Productivity analysis for Vietnam''s textile and garment industry

This paper uses firm-level panel data to study the development and determinants of technical efficiency and productivity in the textile and garment sector in Vietnam during the period 1997-2000.

PRODUCTIVITY ANALYSIS FOR VIETNAM’S TEXTILE

AND GARMENT INDUSTRY

Nguyen Thang

To Trung Thanh

Vu Hoang Dat Remco H Oostendorp

This paper uses firm-level panel data to study the development and determinants of

technical efficiency and productivity in the textile and garment sector in Vietnam

during the period 1997-2000 Applying the methodology of Battese and Coelli

(1995), we find that average technical efficiency of the textile and garment sectors

is relatively high, but that technical efficiency differs significantly across

ownership, location, size, age, and export orientation The most productive firms in

the textile sector are old, mid-sized, private, South-based, export-oriented firms,

while in the garment sector old, medium to large, private or foreign invested,

South-based firms have the highest technical efficiency In the garment sector TFP

has increased over the study period, reflecting changes in technical efficiency as

well as technical progress

Acknowledgements

This paper was written based on a competitiveness study funded by the International Development Research Center of Canada as part of the Vietnam Economic Research Network We like to thank Bernard Decaluwé and John Cockburn for their valuable assistance in the course of the study, and we are also grateful to the seminar participants at the Hue, Halong Bay and Hanoi workshops for their helpful discussions and comments on earlier versions of the paper Any remaining errors remain our own

PART I INTRODUCTION

The 90s decade witnessed the impressive performance of Vietnam Textile and Garment industry Quickly recovered from great difficulties caused by the collapse of the former Soviet Union and Eastern European socialist countries and the resulted loss of CMEA market, Vietnam T&G industry managed to develop with the average growth rate of 10.6% per annum and make up 7.5% of the overall GDP during past years (GSO, 2003) The sector has consistently featured in the list of top exporting industries In 2004, the export value of the T&G industry reached approximately US$4.3 billion and the sector thus became the second largest earner of badly needed foreign exchange in Vietnam (Vietnam Economic Times, 2005) Of equal to importance, the industry is highly labor intensive by nature, generating the largest number of jobs among manufacturing industries and sharing 22.2% of total employment in the manufacturing sector (VLSS 97/98) T&G sector has therefore been regarded as an important and strategic industry in solving the acute problem of unemployment and poverty, by exploiting Vietnam’s comparative advantage in labor intensive production Although these unprecedented achievements are undeniable, the T&G industry is not without problems regarding productivity which could threaten the sustainable development of the sector This has naturally aroused interest of researchers in doing careful studies on efficiency and productivity, and on this basis, making well-founded policy recommendations towards the improvements in the industry’s performance

This research is made in the effort to continue and upgrade the previous VEEM report in productivity analysis for T&G industry, which bears some shortcomings First, the use of total wage bill as measure of labor input may cause problem of identity and non-absoluteness

of controlling the difference in both production function and technical inefficiency effects may lead to bias in estimate results Second, assumption of all firms’s minimizing costs and maximizing profit which is inherent in the Tornqvist index number approach applied may be

a constraint to the analysis Third, the analysis does not take advantage of properties of unbalanced panel data available Fourth, the results of TFP growth and TP could be considerably improved if better deflators can be calculated The methodology applied in this study will partly improve the productivity analysis through obtaining better deflators and using stochastic frontier method to directly calculate Malmquist TFP index and its components

To acquire above objectives, the rest of the study is organized as following: Part II briefly describes the methodology that will be used for the subsequent empirical analysis Based on the regression result and estimates of indices, Part III gives a detail picture on technical efficiency performance of the textile and garment sub-sectors and determinants of technical efficiency And then Part IV will give the whole content growth pattern of TFP over the

years The final part presents main findings and suggestions for further research

PART II METHODOLOGY

This section presents the methodological framework that will be used for the subsequent empirical analysis A brief discussion of sources of output growth, which include input growth and TFP change, will be presented in the beginning Next, the section mentions the Malmquist TFP index that will be applied in the study It then discusses the stochastic frontier method to estimate Malmquist technical efficiency change, technical progress change and TFP growth Some other methodological issues will be also shown in this section

2 1 Source of Output Growth and Mamlquist Index

On the supply side, any output growth is determined by the expansion of productive resources and by the improvement in their use (Micko, M and John, M.P, 1991) The latter is expressed through TFP concept, which measures joint productivity of all inputs used in combination to produce certain goods and/or services TFP change, as mentioned earlier, can

be in turn broken down into technical efficiency change and technical progress The concept

of TFP is closely related to disembodied technological change in that it does not increase the productivity of a particular input but rather that of all input jointly

To have clearer picture on this issue, a production function is specified as:

( ) [ ( ) ] ( )

, t eu tt

Z

F

t

Q = (1)

where Q(t): observed output level at time t

Z(t): the set of inputs used at time t

F[Z(t),t] is potential output at best practice level at time t

eu(t) is the level of technical efficiency with u(t)<=0

By taking derivative with respect to time (t) and dividing both sides of the equation by Q(t), (1) can be re-written as

•

•

•

++

F

F Z

By definition, TFP change reflects the variation of output that cannot be explained by changes in inputs, and therefore, it combines technical progress and technical efficiency change, which are captured by the last two terms of (1’)

