth largest cotton producing country in the
world after China, India and the USA. It is not surprising that
Pakistan's industrialisation began in the 1950s with the textile
industry at its core. Over the years, the textile sector has maintained
its central role in Pakistan's economy. It contributes about 54
percent of the total export earnings of the country, accounts for 46
percent of the total manufacturing sector, and provides employment to 38
percent of the labour force in manufacturing [Pakistan (n.d.)].
Pakistan's textile exports, which were 9.754 billion Dollars in
2009-10, increased to 13.104 billion Dollars in 2010-11, [Pakistan
(n.d.), Table 8.1]. The textile policy (2009-14) targets its exports to
rise to $25 billion by the year 2013-14.
Textile industries have certain peculiarities which make them
especially suitable for a developing country like Pakistan. First, the
raw material used is abundantly available in our agro-based economy.
Second, textile industries are labour intensive, and require relatively
low level of skill from workers. Uneducated/unskilled men and women can
also be employed in these industries. Consequently, these industries
ease the unemployment problem, alleviate poverty, and promote female
empowerment. Third, these industries do not require heavy investment in
plants and machinery, making it easier to enter this business. Fourth,
they provide a wide range of vertical linkages within various subgroups.
Fifth, textiles, especially clothing, both in product material and
design are highly value added. Today textile materials have wide variety
such as nylon, cotton, polyester, silk, and wool. Special combinations
of these materials are used to make high performance clothing and
specialty fabrics. Recent developments in microfiber research have
opened up new horizons for textile industry. These fibres are especially
designed to have desirable attributes of insulation, durability, water
and stain resistance etc. They can perform well even in the most
demanding situations. Due to these reasons their demand is increasing in
areas like sports, military, and industrial clothing.
In view of the importance of the textile sector it would be
necessary to explore the factors that contribute to its performance.
Empirical research indicates that improvement in technical efficiency is
a major contributor to overall factor productivity growth, see e.g.
Wadud (2007).
Technical efficiency measures how optimally a firm (or an industry)
is using inputs to achieve a given level of output. Normally, a frontier
function is estimated to serve as a benchmark against which each firm is
compared to get individual efficiency scores. The firms lying on the
frontier get a score of one while those lying below this frontier get a
score of less than one.
The objective of this paper is to estimate technical efficiency
scores of Pakistani textile manufacturing industries and to analyse the
factors influencing these efficiency scores. The paper contributes to
the empirical literature on technical efficiency of Pakistani textile
industries in two important ways. First, we aim to find technical
efficiency scores for textile industries in particular. Previous studies
have measured technical efficiencies of Pakistani textile industries in
the broader context of overall manufacturing industries. For example,
Din, el al. (2007) estimate technical efficiencies of Pakistani
manufacturing industries. Their production frontier represents all
manufacturing industries. Consequently, their efficiency scores indicate
how a particular industry performs in comparison with all other
manufacturing industries. This paper constructs the production frontier
with reference to textile industries exclusively. Here, efficiency
scores indicate how a particular textile industry performs in comparison
with other textile industries. Second, this paper goes a step further in
exploring the factors which influence these efficiency scores.
From an analytical perspective it would be interesting to observe
how technical efficiency behaves in relation to different input
compositions. Output is almost always positively affected by inputs (up
to certain limits), but how a certain input is used in relation to other
inputs may determine whether technical efficiency has increased or
decreased.
Returns to scale are important in determining technical efficiency
scores. As pointed out by Coelli (1996), in case of constant returns to
scale (CRS) we assume that all decision making units (DMUs) are
operating at the optimal scale. However, factors like imperfect
competition, regulatory requirements and constraints on finance may
cause a DMU to operate at less than the optimal level. This fact favours
the use of variable returns to scale (VRS) model. However, the CRS
approach has its own advantages. The assumption of CRS allows the
comparison between large and small DMUs [Noulas (1997)]. A problem with
the VRS model is that in such models where a few large DMUs are present,
there is the possibility that the frontier will be dominated by these
large DMUs. While in fact these large DMUs may not be efficient [Berg,
et al. (1991)]. With these considerations we use both the CRS and the
VRS assumptions to analyse the data.
The rest of the paper is divided as follows: Section 2 gives a
review of theoretical and empirical literature; data, models, and
variables are discussed in Section 3; results are discussed in Section
4; and finally Section 5 concludes the paper.
2. REVIEW OF LITERATURE
The theory of production frontier provides a suitable framework for
empirical work on technical efficiency. Such work started with Farrell
(1957) who used the concept of frontier production function against
which the performance of productive units could be compared. Following
these early works, many writers tried different techniques to estimate
the production frontier and efficiencies. Broadly, these techniques can
be divided in two major groups:
* Parametric Techniques, and
* Non-Parametric Techniques
Parametric Techniques are based on econometric regression models.
