Absorptive Capacity and Foreign Spillovers: A Stochastic Frontier Approach Jojo Jacob* and Bart Los** *Eindhoven University of Technology, ECIS, P.O. Box 513, 5600 MB Eindhoven, The Netherlands; e-mail: [email protected] **University of Groningen, GGDC, P.O. Box 800, 9700 AV Groningen, The Netherlands; e-mail: [email protected]

Forthcoming in K. Frenken (ed.) “Applied Evolutionary Economics and Economic Geography” (Edward Elgar)

ABSTRACT This chapter is about the role that foreign technology and differences in absorptive capacity played in the labour productivity growth performance of the Indonesian manufacturing sector in the period 1985-1996. We propose a Stochastic Frontier Analysis (SFA) approach to study these issues. In our view, SFA has a few nice features that deserve to be explored in this context. First, SFA offers opportunities to study the effects of technology spillovers for best-practice plants and similar plants that attain lower productivity levels simultaneously. Hence, we can allow for heterogeneity with respect to the presence of absorptive capacity. A second interesting feature is that a decomposition framework can be connected to the estimation results, which allows for industry-level analyses of the contributions of “innovation” (led by technology imports), “assimilation” (affected by changes in absorptive capacity) and the “creation of spillover potential”. This last source takes an argument into account that is put forward in theories of “appropriate technology”: firms can assimilate knowledge usefully only if it is relevant, i.e. originating from a firm operating similar production processes. This argument could well have relevance in a dual economy like Indonesia. *A first version of this chapter was presented at the European Meeting on Applied Evolutionary Economics 2005 (Utrecht, The Netherlands). Useful comments by participants and, in particular, Koen Frenken and Fabio Montobbio are gratefully acknowledged.

1

Absorptive Capacity and Foreign Spillovers: A Stochastic Frontier Approach 1. Introduction In the literature on economic growth in developing countries, international technology flows have gained growing attention. International technology can ‘flow’ from the originating country to the receiving country in several ways. Among them, foreign direct investment and trade in intermediate inputs have been the subject of a great deal of empirical work. Most studies choose firms or plants as units of analysis and adopt a neoclassical production function framework, in which the average response of the endogenous productivity variable to a change in one of the exogenous variables (such as the intensity of FDI and trade in intermediate inputs) is estimated by means of classical regression analysis. Although econometric scrutiny does not always confirm strong anecdotal evidence, the majority of studies find significant positive impacts of international flows indeed.1 Most studies adopt an approach based on production functions, often by means of panel data regressions. One of the most prominent disadvantages is the impossibility to obtain an understanding of the causes of observed heterogeneity. The focus of these studies is on ‘representative behaviour’ (or, ‘average behaviour’) instead. Deviations from this behaviour are merely seen as realisations of a random noise process. If, for example, productivity performances show an increasing variance over time, the production function approach does not yield any insights, as the effect is just an increase in the variance of the stochastic random noise process. In this chapter, we borrow an interesting alternative approach from a subfield in econometrics called “Stochastic Frontier Analysis” (SFA) to study labour productivity growth in Indonesian plants.2 Typical techniques belonging to this subfield do not estimate ‘average relationships’ between variables, but relationships for best-practice plants. Furthermore, they enable researchers to use plant characteristics as potential explanations of the extent to which other plants’ performances fall below best-practice. In the context of our chapter, this implies that we will estimate relationships between inputs (capital, labour) and output (value added) for best-practice firms for several years, to get indications of the degree to which international technology spillovers affected productivity growth. Simultaneously, we will link the underperformance of other firms to 1 2

An extensive overview of empirical studies was recently given in Keller (2004). A modern textbook is Kumbhakar and Lovell (2000).

2

variables that relate to the evolutionary concept of absorptive capacity (Cohen and Levinthal, 1990), such as labour quality, presence and strength of links to foreign markets, ownership, experience etc. The results are quantifications of the failure to fully assimilate international technology spillovers, and thereby to raise productivity to its potential level.3 The organisation of the chapter is as follows. In Section 2, we will shortly review some theories of productivity growth that are relevant for our empirical approach. Section 3 proposes our methodology. It deals with the “appropriate technology” accounting framework and discusses the way in which frontiers and distances to these frontiers are estimated. Section 4 is devoted to data issues. Special attention will be paid to the procedures we adopted to clean the dataset. In Section 5, we will present our results. Section 6 concludes and proposes a few directions for future research.