The following figure helps to visualize how a change in output can be broken down into the above mentioned components In the output-input space, the firm has the production frontiers F1 and F2 for year 1 and year 2 respectively The firm is said to be technically efficient if output is produced at the frontier level, e.g Yl* in period 1 and Y2* in period 2 Otherwise the firm is technically inefficient, for example if realized outputs are Y1 in period 1 and Y2

in period 2, both being lower than the efficient levels, i.e Y1<Yl* and Y2<Y2* In the Figure, technical inefficiencies (or technical efficiency gaps) are abbreviated as TEG1 for period 1 and TEG2 for period 2 respectively The change in technical efficiency over time is the difference between TEG1 and TEG2 Technical progress is measured by the distance

between production frontier F2 and frontier F1, that is, (Yl**- Yl*) at the input level of X1 or (Y2* - Y2**) at the input level of X2

Figure 1 Output Growth Decomposition (Output and Input in Logarithmic Scale)

Therefore the total output growth, (Y2 - Yl), can be decomposed into input growth, technical progress and technical efficiency change, as follows:

Y2 - Yl = [Yl* - Yl] + [Yl**- Yl*] + [Y2 - Yl** ]

= [Yl* - Yl] + [Yl** - Yl*] + [Y2 - Yl** ] + [Y2*- Y2* ]

= [Yl* - Yl] + [Yl** - Yl*] - [Y2*- Y2] + [Y2*- Y1** ]

= {[Yl* - Yl] - [Y2*- Y2]} + [Yl**- Yl*] + [Y2*- Yl** ]

Thus:

Y2 - Yl = -{TEG2 – TEG1} + TC + ∆YX (1’’)

where

Y2 - Yl = Output growth

-(TEG1 - TEG2 ) = Output growth due to due to technical efficiency change

TC = Output growth due to technical progress

∆YX = Output growth due to input growth

The sum of the first two terms in (1’’) is TFP change In other words, TFP change consists of change in technical efficiency and technical progress In the short run, TFP change may be induced by improvement in technical efficiency, which allows firms to increase outputs from

a fixed set of inputs under a given technology Technical efficiency thus provides a measure

of TFP gap for an individual firm relative to the production frontier, which describes the best available technique In the long run, however, the frontier can itself shift with technical progress, leading to aggregate productivity growth

The distinction between TEC and TP is important in the sense that different policy implications may be made from the same TFP change For example, if TEC effect is substantial and dominates TP effect, policies should be focused on providing incentives and appropriate business environment to induce and enable firms to catch up with the best practice On the other hand, if the majority of firms are located in a close distance from the production frontier, priority should be given to policy measures that encourage technical progress These two sets of policies may be quite different from one another

There are some ways to compute TFP index and its decomposition One commonly used traditional method is Divisia index that is based either on cost shares or on input shares in

total revenue (Sadoulet and Janvry (1995)) However this method requires accepting the restrictive assumptions that all firms are cost minimisers and profit maximisers One way to get rid of this issue is to apply Malmquist TFP index Further more, this index number method can take advantage of (balanced and unbalanced) panel data used in the analysis without price data

The Malmquist productivity indexes were first suggested by Caves, Christensen and Diewert (1982), and furthered developed by Fare et al (1989) This index is defined using the Shephard (1953)’s distance functions that describe multi-input and multi-output production technology without the requirement to specify a behavioral objective (such as cost minimization or profit maximization) Fare et al (1994) specified the Malmquist (output-oriented) TFP change index between period s (the base period) and period t as follows:

mo(ys, xs, yt, xt) = [(dos (yt, xt)/dos (ys, xs)) (dot (yt, xt)/dot (ys, xs))] 1/2 (2)

where dso (yt, xt) represents the distance from the period t observation to the period s technology The productivity index may be rewritten in one equivalent way as following

mo(ys, xs, yt, xt) = [dot (yt, xt)/dos (ys, xs)].[(dos (yt, xt)/dot (yt, xt)) (dos (ys, xs)/dot (ys,

The first term measures the change in the output-oriented measure of technical efficiency between periods s and t (catching up to the frontier), whereas another provides a measure of technological change (innovation) It is the geometric mean of the shift in frontier between periods s and t, evaluated at xt and at xs

2 2 Stochastic Production Frontier Method

To estimate Malmquist indices, there are some alternatives The Malmquist TFP index can be calculated using nonparametric methods However this approach assumes constant returns to scale and does not allow for measurement error and does not provide statistical properties The study will use the stochastic frontier method (econometric method) because it may produce TE change, TP change and TFP change directly and take advantage of unbalanced panel data However, due to the panel is two short and small, the random or fixed effects approach is not chosen

This econometric way is to follow Battese and Coelli’s (1995) stochastic frontier production model Accordingly, the production technology of a firm may be specified as follows:

lnY it = β0 + f(ln(X it ), t, β) + (V it - U it ) (4) i=1, ,N, t=1, ,T (4)

Where

Yit : the production output of the ith firm in the tth time period

Xit : vector of inputs

β : vector of parameters to be estimated

Vit : random variables which are assumed to be i.i.d N(0,σV2) (“white noise”) Uit : one-sided non-negative random variables called technical inefficiency effects, which is assumed to be i.i.d as truncations at zero of the N(mit,σU2) distribution, where:

Zit is a px1 vector of variables which may affect the firm-specific efficiency

δ is a 1x p vector of parameters to be estimated

The equations (4) and (5) are denoted as frontier and inefficiency equations, respectively They can be estimated in ML approach with the help of software FRONTIER 4.1 by Tim Coelli, 1996 Technical efficiency is defined as:

e e

X

e X

Y

v t f

u v t f

it

it it

it

it potential

),,(

ββ

By observing that d0t(xit, yit) = TEit and d0s(xis, yis) = TEis, the efficiency change can be derived as TE change = TEit/TEis, which may be directly compared to the first term in equation (3) Technological change index between adjacent period s and t as expressed by a geometric mean [(1+∂f(xis,s,β)/δs)*(1+∂f(xit,t,β)/δt)]1/2.which may be directly compared to the second term in equation (3) The Malmquist TFP index can be derived as the product of these above two indices

To make the frontier model operational for an econometric analysis, we need an explicit functional form of the production function To get start, the transcendental logarithmic (translog) function is used, which can be considered as the second order approximation of any function However, to get rid of burden of this form under small number of observations

in the dataset, other forms of the production function could be used The explicit form of the translog form of the production function as follows:

as Basu and Fernald (1995) pointed out, is that adopting it may yield misleading results if there is imperfect competition or increasing returns to scale Although this is unlikely to be the case in this study of textile and garment sectors, where competition is fierce, value added function does not allow capturing inefficiency related to the usage of intermediate inputs while output production function does Hence in this study, the revenue production function

is adopted as a more flexible function form

Non-neutrality of technical progress is captured by various inputs being interactive with time (i.e βKt ln(Kit)t + βLt ln(Lit)t +βMt ln(Mit)t), in addition to neutral technical progress effect captured by the sum (βtt + βttt2)

It is potential that the technology and factors affecting firm-specific- technical inefficiency in the textile sub-sector and the garment sub-sector are different thus there are two ways to capture this potential differences: (1) estimating the model using separate data sets of the two sub-sectors, this approach could absolutely capture the potential differences in production technology and the affecting factors but it causes the lost in degree of freedom; (2) using dummy variable for one sub-sector and interacting it with input variables in the production function and technical inefficiency effect variables This approach may save degree of freedom, however this approach leads to an assumption of the same gamma and sigma-squared in the two sub-sectors, which may cause distortions in estimated technical efficiency and technical inefficiency effects In this study, the fist approach is chosen and a rational method of detecting and controlling outliers is used to reduce effects of the low in degree of freedom

It should be noted that the specifications shown in Equation (5) is a general form All parameters in the form should be estimated first and variable deletion using LR tests1 on the

basis of the “rules of thumb” principle will be then carried out in order to arrive at the best model specification that will be represented in the following part Some estimated coefficients of Equation (6) may be insignificant and one may think they should be deleted to save degree of freedom when the LR test of a null hypothesis that they are jointly equal to zero is not rejected at a specific level of significance but this process might cause imbalance

in translog form and could not reflect true production function Thus, only the LR test for deletion of insignificant variables in Equation (5) is carried out in this study

2 3 Data

Data source for the study includes two databases: 1999 and 2001 surveys of textile and garment firms carried out by the Institute of Economics in cooperation with other research institutions The former collected data for two years 1997 and 1998 from 96 firms while the latter gathered data for 1999 and 2000 from 150 firms If the former is merged with the latter, there are initial 207 firms with 492 observations for both the textile and garment sub-sectors

statistic has asymptotic chi-square distribution with degrees of freedom equal to the number of restrictions imposed under the null hypothesis

PART III TECHNICAL EFFICIENCY ANALYSIS

This part will be devoted to analyze the performance of technical efficiency in the textile and garment sub-sectors In the beginning, the methodology and variable description will be mentioned Then general technical efficiency performance of the two sub-sectors is in the focus with discussions on technical efficiency frequency and performance by the sub-sectors Finally, the part will examine some firm-level determinants of technical efficiency performance to have a clearer picture on pattern of technical efficiency in the two sub-sectors

3.1 Variables Description and Data Analysis

As said in the Part II, the study will work with one-stage Battese and Coelli’s econometric approach to estimate separately production technology of the two sub-sectors and then compute firm level technical efficiency, TE change, TP and TFP growth

For the frontier model, the variable of output (Y) is the firm’s total revenue derived from selling various outputs There are three inputs used in production: labour, intermediate and capital inputs Labor inputs (LABOR) are measured by the firms’ number of employees, this measure indeed is an imperfect measure of labor inputs as it can not capture difference in labor quality and labor efforts, which could be partially captured by total wage bill However, the measure of total wage bill as labor input causes the problem of identity and leads to estimations of technical efficiency and technical inefficiency effect variable do not have much economic meaning Consequently, the measure of labor input as number of employees could be regarded as a logical measure of labor input Intermediate inputs (INTER) include costs of raw materials, fuel, electricity, and water Values of these inputs, like output value, are drawn from the financial statement of the firm2 Capital (CAP) as a proxy for capital input

is measured by purchased value of machinery, equipment and buildings net of accumulative depreciation This measure of capital, however, is not the best proxy The different depreciation schemes adopted across firms may result in inaccurate and hardly comparable estimates of true capital stock In addition, this proxy cannot reflect the flow of the service provided, which is the true measure of capital input A much better measure of capital is the replacement value of the capital stock, corrected for capacity utilization (Lundrall and Battese, 1998) This measure can reflect the flow of capital service, yet allows capturing the difference in the quality of capital Unfortunately, data on replacement value and capacity utilization in the survey do not appear to be reliable