Usually a stochastic production, cost, or profit frontier is used, and
efficiencies are estimated with reference to that frontier. Parametric
techniques require a functional form, and random disturbances are
allowed for in the model. The usual tests of significance can be
performed in these models. Non-parametric techniques, on the other hand,
do not require a functional form. They do not allow for random factors,
and all deviations from the frontier are taken as inefficiencies.
Consequently, inefficiencies in non- parametric techniques are expected
to be higher than those in parametric techniques. Moreover, tests of
significance cannot be performed in non-parametric techniques.
The commonly used parametric efficiency techniques are the
stochastic frontier analysis (SFA), the thick frontier approach (TFA),
and the distribution-flee approach (DFA). Whereas, among non-parametric
techniques, data envelopment analysis (DEA) and free disposal hull (FDH)
are more commonly used. To keep the analysis simple we shall use a
single non-parametric technique viz. DEA assuming both CRS and VRS. The
CRS model is attributed to Charnes, Cooper, and Rhodes (1978), while the
VRS model was proposed by Banker, Charnes, and Cooper (1984) by imposing
an additional convexity constraint to obtain that model.
Once we get technical efficiency scores, the next stage involves
the analysis of the factors which may be influencing these efficiency
scores. The Ordinary Least Square estimation might appear to be the
obvious way. However there is a problem with such estimation; technical
efficiency scores are bounded between zero and one, and Ordinary Least
Squares with such a dependent variable may predict values greater than
one [Coelli, et al., p. 194]. Different techniques have been suggested
to solve this problem. This paper follows the technique used by Bjurek,
et al. (1992), and McCarty and Yaisawarng (1993) who applied a censored
regression model to analyse the technical efficiency scores obtained
through application of the DEA technique.
Censored regression models are designed to estimate linear
relationships between variables when the dependent variable is bounded
by either a minimum value or a maximum value (or both). In the case of
censoring from above the dependent variable lies at or below some
threshold value. Similarly, in the case of censoring from below, values
of dependent variable lie at or above some threshold value. The Tobit
model developed by James Tobin (1958) is employed here to analyse the
factors influencing efficiency scores.
This two-stage approach of efficiency analysis has been widely used
in different areas of empirical research. Oum and Yue (1994) use DEA
efficiency scores with a Tobit model to analyse the effects of
government intervention and subsidisation on the efficiency of railways
systems in 19 OECD countries. Chilingerian (1995) analyses the clinical
efficiency of 36 physicians in a single hospital using DEA and a multi-
factor Tobit analysis. Luoma, et al. (1996) examine the efficiencies of
Finnish health centres by applying DEA and the Tobit model to find out
how the various economic, structural and demographic factors affect
these efficiencies.
During recent years quite a few studies have explored the
performance of textile manufacturing activities. Some of these are
briefly reviewed below.
Murugeshwar (2011) analyses growth in total factor productivity in
Indian textile industry. The study is based upon the data collected by
Annual Survey of Industries (ASI) and published by Central Statistical
Organisation (CSO). There are 6 sub- sectors identified on three and
four-digit classification. Cross-sectional and time series data is used
for the period 1980-2005. The author estimates Malmquist Productivity
Indices, and the break total factor productivity growth in case of
change in technical efficiency and change in technology.
Samad and Patwary (2003) estimate technical efficiencies for the
textile industry of Bangladesh using translog stochastic production
frontier. The study uses panel data for the period from 1988-89 through
1993-94. The data are taken from Census of Manufacturing Industries
(CMI) published by Bangladesh Bureau of Statistics (BBS). The value of
gross output is taken as the dependent variable whereas, total fixed
assets, total number of persons engaged, and the cost of raw material
and packaging are used as independent variables. Woolen textiles, jute
textiles, and carpets and rugs are found to be highly efficient groups
of industries. Cordage, rope and twine, and spooling and thread ball
score least in efficiency ranking. The authors attribute these low
efficiency scores to low level of technology used in the industries.
Wadud (2004) analyses technical efficiency of Australian textile
and clothing firms based on a longitudinal survey covering the period of
1995-1998. The author uses a Cobb Douglas stochastic production frontier
to examine firm level technical efficiencies. Analysis of inefficiency
effects indicates that firms' age, size, capital intensity,
proportion of non-production to total workers and type of legal status
significantly affect technical efficiencies of the firms. In a
subsequent paper [Wadud (2007)], the author decomposes the total factor
productivity growth into changes in technology, changes in technical
efficiency, and scale effects. It has been found that changes in
technical efficiency mostly dominated the overall growth in total factor
productivity in textile and clothing firms.