2. Selected Theories about Productivity Growth Convergence (or its absence) of labour productivity levels has attracted a lot of attention, both from economic theorists and from more empirically oriented scholars. Although it is hard to classify theories in a field characterised by synthesis and hybridisation, roughly two categories of theories can be discerned. We follow Nelson and Pack (1999) in using the labels “accumulation theories” and “assimilation theories”. Accumulation theories basically assume that raising capital intensities (be it physical capital or human capital) automatically leads to labour productivity growth, although increasingly more investment is required for a given productivity gain. In this view, labour productivity growth is governed by movements along the production function of a given country, sector or firm under consideration. This perspective implies that accumulation of capital is the cause for growth in labour productivity. Assimilation theories challenge this view. Here, technology is seen as something that does not automatically and immediately flow across firms or countries. Instead, only firms or countries that have invested sufficiently in their “absorptive capacities” will be able to turn innovations developed elsewhere into productivity gains for themselves. In the view of assimilationists, policies to stimulate entrepreneurship and eagerness to learn have been much more important. Such a view on macroeceonomic performance can, with relatively minor modifications, be transferred to studies at firm or plant level. The resource-based view of the firm (see Teece, 2000, for example) stresses that long-run firm performance is mainly determined by learning capabilities.

3

The analysis can be conceived as an empirical approach to part of the technology-assimilation model of Nelson and Pack (1999).

3

In this chapter, we will differentiate between two barriers to attaining productivity levels attained by better performing plants. The first type of barrier relates most strongly to issues mentioned above. Pack (1987) and Van Dijk (2005, Ch. 8) show that plants that are similar in terms of the types of machines installed attain widely varying productivity levels.4 Apparently, learning and organisational capabilities are not identically distributed across plants, which shows up in different productivity figures for plants with more or less identical equipment installed. The second type of barrier is quite closely associated with what Abramovitz (1989) labelled “technological incongruence”. A similar idea has recently been proposed in the form of a formal model by Basu and Weil (1998). They defined technologies as particular combinations of inputs, or, in other words, capitallabour ratios. New knowledge is only “appropriate” for a range of such technologies. Firms or countries will in the short run only be permitted to benefit from innovations if these relate to technologies that are comparable to the ones operated.5 In the longer run, non-appropriate innovations can become appropriate if the firm or country invests to such an extent that it shifts its technology to a capital intensity level comparable to the innovating firm or country.6 The predictions concerning convergence and divergence of productivity levels that follow from the Basu and Weil model are based on the assumption that more capital intensive technologies allow producers to attain higher maximum levels of labour productivity. Los and Timmer (2005) showed that both types of barriers to catch-up play an important role in the empirics of macroeconomic growth. We adopt several parts of their methodology to investigate how innovations, changes in absorptive capacity and technologies operated contribute to the productivity growth experiences at the plant level in Indonesia’s manufacturing sector.

3. Methodology In this section, we will describe the empirical methodology we use. It contains two parts. First, we will outline how we decompose labour productivity growth (or decline) of a plant into the effects of innovation, assimilation and equipment upgrading which creates 4

5

Pack (1987) studied the performance of textile plants in Kenya, the Philippines and the UK. Van Dijk (2005) focused on the productivity levels of paper-making plants in Indonesia and Finland. Basu and Weil (1998) illustrate this concept by arguing that new knowledge pertaining to the very capital-intensive maglev-trains in Japan will not be useful to transporters in Bangladesh using very capital-extensive bullock carts technologies.

6

Atkinson and Stiglitz (1969) introduced the concept of ‘localised learning by doing’ by which they suggested that firms improve the productivity of a particular mix of capital and labour over time. Basu and Weil (1998) extended this notion by emphasizing the importance of ‘localized knowledge spillovers’.

4

potential for spillovers. Next, we will discuss the estimation methods required to arrive at quantifications of these effects.

3.a Identifying the Sources of Growth Los and Timmer (2005) decomposed labour productivity growths rates of a group of countries, between 1970 and 1990, into the effects of movements towards the frontier, or changes in technical efficiency (assimilation), movements of the frontier (innovation), and capital deepening (creating potential). The decomposition form itself was popularised by Kumar and Russell (2002), but Los and Timmer were the first to link their results to the theories discussed in the previous section. Our approach starts from a similar perspective. It is novel in the sense that it explicitly relates the decomposition results to observable characteristics of the plants. Figure 1 shows a plant’s actual labour productivity levels y0 and y1, in an industry with production frontiers f0 and f1, for periods 0 and 1, respectively. The labour productivity change (y1/y0) of this plant can be decomposed in the following way: y1  y1 y a   y c y d   ⋅ = ⋅ ⋅ y 0  y d y 0   y a y b 