Next is the description of variables used in the inefficiency model These explanatory variables are identified on the basis of a survey of the literature (IMPR, 2001a) and the authors They include firm’s characteristics such as size, age, ownership structure, location and targeted markets (export vs domestic sales) etc In the literature, there are a number measures of firm size that are used, namely the number of workers (Pitt and Lee, 1981 and Chen and Tang, 1987), total sales (Haddad and Harrison, 1993), value added (Brada, King and Ying Ma, 1997), intermediate inputs (Coelli and Battese, 1996) In the current study, firms are classified into small, medium and large firm that means the size of firm is measured

2

Here, a word of caution with regards to aggregation of outputs and inputs must be said explicitly Schmidt and

the textile and garment industry, whose products are highly differentiated, this problem may be even more severe Specifically, as long as the aggregation uses the value weights, technical (in)efficiency may encompass output-specific allocative (in)efficiency, which is related to the firm’s choice of output mix Similarly, highly aggregated “material inputs” may mix up technical (in)efficiency and inputs-specific allocative (in)efficiency In short, technical (in)efficiency may not be measured accurately due to the aggregation-related problems

by dummy variables A firm is classified as small one (SIZE1) if it has less than 300 employees; a firm is regarded as medium one (SIZE2) if it has number of employees is between 300 and 1000; and the rest are large one The justification for this specification is that the numbers of employees as firm size is a better proxy which could reflect the labor intensive nature of the sub-sectors and consistent with definition adopted in the sampling process of the 2001 survey, however, if the number of employees was directly employed as firm size, there would be the serious problem of multicollinearity and influential observations As a result, dummy variable for firm size could partially employ advantage of firm size with number of employees and does not significantly suffer from the problems Firm age (AGE) is the number of years from the firm’s establishment year to the reported year In addition to individual effects of firm size and firm age, the impact of interaction between the two factors on technical inefficiency is also important and therefore also investigated in a number of other studies This interaction is thus also adopted in the study With regards to ownership structure, in the survey questionnaire, firms are classified into one

of 6 ownership forms including SOEs of central management, SOEs of local management, joint ventures, 100% foreign-owned firms, cooperatives and private enterprises However, it has been found that cooperatives are fairly similar to domestic private firms and joint ventures are in many ways similar to firms with 100% foreign capital These similar types are merged to increase the degrees of freedom and in this analysis, three types of ownership forms are used to classify firms into state-owned enterprises (SOE), foreign-invested enterprises (FIE) and local private enterprises (PRI) Inefficiency models use SOEs as the reference group and therefore have 2 dummy variables of D_PRI and D_FOR to describe ownership type of the firm With regards to location, the study uses two dummy variables of D_LOC and CENTER to classify 2 firms’ locations of the South and the Center With regards

to targeted market (export vs domestic sales), like the size variable, the literature does not provide a unique definition either Cheng and Tang (1997) suggest that only firms that commit to sell all their products in the world market are defined as export-oriented This means that the international markets still remain important for inward-looking firms Kalirajan (1993) uses the export share threshold of 50% and over to define export orientation

In this study, we also use a dummy variable (D_EXP) with an export share threshold to define whether a firm is export-oriented or inward-looking3 However, instead of 50% threshold, median value of export shares is used to separate out these two types of firms Firms that have export share above the median are defined as export-oriented, and otherwise – inward-looking This definition appears to better reflect the strong export-orientation of Vietnamese textile and garment firms It should be noted, however, that different thresholds based on 50%, mean or median values do not appear to matter much, given the peculiar shape

of distribution of export share, which has a small proportion of firms that are located between these thresholds (see Appendix VII) The next variable in technical inefficiency effect model

is capital structure (CS) which is measured by percents of external capital to total capital The last variable in the inefficiency model is the equipment level (EQUI) which is measured by fixed capital per employee Conclusively, we have explicit form of the technical inefficiency effect model:

EQUI CS

EXP D FOR

D PRI

D

CENTER LOC

D AGE

SIZE

AGE SIZE

AGE SIZE

SIZE

it

*

*_

*_

*_

*

*_

*1

*

12 11

10 9

8

7 6

5

4 3

2 1

0

αα

αα

α

αα

α

αα

αα

α

µ

++

++

+

++

+

++

++

=

(7)

3

Due to the relatively small variation of export shares, and small size of the sample, dummy variable appears to

be better than continuous variable, as it allows saving degrees of freedom while still reasonably well capturing

With respect to sub-sector classification, it is clear that the technology in the textile and garment sub-sectors and factors affecting technical inefficiency are potentially different in the textile firms and the garment firms as mentioned above Nevertheless, an outstanding feature

of the T&G industry is that many textile firms also install garment production line The reason for mixed production has been discussed in IOE (2000) As a consequence, we cannot distinguish textile and garment firms simply on the basis of information on their registration license and thus separate estimation of production function for the textile sub-sector and the garment sub-sector may be misleading A good strategy is to pool both textile and garment firms in one sample and use dummy variable(s) to separate these sub-sectors apart However, the assumption of the sigma-squared and gamma are identical in two sub-sectors under this approach might be violated Thus, one criterion for separating firms into the two sub-sectors