Din, et al. (2007) estimate technical efficiencies of Pakistani
manufacturing industries using industry level data from Census of
Manufacturing Industries for the years 1995-96 and 2000-01. The
efficiencies of textile industries are estimated in the broader context
of overall manufacturing industries. The study uses stochastic frontier
as well as DEA technique. This technique is used under the assumptions
of CRS and VRS. Results show low technical efficiency scores for the
textile sector. The average efficiency scores for this sector are 0.12
and 0.30 for 1995-96 and 2000-01 respectively under the assumption of
constant returns to scale; whereas, for overall manufacturing industries
these scores turn out to be 0.23 and 0.42 respectively.
Khalil (2011) measures technical efficiency of 45 textile
processing units located in Karachi. The paper uses data from a survey
conducted in 2008. Data envelopment analysis is used to estimate
efficiency scores while taking into account both desirable and
undesirable outputs (polluting factors which need to be reduced to
increase the performance). The results indicate that when undesirable
outputs are included in the model, the number of efficient producers
increases. From this the author concludes that some producers do give
consideration to the reduction in undesirable outputs.
3. DATA AND METHODOLOGY
Data and Variables
The data used in this paper are taken from Census of Manufacturing
Industries (2005-06), published by the Federal Bureau of Statistics (now
Pakistan Bureau of Statistics). Industries are identified at 5-digit
level according to Pakistan Standard Industrial Classification (PSIC),
2007. Twenty-seven industries are included in the analysis. (1) The data
used are briefly described below:
Output
Value added reported in CMI reports does not allow for
non-industrial costs. However, another variable, contribution to GDP,
takes care of industrial as well as non-industrial costs. This
definition of output is adopted as it seems more appropriate in the
context of the present study.
Capital
Capital consists of all fixed assets which are expected to have a
productive life of more than one year, and are in use by the
establishment for the manufacturing activity. These include land,
building, plant and machinery etc.
Labour
Labour includes employees, working proprietors, unpaid family
workers and home workers. Labour data have been adjusted to allow for
number of shifts as reported in CMI.
Raw Materials
As defined in CMI (2005-06): 'Raw-materials include raw and
semi-finished materials, assenlbling parts etc., which are physically
incorporated in the products and by-products made. Chemicals, lubricants
and packing materials which are consumed in the production and spare
parts charged to current operating expenses are included. The raw
material given to other establishments for manufacturing goods (semi-
finished and finished) on behalf of the establishment is included,
whereas raw material supplied by others for manufacturing goods oll
their behalf is excluded. The CMI gives data on imported raw materials
as well as on those domestically produced.
Energy
This input is obtained by adding cost on fuel and cost on
electricity as reported in CMI. Fuel is defined as 'firewood, coal,
charcoal, kerosene oil, petrol, diesel, gas and other such items which
are consumed in generating heat and power.'
Industrial Costs
The CMI includes cost of the raw materials, fuels and electricity
consumed, payments for work done, payments for repairs and maintenance
and the cost of goods purchased for resale in the category of industrial
costs.
Non-Industrial Costs
These consist of payments for transport, insurance, copy
rights/royalties, postage, telephone, fax and internet charges, printing
and stationery, legal and professional services, advertising and selling
services, travelling, etc.
Methodology
A two-stage methodology is used to analyse technical efficiency at
the industry level. In the first stage technical efficiency scores are
obtained using the DEA model. In the 2nd stage the effects of various
variables are analysed through the TOBIT model. The models are briefly
described below:
DEA Model
We use the DEA model to estimate the technical efficiency score
under the CRS and VRS assumptions. It is assumed that the industries try
to maximise output with a given combination of inputs. Under the
assumption of CRS, the following n linear programming problems are
solved to get efficiency score for each industry.
Max [PHI], [[lambda].sup.[PHI]]
s.t.
-[PHI] [y.sub.i] + Y [lambda] [greater than or equal to] 0
[x.sub.i] - X [lambda] [greater than or equal to] 0
[lambda] [greater than or equal to] 0
Where ([PHI]) is a scalar, and [lambda]. is a vector of constants.
X and Y represent input and output matrices for all industries. The
symbols [y.sub.i] and [x.sub.i] represent output and input vectors of
ith industry respectively. The contribution to GDP is used as output.