0 .5

 y y  ⋅  b ⋅ d   ya yc 

0 .5

(1)

or

(1 + ˆy ) = (1 + ˆy )⋅ (1 + ˆy )⋅ (1 + ˆy ) T

A

C

I

(2)

(

)

In the first term on the right hand side 1 + yˆ A , a value of yˆ A larger than 0 indicates that the plant under consideration has increased its labour productivity for the technology operated. In other words, it indicates that the plant has been able to bring about an increased exploitation of technological potential as compared to the maximum productivity observed for the equipment operated. We call this the assimilation effect.7 The second explanatory factor (1 + yˆ C ) indicates the changes in labour productivity due to increases in capital intensity alone. While a higher capital intensity in itself does not generate higher labour productivity, it can lead to an upward shift in the attainable or the ‘target’ productivity levels, depending on the slope of the frontier. Therefore, a value greater than 0 for yˆ C can be interpreted as creating potential. The third 7

Below, we will argue that our estimation framework allows us to decompose assimilation effects into “explained assimilation” (explained by means of absorptive capacity indicators) and “unexplained assimilation”.

5

factor (1 + yˆ I ) points to the effect of localised technological change that results in the upward shift of the production frontier. Assuming that the plant’s capital intensity remains constant, a positive value for yˆ I indicates that it has benefited from an increase in the maximum attainable labour productivity levels for the given technologies. We call this the innovation effect. Figure 1: Labour productivity growth decomposition

Y/L

yd yb

F(1)

yc

F(0)

ya y1 y0

*(1) *(0)

C/L

Los and Timmer (2005) estimated the productivity frontiers for the beginning and end periods using data envelopment analysis (DEA). We follow a similar approach, but with the key difference that we derive the frontier labour productivity levels by means of stochastic frontier analysis (SFA). This change of method has advantages and drawbacks. The major drawback is that truly localised innovation cannot be modelled, as the

6

estimated elasticity of foreign R&D spillovers (the source of technological change) is the same across the full range of technologies. As a result, the shifts in the frontier labour productivity levels always amount to an identical proportional growth rate across the full range of technologies. The distance to the frontier, however, can well change, thereby allowing potentials for spillovers to change. The major advantage is that the location of the frontier is not very sensitive to measurement errors for a small number of firms. As is well known (see, e.g., Coelli et al., 1998), DEA results can be distorted quite a bit. In view of the sizeable measurement and reporting errors that are often found in plant level surveys, especially in developing countries, we feel that the net advantage of SFA as compared to DEA is clearly positive.

3.b Estimation method In recent years, a number of studies have employed stochastic frontier estimations for estimating and explaining inefficiencies of firms and plants in industries. Until recently, the standard approach was a two-stage estimation procedure, in which the production frontier is first estimated. In the second stage, the resulting inefficiencies (the vertical distances from the observed productivities to the estimated frontier) are regressed on firm-specific variables (see e.g. Pitt and Lee, 1981).8 Estimation in the second stage, however, contradicts the assumption of identically distributed inefficiency effects that underlies the estimation of the stochastic frontier in the first stage. To overcome this methodological problem, several authors have suggested single-stage procedures for simultaneously estimating both the stochastic frontier and inefficiency functions. The Battese and Coelli (1995) model is one such approach. Consider the following production function for panel data: yit = α + X it β + ε it

(3)

where yit is the dependent variable corresponding to the ith plant and time t, X is a vector of explanatory variables, and e it is the composite error term. It consists of a white noise error vit: vit ~ iid N ( 0, σ v2 ) and uit . The two sets of disturbances are assumed to be independent. The uits are non-negative random variables associated with technical inefficiencies, and are assumed to be independently (but not identically) distributed as truncations (at zero) of the N ( µ it , σ u2 ) distribution, with: µit = Z itδ

8

(4)

For a recent survey, see Wang (2003).