is needed In this study, the threshold of 50% of average production share over the study period to set boundary between textile and garment firms is employed This means that firms whose average share of clothing production out of total production value is 50% or higher are defined as garment (or garment-oriented) firms and firms whose share is below 50% are defined as textile (or textile-oriented) ones This threshold appears to be reasonable for separating textile and garment firms given the peculiar shape of distribution of garment share, which is clearly skewed towards two tails of the distribution (see Appendix VI) In addition, the use of the average value of production allows ensuring consistence in identification of production technology of firms as the study period is short When firms are divided into textile and garment firms, two datasets of the two sub-sectors are separately estimated One problem of small sample size is outliers that one or some observations could significantly influence estimated results Accordingly, reasonable methods of detecting them should be used to enhance robustness of regression results In this study, change in estimated coefficients of variables when one observation is included and excluded is used as a criterion for detecting outliers Consequently, ones are regarded as outliers if they cause changes in at least one of estimate coefficients of variables in production function larger than their standard errors With this approach and the translog form of the production function, there are 27 out

of 140 observations are regarded as outliers in the textile dataset This abnormally large number is likely to reveal that there are problems of data or the translog form is a burden of the small dataset Thus, the Cobb-Douglas form is also employed as an alternative and the number of influential observations under the Cobb-Douglas production function is only two Appendix II lists the identity of the outliers

Cobb-Douglas is likely to be too restrictive to be the reasonable form of the production function of the textile sector; meanwhile the translog form is likely to be a burden of small dataset Accordantly, the study concentrates on analyzing estimate results of the textile sub-sector which do not depend on the forms of the production function Table 1 shows the Maximum Likelihood estimates of the stochastic production frontiers for the textile sub-sector after controlling outliers of the two forms of production function with the help of the software FRONTIER 4.1 It should be noted that all variables are deflated to the base year -

Table 1: Maximum Likelihood Estimates of Coefficients for the textile sub-sector

Cobb-Douglas production function

Variable Translog production functionCoeffi Stand.Err T_statistic Coeffi Stand.Err T_statistic

Mean efficiency = 0.80082315 Mean efficiency = 0.81528668

Log likelihood function = -8.8621 Log likelihood function = 41.901866 (*, **, ***: statistically significant at 10%, 5% and 1% respectively)

Note: L, K, M are denoted for logged values of LABOR, CAP and INTER respectively; SIZE and AGE have been logged as well

(Source: authors basing on regression results)

These results are remarkably different from the results of the initial dataset (see Appendix IV for the estimate results using initial datasets), this fact shows that the influential observations are really problematic in the initial datasets

Generally, in estimate results of the textile sub-sector, at least one of the coefficients (absolute term of square term) of inputs are statistically significant at 10 % level of significance but the coefficients of the capital input in the Cobb-Douglas form which is significant at 10.5 % level This high level of significance of the coefficient of the capital input is probably caused by the problem of measurement error All time related variables are statistically insignificant at any conventional level of significance which may imply the technological change over the study period is not significant This fact is clearly due to the short of study period

To save degrees of freedom in estimating small number of observations, one may test to delete insignificant variables as mentioned above However, the current study would like to use initial results for the textile sub-sector which are better for comparisons between two forms of the production function It could be seen that from the two results, the coefficients of variables of the technical inefficiency effects model under two specifications of the production function are almost the same in terms of signs and significance even the magnitudes are not considerably different except ones for the dummy variable for foreign invested firms The difference in the signs of dummy variable for the foreign invested firms is probably caused by problems in measurements of both input and output sides- these problems shall be analyzed in detail later However, it should be noted that the coefficients in the two specifications are not statistically significant

For the garment sub-sector, things are much easier as there are 352 initial observations and the number of outliers under translog form of the production function is only two that means the translog form is reasonable for describing production technology of garment firms observed in the dataset Table 2 presents the estimate results for the sub-sector As the translog form is reliable for the garment sub-sector, the technique of dropping insignificant variables to save degree of freedom could be applied without losing any properties and be convenient for analyzing In the initial estimate result, the coefficients of firm size and export orientation are not significant, they are tested for dropping In fact, the dropping of those variables does not affect significantly the result (see appendix IV for the estimate of the initial specification for the garment)

At least one of the coefficients of the inputs is statistically significant at 5% level of significance and these coefficients are positive that is consistent with economic theories In addition, the coefficient of the absolute term of time trend is also statistically significant at 5% level of significance, this result shows that the technology of the garment sub-sector have significantly changed in the study period Further more, the negative sign of the coefficients

of interactive variables between labor and intermediate inputs and time trend and positive sign of one for capital input shows the that firms in the sub-sector have changed their production technology into more capital intensive ones, however this tendency is not significant as the study period is short

Table 2: Maximum Likelihood Estimates of Coefficients for the garment sub-sector

Variable Coefficient Stand.Err T_statistic

Log Likelihood function = -93.553923

Note: (*, **, ***: statistically significant at 10%, 5% and 1% respectively)

(Source: authors basing on regression results)

3.2 Hypothesis tests

Hypothesis tests of both the stochastic production function and technical inefficiency effects are summarized in Table 3 The null hypotheses are tested using the LR tests above mentioned The null hypotheses that production technology of the garment is corresponding

to Douglas production function are rejected 5% level of significance Thus, Douglas production function is not an adequate specification for the garment sub-sector under given assumptions of the stochastic production function model The following null hypothesis

Cobb-of no technical inefficiency is rejected at 1% level Cobb-of significant Under this hypothesis, all coefficients including constant term and gamma are jointly equal to zero and an average production function, which could be estimated using OLS procedure, would be an adequate specification of production in the textile and garment sub-sectors To have a true picture of technical inefficiency effects and convenience in analysis, insignificance variables in the estimate results of the garment sub-sector are tested to be dropped, the null hypotheses of

joint insignificance of variables in the technical inefficiency effect model are accepted at 5% level of significance The last null hypothesis is that the remaining explanatory variables in the technical inefficiency effect model are jointly equal to zero is rejected at 1% level of significance