Five inputs are identified viz, labour, capital, raw materials, energy,
and non-industrial costs. The scalar [PHI] is the largest factor by
which all outputs of industry i can be raised. The reciprocal of [PHI]
is the technical efficiency of the ith industry. (2) It represents the
proportional increase in output that could be achieved by the ith
industry, with inputs being held constant. For VRS, additional convexity
constraint (e [lambda] = 1) is imposed in the model. The VRS model is
written as:
Max [PHI], [[lambda].sup.[PHI]]
s.t.
-[PHI] [y.sub.i] + Y [lambda] [greater than or equal to] 0
[x.sub.i] - X [lambda] [greater than or equal to] 0
[lambda] [greater than or equal to] 0
e[lambda] = 1
Where e is a vector of one.
The convexity constraint ensures that an inefficient industry is
only 'benchmarked' against an industry of a similar size. That
is, the projected point for that industry on the DEA frontier is a
convex combination of observed industries [Coelli (2005), p. 172].
These models can be computed by running a linear programme for each
industry. This study uses the computer programme DEAP developed by
Coelli (1996) to compute technical efficiency scores.
Tobit Model
Since technical efficiency scores are restricted by an upper and
lower limit, viz. zero and one, but are continuous between the two
limits, the two-limit Tobit model is used here) Such a model can be
represented in general form by the following equation:
[Z.sub.i.sup.*] = [beta]' w + [[epsilon].sub.i]
Where [z.sub.i.sup.*] is unobserved or latent dependent variable.
Observed DEA efficiency score of ith industry, denoted by [z.sub.i], in
this model are used in place of [z.sub.i.sup.*].
w is a vector of explanatory variables,
[beta] is a vector of parameters to be estimated, and
[[epsilon].sub.i] ~ N(0, [[sigma].sup.2]) is the random term.
We denote lower limit by [L.sub.1], upper limit by [L.sub.2], such
that:
[z.sub.i] = [L.sub.1i] when [z.sub.i.sup.* [less than or equal to]
[L.sub.1i]
[z.sub.i] = [L.sub.2i]; when [z.sub.i.sup.*] [greater than or equal
to] [L.sub.2i]
[z.sub.i] = [z.sub.i.sup.*] when [L.sub.1i] < [Z.sup.*] <
[L.sub.2i]
The model is estimated through the Maximum Likelihood technique.
The likelihood function of this model is given by:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
The technical efficiency scores obtained in the first stage of the
analysis are used as dependent variable in the following empirical
equation.
[z.sub.i] = [[beta].sub.0] + [[beta.sub.1]MachK +
[[beta].sub.2]DimpRm+ [[beta].sub.3]ElecEner + [[beta].sub.4]NicTc +
[u.sub.i]
Where [u.sub.i] is the random term.
The variables used in this regression are explained below:
[z.sub.i] is the dependent variable taking values of the ith
industry's technical efficiency scores obtained from the DEA model.
It may take values between zero and one. However, in actual practice,
the technical efficiency score is never zero. Hence the lower limit in
Tobit estimation is fixed at the minimum value.
MachK is the ratio of value of purchase of plant and machinery to
total value of capital. This variable is used to measure the effect of
new technology in the production process. Other expenditures on capital
like land, building, and furniture and fixtures are also essential for
production process, but these forms of capital are often used in
production activity in indirect ways. New and modern machines are
expected to make efficient use of other inputs like raw material and
labour. Liberman and Johnson (1999) find that investment in new
equipment by Japanese steel firms led to a higher level of labour
productivity in comparison with U.S. firms. In contrast, Dijk and
Szirmai (2006) find that plants operating under the latest technologies
have lower levels of efficiency than mills operating under outdated
equipment in the Indonesian pulp and paper industry. But, such behaviour
is not likely to occur at industry level. So, we may reasonably expect
that this variable will take positive sign in the regression.
DimpRM is the dummy variable used to capture the effect of imported
raw material in the production process. The variable takes the value of
one if imported raw material is used, zero otherwise. The sign of this
variable is an empirical matter. One might expect that imported raw
material, being of better quality, would positively affect technical
efficiency. Mazumdar, Rajeev, and Ray (2009) find positive effect of
imported raw material on efficiency of Indian pharmaceutical firms.
However, if the imported raw material happens to be of low quality, or
it does not quite suit domestic technology, then its effect on technical
efficiency might be negative.