7

in which Z is a vector of observable, non-stochastic explanatory variables associated with technical inefficiency, and d is a vector of unknown coefficients. In this model, the maximum likelihood method is used for the simultaneous estimation of the parameters of the frontier and technical inefficiency models, i.e., estimation of the values of the unknown parameters ßs, ds, s u2 and s v2. We computed the estimates using the FRONTIER software package (Coelli, 1996). Battese and Coelli (1995) also provide an expression for the conditional expectation of exp(-uit) given eit. The maximum likelihood estimation of this function is used to estimate the technical efficiency index of the ith firm at time t, based on expected values conditional on the observed values of the explanatory variables in X and Z. When the productivity frontier is expressed in logarithms, the technical efficiency index (TEI) can be expressed as follows: TEI it = exp(−uit )

(5)

This index has a value between 0 and 1, with 0 (uit? 8 ) indicating the least efficient, and 1 (uit=0) the most efficient plants. Changes in TEI as defined in equation (5) denotes a part of the actual shift in labour productivity. When a change in TEI causes an upward shift, as in figure 1, it can be interpreted as associated with the assimilation of technology-specific knowledge. It is that part of the assimilation effect which can be explained by the changes in the indicators of absorptive capacity, given their estimated coefficients from the SFA model. The remainder of the upward shift cannot be explained, and is calculated as the difference between the actual growth in labour productivity and the predicted growth in labour productivity derived from the SFA model.

3.c The Empirical Model We begin with a production frontier of the Cobb-Douglas form, augmented to account for technological change. In most developing countries, and especially so in Indonesia, foreign technology is the key source of technological progress, because of the virtual absence of own technological efforts. We account for technology flows from abroad by constructing a measure of international R&D stock (IRD). The augmented production function is defined as follows:

Yit = AEαit Litβ IRDtς

(6)

where, Yit is value added of plant i at time t, E total energy use (as a proxy for capital goods use, as further explained below), L total number of workers, and IRD the 8

international R&D stock representing the technology flows available to all plants in the industry (see the following section for a fuller description of the variables). The variable IRD is interpreted to be the driver of shifts in the production frontier in a given industry. Dividing Y and E by L and taking logarithms, equation (6) becomes

y − l = a +α (e − l ) + ς ird + ε it

it

it

it

t

it

(7)

where the lowercase symbols denote variables in logarithms. In the transformation of equation (6) to (7), we impose the assumption of constant returns to scale in the rival inputs labour and energy.9 We use equation (7) as the frontier function that will be estimated simultaneously with the inefficiency function, based on the procedure described in the previous subsection. Given that technology-embodied inputs have often shown to be an important channel of foreign technology diffusion, a plant’s access to imports might be a good proxy of its ‘access to foreign technology’. Access to a source of technology does not, however, imply that the acquisition of technology is guaranteed. This is because technology is not entirely ‘codified’, and indeed often takes a highly ‘tacit’ form (Polanyi, 1958). Therefore, the extent to which a plant is able to ‘absorb’ knowledge related to new technologies, also known as a plant’s absorptive capacity (Cohen and Levinthal, 1990), can depend on the quality of its labour force. Evenson and Westphal (1995) proxied this quality by the proportion of scientists and engineers in a plant’s work-force. The ‘ownership structure’ of a plant can also be a significant factor influencing the capacity to assimilate knowledge. A plant with foreign management control might be expected to run more productively than, for example, a non-professional, familycontrolled enterprise. The ‘foreign-connection’ may enable the former to adapt itself much more quickly than the latter to global changes in technology, production relations, etc. The performance of enterprises as compared to other enterprises with similar technologies may also depend on its ‘size’. As noted by Tybout (2000), in many developing countries, the demand for manufactured products is skewed towards simple items which can be efficiently produced using cottage techniques. An opposite effect would be the operation of Schumpeterian dynamics that leads to greater learning efforts by large firms. This may result from scale economies, availability of internal resources in the presence of imperfect markets and/or uncertainties, synergies between technological, 9

The scale of operation could be important in learning, for instance because because big firms often have more contacts with suppliers, are often more strongly represented in professional associations, etc. To investigate the learning effect of scale, we include the variable ‘plant size’ in our inefficiency function.

9

production, marketing and distribution activities, etc. The empirical evidence, mostly pertaining to advanced economies, shows no consensus, however (see Marsili, 2001, for an overview). Another factor that may influence technical efficiency is the ‘age’ of a plant. Experienced plants may enjoy the benefits of learning-by-doing. As Klepper (2002) argues, with increases in competitive conditions firms with greater experience have greater leeway in enhancing their capabilities. Keeping these considerations in mind, we consider the following absorptive capacity variables for the mean inefficiency model represented by equation (4): Zit : Accessit ; LQualit ; Foreignit ; Agei ; Sizeit where, Access represents access to technology spillover, defined as the share of imported intermediate inputs in total intermediate inputs; LQual stands for the quality of labour at a plant, defined as the share of non-production (white-collar) workers in total employment; Foreign represents the proportion of foreign ownership in a plant; Age is measured as the difference between the year of inception and the year of operation, and; Size is defined as the logarithm of the total number of workers. A final aspect to consider is the influence on technical efficiency of factors observable only to the managers of a plant, which are not reflected in survey-based dataset like ours. As a result, such firm-specific effects (or heterogeneities) may be related to other regressors of the model, which may cause biased results. To overcome this problem, we adopt a specification that incorporates plant fixed effect in the inefficiency model.