Table 3: Results of hypothesis tests

The Textile Douglas form

-Cobb-The Textile- Translog

form

The Garment Null hypothesis Calculated

value

Critical Value

Calculated value

Critical Value

Calculated value

Critical Value Cobb-Douglas

production function

40.05 18.31

No technical inefficiency 65.09 23.68 66.18 23.68 80.59 21.03 Joint insignificance of

(Source: authors basing on regression results)

3.3 Technical Efficiency Performance

After having estimates of technical efficiency derived from the regression results, the study computes geometric means of all firms in two sub-sectors, it should be noted that textile firms are calculated under both specifications of the production function These outcomes are shown in table 4 There two kinds of mean are considered, the first is the unweighted mean that is simple average value of estimated technical efficiency of firms in the sub-sector, the second is the weighted means with the weight is ratio of firm to total revenue of the firms in the sample The weighted mean may reveal a truer picture of technical efficiency performance of the sub-sectors as it incorporates position of a specific firm in overall production of the sub-sector However, unweighted technical efficiency could be regarded as

an informal indicator of importance of factors affecting technical inefficiency and it is useful

in comparing with estimated results of other works as they often present simple average technical efficiency Moreover, comparisons of the two indicators will reveal some characteristics of relationship between size of firms and technical efficiency

From those indicators, some comments could be made for the textile sub-sector Firstly, there has been an improvement in technical efficiency over the study period except that of year

1998 under translog specification of the production function In detail, the biggest improvement year has been 1999 in both unweighted and weighted indicators This fact is explained by the recovery of Vietnam’s textile and garment industry after Asian Financial Crisis in 1997 The weighted technical efficiency is somewhat higher than unweighted ones but the difference is not significant implying that the size impact on technical efficiency is not significant; this finding is consistent with the insignificant coefficient of dummy variable for small firms and a mixed picture of technical efficiency of medium firms in comparing with large one, which shall be analyzed later in detail Moreover, the gap between weighted and unweighted ones tends to be miniature that means the small firms have better improved their technical efficiency Thirdly, the both weighted and unweighted mean technical efficiency are relatively high in comparing other sectors of Vietnam, for example Minh et al (2005) found that the technical efficiency of Vietnam food processing industry in period of 2000-2003 was 60,47%, Thang (2005) stated the estimate technical efficiency of Vietnam iron and steel industry in period of 2000-2003 was 51% The high and concentrative technical efficiency of the textile sub-sector is explained by the high competitiveness of the textile market as it forces firms to catch with the frontier to exist

Table 4: Technical Efficiency Performance

Unweighted technical efficiency Weighted technical efficiency

Textile- CD

The Textile-Translog

The Garment

The Textile- CD

The Textile-Translog

The Garment

(Source: authors basing on regression results)

The increasing trend in means of technical efficiency is also seen in the garment sub-sector but the unweighted indicator of year 1998 This means that the small firms in 1998 have decreased their technical efficiency; but they has much improved their technical efficiency in the year 1999 and 2000 and small firms have been faster in increasing their technical efficiency The wider difference between the weighted and unweighted indicators in the garment sub-sector in comparing with that of the textile sub-sector shows that the size effect have worked more effectively in the garment sub-sector In addition, the significantly higher

of the weighted is likely to reveal there is a positive relationship between firm size and technical efficiency

One again, the relative high of the average efficiency in the garment sub-sector shows that garment firms are relatively good at leaning each other However, an average garment firm is still from 19.76% to 22.35% below the production frontier It is a room for the firm to improve its technical efficiency

3.4 Determinants of Technical Efficiency

In this section, the study makes use of the regression result to analyze the relationships between firm’s characteristics and technical efficiency For the garment sub-sector, analyses are quite straight forward as there is the reliably unique model and all variables of the technical inefficiency effect model, which have statistically insignificant coefficients have been tested to be dropped While, there are some difficulties in analyzing the textile sub-sector as the use of the two specifications of the production function Thus, beside concentrations on the variables of which the signs, significance and even the magnitudes are similar in the two specifications only some tentative analyses are made for remaining variables in the model with Cobb-Douglas production function

3.4.1 Firm Size and Age

As mentioned earlier, in the inefficiency model, together with variables of age and size, the interactive term for these variables are also included This allows investigating how firm size and age (proxy for experience) interacts to influence the age-efficiency and size-efficiency relationship respectively In the estimate results of the textile sub-sector, the coefficients of the dummy variable for small firms and interaction variable with AGE are both insignificant that means technical efficiency of small textile firms are not significantly different from that

of large firms, ceteris paribus This fact is true only for the average small firm and the large

one and there is no information about relationship between firm size and technical efficiency among small textile firms or large ones