ElecEner is the proportion of cost of electricity to total energy
cost used in the industry. Electricity is usually considered a better
option than other sources of fuel. This source of energy is highly
flexible and convenient. Literature indicates that electricity-intensive
technologies have been replacing other energy-intensive technologies
(which rely on fossil fuels to a greater extent) in manufacturing [Donas
and Dunne (1995)]. A higher proportion of electricity used is expected
to influence efficiency in a positive way. However in Pakistan economy,
due to shortage of electricity, this important input may not be able to
play its due role. Frequent power failures in electric supply and
'load shedding' may result in disruptions in production
process, and may even force industrial users to seek other relatively
inefficient sources of energy. The sign and significance of this
variable may, therefore, be different from what the theory suggests. In
other words, the proportion of electricity in total energy used by the
industry indicates the level of dependence on electricity. When the
supply of electricity becomes unreliable, the industries which depend
more on electricity suffer more. This implies possibility of negative
relationship between the proportion of electricity in total energy use
and efficiency scores.
NicTc is the proportion of non-industrial costs to total costs
(industrial and nonindustrial). As described above, CMI includes costs
like payments for transport, insurance, copy rights/royalties, postage,
telephone, fax and internet charges, printing and stationery, legal and
professional services, advertising and selling services, travelling in
the category if non-industrial costs. However, other costs like
corruption, bureaucratic hassles, litigation, and dispute settlements
might also be contributing to this type of cost. All these things are
expected to cause hurdles in smooth functioning of a business. So, we
might expect this variable to take a negative sign. All variables used
in Tobit regression are in the natural logarithmic form. The computer
package STATA is used to run the Tobit model.
4. RESULTS
DEA Model
Technical efficiency scores from DEA models are reported in Table
1. The scores obtained through VRS are slightly higher than those
through CRS model. This is due to the fact that the envelop obtained
through tile VRS model encloses the data in a more compact way than that
from the CRS model. Consequently more observations are likely to lie on
or near tile frontier. The average technical efficiency turns out to be
0.73 in case of tile CRS model and 0.81 in tile VRS model. These
averages are much higher than those reported by Din, et al. (2007).
Further comparison shows that efficiency scores for individual
industries are also, in general, higher in present study. The reason for
this discrepancy is that the mentioned study constructs the production
frontiers for the whole manufacturing sector, and the technical
efficiencies of textile industries are computed with reference to these
general frontiers. In the present study the frontiers are constructed
for tile textile industries only, and technical efficiency scores are
computed with reference to these specific frontiers.
Individual efficiency scores (Table I) indicate that Cotton
Fabrics, Printing Services of Fabrics, Made-up Textile Articles for
Household, Cordage, Rope, Twine and Netting, Embroidery and Zari Work by
Hand, Knitted and Crocheted Fabrics, and Other Textiles n.e.c, are the
most efficient industries. Among the least efficient industries are:
Carpets and Rugs (other than by hand), Processing of Textile Waste,
Knitted/Crocheted Synthetic Articles, and Other Textile Finishing n.e.c.
There may be a number of causes of these differences in efficiency
scores. Unfortunately the CMI data is not detailed enough to undertake
an exhaustive analysis of the factors influencing technical efficiencies
of all textile industries. The present study limits itself to analysis
of tile effect of input proportions on efficiency scores; i.e., to
explore what type of input proportions are beneficial or detrimental to
the efficiencies of textile industries. In the following pages we try to
tackle this issue through Tobit analysis. These efficiency scores are
quite high in comparison with Din, et al. (2007). As mentioned
previously, Din, et al. (2007) estimate technical efficiencies of
Pakistani manufacturing industries. Their production frontier represents
all manufacturing industries. Consequently, their efficiency scores
indicate how a particular industry performs in comparison with all other
manufacturing industries. This paper constructs the production frontier
with reference to textile industries. Here, the efficiency scores
indicate how a particular textile industry performs in comparison with
other textile industries. Due to fewer variations in the nature of
industries, the production points do not lie very far front the
frontier. Therefore, these efficiency scores are relatively higher.
Tobit Results
The results of Tobit regressions are reported in Table 2 and Table
3. Table 2 shows the results when DEA scores are obtained under the
assumption of constant returns to scale. The Likelihood Ratio (LR)
Chi-Square test is conducted to check the null hypothesis that all
predictors' regression coefficients are equal to zero. The number
in the parentheses indicates the degrees of freedom of the Chi-Square
distribution used to test the LR Chi-Square statistic and is defined by
the number of coefficients in the model. The null hypothesis is rejected
at 0.0242 and 0.0009 levels of significance for CRS and VRS cases
respectively. This leads us to conclude that at least one of the
regression coefficients in both models is not equal to zero. As argued
by Coelli (1996), in CRS we assume that all decision making units are
operating at optimal scale. However, there are many factors like
imperfect competition, and constraints on finance that may cause a
decision making unit to operate at less than optimal level.