Both coefficients of dummy for medium textile firms and interaction variable between medium textile firms and lnAGE are statistically significant, thus it must be calculated total impact of firm size on mean value of truncated normal distribution of technical inefficient elements to have direction of size effect on technical efficiency, which depend on AGE:

ln

*3559.09192.0Impacts

in comparing with large ones

For the garment sub-sector, both of average small and medium firms have lower technical efficiency in comparing with large one as signs of two coefficients of interactive term are positive and no independent firm size effect This result is somewhat inconsistent with the result of the textile that the technical efficiency of small firms is not significantly different from that of large ones This difference is likely to be explained by the specification of the firm size in the models The equivalent values of the dummy variables are the average of the labor of firms under criteria The average labors of small firms in the textile and garment sub-sectors are 176.25 and 176 respectively These numbers are not significant different from each other But these numbers are large size in turn 2911.475 and 1868.608 for the two sub-sectors or the average large textile firm has considerably larger size than the garment one Given the current level of the managerial skills in Vietnam, it is likely that the optimal firm size under inverted-U- shape relationship between firm size and technical inefficiency in the Vietnam’s textile and garment industry is smaller than the average size of large textile firms Thus there is possible deterioration in technical efficiency of the large textile firms Meanwhile, the smaller average size of the large size garment firms is under the optimal size

or just lightly larger than the optimal and there is no significantly negative effect of the large size firms in the garment sub-sector

As signs of the coefficients of AGE and interactive term between age and small size in the estimate results of the textile sub-sector under two specifications are positive, the AGE has negative impacts on technical efficiency It is somewhat controversial to the theories that the AGE- proxy of experience- often has positive relationship with efficiency However, it should be noted that both the two coefficients are not statistically significant The result of the coefficients of medium size of textile firms are quite consistent with the theories and above mentioned findings that the SIZE2 is taken 1 for medium firm that leads to the sum of AGE coefficient and coefficient SIZE2_AGE are negative, or AGE have positive impacts on technical efficiency of the textile firms

There is a mixed picture of the relationship between firm age and the technical efficiency in the garment sub-sector As all coefficients of AGE and its interactions are statistically significant, the impact of the AGE is measured by the total coefficients of the AGE variable and its interaction For the small firm, the total value of coefficients is positive, that implies the relationship is negative or small-old firms are less technical efficient than small-young firms Meanwhile, the relationship is positive for the medium and large garment firms These results are explained by the fact that the experience does not play importance role in small ones as it play in the medium and large ones In addition, if an old firm is small, it has outdated technology and can not invest to improve its technology and no conditions for rationalizing its production that lead to low technical efficiency

3.4.2 Location

The statistically significant and negative coefficients of the D_LOC in the estimate results for the textile sub-sector in the both specifications of the production function as well as in the

estimate result of the garment sub-sector implies that the technical efficiency of the South- based textile and garment firms is steadily higher than that of the firms on other regions Meanwhile, the coefficients of dummy variables for the Centre-based textile firms are statistically insignificant in both specifications but the coefficient is statistically significant and positive in the estimate result for the garment sub-sector These results mean that the technical efficiency of the textile firms in the centre is not significant different in comparing with the North based ones and the centre-based garment firms is worse in technical efficiency performance than ones of the North The poor business, infrastructure and other production conditions of the central region in comparing with ones of the North have caused the worse performance of the centre-based garment firms and the small numbers of observations of the firms in this region is a rational reason for the insignificance of the estimated coefficients

It is the fact that the South is widely perceived to have better business environment and infrastructures than the North does In addition, with as much as 50% of the total industry outputs (MPI, 1998), geographical concentration is expected to enable South-based firms to enjoy the positive externality and learning effects through taking advantages of other firm’s invention The concentration is also expected to result in fiercer competition among closely located firms, forcing them to improve technical efficiency Thus, the higher technical efficiency of the South-based textile firms is not new Additionally, the difference in average technical efficiency of firms in the two regions is also reinforced by the use of number of employees as labor input Under an assumption of wage bill partially reflecting labor quality, the labor quality of the South-based textile firms is higher than it of their North counterparts

as the wage bill per labor of the South-based firms is higher (see Appendix V) Therefore, it has understated labor inputs of the South based textile firms and overstated technical efficiency of the North-based firms

3.4.3 Ownership

The regression results in the textile sub-sector are somewhat inconsistent between two specifications for dummy variables for the ownership of firms, the coefficients of the dummy variable for the private firms changes from statistical significance in the Cobb-Douglas model into insignificance in the translog model Meanwhile the coefficients of the dummy variable for the foreign invested firm change it signs although both coefficients are statistically insignificant This fact might be explained by the results of the outlier detection procedure with the translog specification of production function Under the used outlier detection procedure, the ratio of observations foreign invested firms and private firm regarded as outliers is higher than average5 This may lead to bias in estimation of the coefficients of these dummy variables In addition, there are two distortions in the measurements of output and inputs of the foreign invested firms which may affect estimate results of the dummy variables for foreign invested firms This leads to a question of possible measurement problems On the inputs side, price distortions against the foreign sector may result in overestimation of input quantities that are proxied by values of inputs as done in this study Indeed, due to the existing dual pricing system in Vietnam, foreign firms have to pay higher dollar indexed prices for a number of infrastructure services such as electricity, water, telecommunication, airfare etc and thus their quantities of inputs used for production are inflated as compared to domestic firms On the output side, the “transfer pricing” phenomenon in the foreign sector, i.e deflating price of final products that are exported to mother enterprises in the countries of origin is quite often reported in the press (for example, The Evening News, 12 July 2002) Transfer pricing also affects costs of imported inputs,

5

The ratio of outliers for all observations and observations of foreign and private firms in the textile sub-sector are 19.26%, 26.67% and 23.68% respectively

which are often overstated by foreign invested firms Therefore, to establish a more reliable ownership-efficiency link, all these distortions should be taken into account and somehow quantified, but it is clearly beyond the scope of this study

The number of outliers in the dataset of the garment sub-sector is only two Thus, the random assumption in sampling still holds with acceptable level of confidence and estimate coefficients hold unbias property The negative signs of coefficients of dummy variables for foreign invested and private firms imply that the technical efficiency of firms of these kinds

of ownership is higher than that of state-owned ones For the case of the foreign invested firms, advances in managerial skills, technology and marketing knowledge are factors which are clear to be positive impacts upon efficiency Another possible explanation for brilliantly measured performance of foreign invested ones is the use of number of employees as labor input The average wage per employee have been much higher in the foreign invested garment firms in comparing with that of private and state owned firms (see APPENDIX V) and if the labor quality is reflected by the wage, thus the labor quality of the foreign invested firms is higher than that of the private and foreign invested ones This leads to the number of employees has been underestimated the labor inputs as the wage could

In this period, the private garment firms have coped with many disadvantages and it could be expected to have lower technical efficiency in comparing with that of the state-owned ones However, there have been several possible reasons for explaining the better performance of the private garment firms in term of technical efficiency Firstly, the dynamic characteristics

of private firms in general and the private textile firms in particular have been partially made

up disadvantages, especially when the business environment has changed significantly in the study Secondly, the existence of family labor which have been more popular in the private garment firms could underestimate labor input of this kind of firms as family labor create more effort (Page, 1984) that result in overestimating efficiency performance of the private firms; The Vietnam private garment firms have often used labor without contract for temporary production (IFC, 2004) and may not report this in the questionnaire, thus this also leads to under-report labor input of the private and over-report efficiency Thirdly, the liberalization of regulations in the 1999-2000 has been most positive impacts on private firms

in general and garment firms in particular

3.4.4 Export Orientation

The dummy variables for export orientation in the estimate results of the textile sub-sector do not hold consistence in significance across the two specifications that the coefficient is negative and statistically significant at 10% level of significance under translog specification

of the production function and negative but statistically insignificant under Cobb-Douglas specification Tentatively, these results imply there is positive impact of export orientation upon the technical efficiency of textile firms However, the small number of observations and change in the ratio of export orientation firms in the dataset after dropping influential observations under translog specification of production function6 are two possible reasons for insignificance of the coefficient in the Cobb-Douglas production function model

The coefficient of the dummy variable for export orientation in the estimate result for the garment sub-sector is positive but statistically insignificant that shows the export orientation does not have significance impact upon technical efficiency In fact, the garment sub-sector is extremely export orientation that the median of export share is 97, 55% and 334 out of 350 observations have export share is larger than 50%, these figure shows that export orientation

6

The ratio of firms are regarded as export orientation ones in initial textile dataset is 50% and the ratio for after dropping outliers under translog form of production function is 46.9%

is not a distinction feature of garment firms As a result, the export orientation dummy variable is statistically insignificant

3.4.5 Capital Structure

The coefficients of the capital structure are negative statistically insignificant in the estimate results of both specifications of the textile sub-sector or relationship between capital structure and technical efficiency is not statistically different from zero But the coefficient of the variable in the garment sub-sector is negative and statistically significant at 5% level of significance This means the capital structure have steadily positive relationship with technical efficiency in the garment sub-sector This difference in relationships between capital structure and technical efficiency in two sub-sectors is caused by their nature The textile sub-sector includes relatively small number of large firms, thus it is easier for them to have external capital when the banking system prefer to lend large firms, this leads to the textile firm themselves do not sufficiently manage the external capital Additionally, the textile sub-sector has relatively greater number of state-owned firms and the weak monitoring mechanism of state-owned banks against state-owned firms has worsened internal management of external capital in the state-owned textile firms These negative elements have naturalized the positive effects of winner-picking principal of lending process of banking system

The garment sub-sector contains a greater number of small firms, the small garment firms themselves are lack of internal capital and they have to cope with others to have external capital The winner-picking principal would result in positive relationship between external capital and technical efficiency

3.4.6 Equipment Level

This firms’ characteristic is not, indeed, well supported in the literature to have relationship with technical efficiency However, attempts to drop it have not been successful In some papers (Minh, 2004, 2005) equipment level is explained as a proxy for embodied technology and it links to ability of worker to internalize the advance technology Others conjecture it as

a representative of labor quality, however, there is no reliable information of labor quality and it is impossible to test this statement Others propose the equipment level as an omitted variable which tranlog form production function ignores Therefore, it needs further information to explain estimated results of this firm’s characteristic

3.5 TFP Growth Analysis

This sub-section shall use the methodology described in the sub-section 2.1 to investigate TFP growth and its components- technical efficiency change and technical progress It should however be noted that the panel datasets from which the relevant rates are derived are heavily unbalanced: the textile panel data set under translog production function consists of 20 observations (firms) for 1997-1998, 7 observations (firms) for 1998-1999 and 26 firms for 1999-2000; these number under Cobb-Douglas production function are 29, 13 and 39 in turn

In the garment sub-sector, the panel data is better that the panel observations for 1998-1997, 1999-1998 and 2000-1999 are 65, 26 and 109 respectively Too small size of the panel data set may result in large measurement errors and hence, unreliable aggregate estimates Therefore aggregates derived from estimates for the textile firms do not appear to be reliable This is also exacerbated by the fact that estimates for textile firms are very sensitive to measurement errors associated with capital inputs, which in turn constitute a relatively large part of the total cost Thus, the results of the subsequent analysis should therefore be interpreted as preliminary and indicative only However, the study still attempt at using some other analyses to use up the information contents and there are two analyses are quite reliable,