This fact may explain the weak results of the CRS model. The
magnitude of Pseudo [R.sup.2] also indicates that the VRS model better
explains the variations in efficiency scores across industries.
The effect of expenditure on machinery and equipment is positive
and significant in case of VRS (Table 3). This is in line with Liberman
and Johnson (1999) who find that investment in new equipment by Japanese
steel firms led to a higher level of labour productivity in comparison
with U.S. firms. The sign of the dummy variable for imported raw
material is positive and significant for both CRS and VRS indicating
serious issues regarding availability of high quality raw material in
domestic market. As mentioned above, Mazumdar, Rajeev, and Ray (2009)
also find positive effect of imported raw material on efficiency of
Indian pharmaceutical firms.
The proportion of electricity in total energy used has no
significant effect on technical efficiency in case of CRS as well as
VRS. The sign also turns out to be ambiguous; positive in CRS and
negative in VRS. These results indicate that electricity as an efficient
form of energy is not playing its due role in our textile industries. In
recent years shortages in power supply have adversely affected almost
all sectors of the economy. Textile industries are especially hurt due
to two reasons. First, they heavily rely on electricity, and second most
of them being small scale units find it difficult to produce their own
electricity at an affordable price.
The effect of non-industrial costs is also found to be negative.
This is probably due to the factors mentioned above viz. corruption,
bureaucratic hassles, litigation, and dispute settlements which are
contributing to efficiency losses.
The size, sign and significance of the intercept indicate missing
factor(s) influencing technical efficiency in a negative way.
Unfortunately data on many inputs in the CMI is not detailed enough to
include all possible factors. Information on education of entrepreneurs,
technical skills of workers, working environment of the factories,
labour-management relationships, and grievance resolution procedures are
some of the issues about which information is crucial to pinpoint the
sources of inefficiencies.
Despite these issues, it must be pointed out that in the
complications of the actual world, no regression call provide an
exhaustive list of variables affecting technical efficiencies. In fact,
studies with significant intercept terms are quite common in the
literature on determinants of technical efficiency, see for example,
Mazumdar, et al. (2009), Wouterse (2008) etc. One of the objectives of
this paper is to analyse tile effect of input composition on technical
efficiencies, and in this regard the exercise is useful.
Like other businesses in Pakistan, textile industries are mostly
family-owned enterprises. As pointed by Gani and Asbraf (2005),
"The business groups in Pakistan (previously known as twenty-two
families) are informal combinations of legally independent business
entities run by families. The family patriarch is the dominant
shareholder and manager whereas the immediate and distant family-members
help operate various firms within the business group". Obviously,
when boards of directors and other management structures are riddled
with nepotism, efficiency becomes a low priority issue.
5. SUMMARY AND CONCLUSIONS
In this paper we have examined technical efficiencies of textile
manufacturing industries in Pakistan using 5-digit level industry data.
Technical efficiencies are computed by Data Envelopment Analysis
technique under the assumption of constant returns to scale as well as
variable returns to scale. The efficiency scores thus obtained are
analysed by Tobit regression technique to determine the factors which
influence these efficiency scores. DEA results show that Cotton Fabrics,
Printing Services of Fabrics, Made-up Textile Articles for Household,
Cordage, Rope, Twine and Netting, Embroidery and Zari Work by Hand,
Knitted and Crocheted Fabrics, and other Textiles are the most efficient
industries; whereas, Carpets and Rugs (other than by band), Weaving of
Fabrics on Khadi/Handloom, Processing of Textile Waste,
Knitted/Crocheted Synthetic Articles, and Other Textile Finishing n.e.c,
tuna out to be the least efficient industries.
In the Tobit model the proportion of machinery in total capital and
dummy for imported raw material are found to have positive effect on
technical efficiencies, while non-industrial costs as a proportion of
total cost have a negative effect. The proportion of electricity to
total energy does not seem to play any significant role.
The issue of raw material needs both short-run as well long-run
strategies. First, import restrictions on raw material used in textile
industries should be removed as a short-run solution. Second, as a
long-term strategy domestic production of such raw material should be
encouraged through research and development, technology diffusion, and
human resource development. Similar policy measures are recommended for
machinery and equipment. The shortage of electricity needs urgent
measures. Cheap and reliable supply of electricity is necessary for the
survival of our textile industry in present day environment of openness
and competition. Eradication of corruption and better governance,
especially simplification of bureaucratic and legal procedures, will
definitely contribute to efficiency in a positive way.
REFERENCES
Banker, R. D., A. Charnes, and W. W. Cooper (1984) Some Models for
Estimating Technical and Scale Inefficiencies in Data Envelopment
Analysis. Management Science, 30:9&10, 78-92.
Berg, S. A., F. Forsund, and E. Jansen (1991) Technical Efficiency
of Norwegian Banks: A Nonparametric Approach to Efficiency Measurement.
Journal of Productivity Analysis 2, 127-143.
Bjurek, H., U. Kjulin, and B. Gustafson (1992) Efficiency,
Productivity and Determinants of Inefficiency at Public Day-Care Centres
in Sweden. Scandinavian Journal of Economics 94, (Supplement): 173-187.
Charnes, A., W. W. Cooper, and E. Rhodes (1978) Measuring the
Efficiency of Decision-Making Units. European Journal of Operational
Research 2, 429-444.
Chilingerian, J. A. (1995) Evaluating Physician Efficiency in
Hospitals: A Multivariate Analysis of Best Practices. European Journal
of Operational Research 80, 548- 574.
Coelli, T. J. (1996) A Guide to DEAP Version 2.1: A Data
Envelopment Analysis (Computer) Programme. Department of Econometrics,
University of New England, Armidale, Australia. (CEPA Working Paper
96/08).
Coelli, T. J., Parsada D. S. Rao, C. J. O'Donnell, and G.
Battese (2005) An Introduction to Efficiency and Productivity Analysis.
2nd Edition. New York: Springer.
Dijk, Michiel Van and Adam Szirmai (2006) Technical Efficiency and
Embodied Technical Change in the Indonesian Pulp and Paper Industry.
Journal of International Development 18:2, 163-178.
Din, Musleh-ud, E. Ghani, and T. Mahmood (2007) Technical
Efficiency of Pakistan's Manufacturing Sector: A Stochastic
Frontier and Data Envelopment Analysis. The Pakistan Development Review
46:1, 1-28.
Doms, M. E. and T. Dunne (1995) Energy Intensity, Electricity
Consumption, and Advanced Manufacturing Technology Usage. Technological
Forecasting and Social Change 49, 297-310.
Gani, W. I. and J. Ashraf (2005) Corporate Governance, Business
Group Affiliation, and Firm Performance: Descriptive Evidence from
Pakistan. Lahore University of Management Sciences. (CMER Working Paper
No. 05-35).
Khalil, S. (2012) Relative Efficiency of Decision Making Units
Producing both Desirable and Undesirable Outputs: A Case Study of
Textile Processing Units in Pakistan. The Pakistan Development Review
51:4.
Liberman, M. and D. Johnson (1999) Comparative Productivity of
Japanese and U.S. Steel Producers 1958-1993. Japan and World Economy 11,
1-27.
Luoma, K., M. L. Jarvio, I. Suoniemi, and R. T. Hjerppe (1996)
Financial Incentives and Productive Efficiency in Finnish Health
Centres. Health Economics 5, 435-445.
Mazumdar, M., M. Rajeev, and Subhash C. Ray (2009) Output and Input
Efficiency of Manufacturing Firms in India: A Case of the Indian
Pharmaceutical Sector. The Institute for Social and Economic Change,
Bangalore. (Working Paper 219).
McCarty, A. Therese and Suthathip Yaisawarng (1993) Technical
Efficiency in New Jersey School Districts. In H. O. Fried, Knox C. A.
Lovell, and Shelton S. Schmidt (eds.) The Measurement of Productive
Efficiency: Techniques and Applications. Oxford University Press.
Noulas, A. G. (1997) Productivity Growth in the Hellenic Banking
Industry: State versus Private Banks. Applied Financial Economics 7,
223-228.
Oum, T. H. and C. Yue (1994) Economic Efficiency of Railways and
Implications for Public Policy: A Comparative of the OECD
Countries' Railways. Journal of Transport Economics and Policy
28:2, 121-138.
Pakistan, Government of (n.d.) Census of Manufacturing Industries
(2005-06). lslamabad: Federal Bureau of Statistics.
Pakistan, Government of (n.d.) Economic Survey (2011-12).
Islamabad: Economic Advisor's Wing, Ministry of Finance.
Pakistan, Government of (n.d.) Strategic Trade Policy, Framework
(2009-12). Islamabad: Ministry of Commerce.
Ray, Subhash C. and Zhang Ping (2001) Technical Efficiency of State
Owned Enterprises in China (1980-1989): An Assessment of the Economic
Reforms. Department of Economics, University of Connecticut. (Economics
Working Paper 2001-05). downloaded from:
http://digitalcommons.uconn.edu/econ_wpapers/200105/
Rosett, R. N. and F. D. Nelson (1975) Estimation of the Two Limit
Probit Regression Model. Econometrica 43:1, 141-146.
Samad Q. A. and F. K. Patwary (2003) Technical Efficiency in the
Textile Industry of Bangladesh: an Application of Frontier Production
Function. Information and Management Sciences 14:1, 19-30.
Tobin, James (1958)Estimation of Relationships for Limited
Dependent Variables. Econometrica (The Econometric Society) 26:1, 24-36.
Wadud, Mokhtarul (2004) Technical Efficiency in Australian Textile
and Clothing Firms: Evidence from the Business Longitudinal Survey.
Australian Economic Papers 43:3, 357-378.
Wadud, Mokhtarul (2007) Sources of Productivity Growth in
Australian Textile and Clothing Firms. Australian Economic Papers 46:3,
254-281.
Wouterse, Fleur S. (2008) Migration and Technical Efficiency in
Cereal Production Evidence from Burkina Faso. (IFPRI Discussion Paper
00815).
Tariq Mahmood <
[email protected]> is Senior Research
Economist at the Pakistan Institute of Development Economics, lslamabad.
(1) CMI reports 28 textile industries at 5-digits level. One
industry, viz., Carpets and rugs (hand made) turned out to be an outlier
in preliminary estimation, so it was excluded from the analysis.
(2) For detail see Coelli, et al. (2005), p. 180.
(3) For details on two-limit TOBIT model, see Rosett and Nelson
(1975).
Table 1
Efficiency Scores of Textile Industries
Industries CRS VRS
1 Spinning of Natural Textile Fibres 1 1
2 Spinning of Man-made Staple Fibres 0.823 1
3 Textile Yarn and Thread of Natural Fibres 0.762 1
4 Text. Yarn and Thread of Man-made Staple Fibres 0.57 0.783
5 Processing of Textile Waste 0.38 0.387
6 Fabrics Other than Cotton 0.869 0.923
7 Cotton Fabrics 1 1
8 Fabrics of Man-made Filaments 0.716 0.781
9 Pile Fabrics, Terry Towelling etc. 0.777 0.81
10 Weaving of Fabrics on Khadi /Handloom 0.398 1
11 Finishing of Textile Fibres and Yarn 0.703 0.827
12 Bleaching and Dyeing of Fabrics 0.569 0.612
13 Printing Services of Fabrics 1 1
14 Finishing of Textiles (Khadi/Handloom) 0.97 1
15 Other Textile Finishing n.e.c. 0.242 0.468
16 Made-up Textile Articles for Household 1 1
17 Other Made-up Textile Articles 0.562 0.574
18 Carpets and Rugs (other than by hand) 0.497 0.513
19 Cordage, Rope, Twine and Netting 1 1
20 Embroidery and Zari Work by Hand 1 1
21 Narrow Woven Fabrics and Embroidery 0.588 0.604
22 Other Textiles n.e.c. 1 1
23 Knitted and Crocheted Fabrics 1 1
24 Knitted/Crocheted Cotton Text. Articles 0.553 0.555
25 Knitted/Crocheted Woollen Text. Articles 0.722 0.742
26 Knitted/Crocheted Synthetic Articles 0.356 0.605
27 Knitted/Crocheted Articles n.e.c. 0.703 0.736
Table 2
Tobit Regression Results for Constant Returns to Scale
Log likelihood = -15.70
LR [Chi.sup.2](5) = 11.22
Prob> [Chi.sup.2] = 0.0242
Pseudo [R.sup.2] = 0.2632
Standard P > [absolute 95% Confidence
crste Coeff error t-values value of t] Interval
Constant -1.48 0.56 -2.63 0.016 -2.66 -0.31
MachK 0.09 0.09 1.03 0.313 -0.91 0.27
Dimprm 0.48 0.26 1.88 0.074 -0.05 1.01
ElectEn 0.26 0.17 1.52 0.143 0.10 0.61
NicTc -0.47 0.20 -2.35 0.029 -0.89 -0.05
Table 3
Tobit Regression Results for Variable Returns to Scale
Log likelihood = -9.14
LR [Chi.sup.2] (5) = 18.63
Prob> [Chi.sup.2] = 0.0009
Pseudo [R.sup.2] = 0.5047
Standard P > [absolute 95% Confidence
vrste Coeff error t-values value of t] Interval
Constant -1.84 0.51 3.58 0.002 -2.90 -0.76
MachK 0.11 0.06 1.84 0.079 -0.01 0.24
Dimprm 0.65 0.19 3.40 0.003 0.25 1.05
ElectEner -0.02 0.12 -0.21 0.837 -0.27 0.23
NicTc -0.54 0.18 -2.98 0.007 -0.92 -0.16