4. Data Issues Our analysis focuses on the manufacturing sector of a developing country, Indonesia. Our main data sources are two large plant-level data sets, backcast and Statistic Industri (SI) data sets constructed by the Indonesian Bureau of Statistics (Badan Pusat Statistik, BPS) (See Appendix A of Jacob and Los, 2005, for a detailed description of the data sources, variables, cleaning processes, etc.). The data sets cover all large and medium-sized plants in the manufacturing sector of the country, from 1975 to 2001.10 After applying cleaning procedures to account for duplications, reporting errors and data entry errors, we focus

10

We will limit our analysis to the period 1985-1996, due to the better quality of the data since 1985 and to the crisis of 1997-1999.

10

our analysis on industries defined at a low level of aggregation (5-digit classification). This allows us to investigate productivity for sets of plants with homogeneous activities. Since the panel data SFA-approach is data-intensive, we select 17 industries for which at least 10 plants are included in the data set (see Appendix Table A.1). The industries under investigation are quite diverse, which allows us to identify inter-industry differences in the importance of absorptive capacity for productivity performance. Studies aiming at explaining total productivity (TFP) growth are often hindered by immeasurable fluctuations in capacity utilisation. Although we do not study TFP growth but labour productivity growth, we encounter similar problems: Our technologies are defined by capital intensities, that is, ratios between a (quasi-) fixed input and a much more variable input. We circumvent this problem by proxying a plant’s capital use by its energy consumption, about which much information is available in our dataset.11 This is also convenient from a practical point of view, since our data on energy use cover a much longer time span than those on investment on which we should base our capital stock estimates. As a determinant of labour quality, we do not have detailed information about skills in our database for a sufficiently long period. Therefore, we proxy differences in skill level of labour force across plants by differences in the share of non-production workers in the total work force. Finally, we should describe how we estimate international R&D stocks that capture technology flows. Since Indonesian firms do generally not undertake any formal R&D activities themselves, it can safely be assumed that new technology must come from abroad (Hill, 1996). Our admittedly poor, but widely accepted assumption if suitable output indicators of innovation are not available, is that technology production is proportional to R&D expenditures. We have data on R&D expenditures by industry for ten countries that together account for approximately 60% of the imports to Indonesia and about 85% of the total OECD R&D expenditure. The selection of this sample is justified because empirical evidence suggests that “it is not the intensity of import per se that matters, but rather the distribution of the countries of origin. The more you import from highly R&D intensive countries, the larger the impact of foreign R&D” (Lichtenberg and van Pottelsberghe 1998, p.1483). Apart from imports, technology purchase, technology collaboration and exports by Indonesian firms as well as foreign investment in the domestic market can all act as carriers of technology spillovers. To accommodate these different channels of technology flow, we do not weight foreign R&D stock by the volume of imports to Indonesia, which is a standard approach in the spillover literature. Instead, we consider solely the technological proximity between the 11

This approach dates back to the seminal neoclassical macro-economic growth accounting study by Jorgenson and Griliches (1967).

11

foreign and domestic industries to weight the foreign R&D stock. The specific channels of foreign technology flows are introduced in the inefficiency function. In the first step, foreign R&D stock is weighted by an index of technological distance between the sector of origin and the sector of destination. We use a patent-based measure of technological distance between sectors derived by Verspagen (1997) from EPO (European Patent Office) data. In the second step, the resulting R&D stock is weighted by an index of technological congruence between sector in the advanced economy and Indonesia that are comparable in terms of their classification code. This weight accounts for inter-country differences between sectors. It captures the idea that an industry in a follower country benefits more from the global pool of technology, the greater its technological congruence with industries in advanced countries. The resulting international R&D stock at the industry level can be expressed in terms of the following equation:

IRD j (t ) = ∑ ( RDck Pkj Scj )(t )

(8)

c, k

where IRDj is the international R&D stock resulting from technology flows available to all plants in the Indonesian industry j; RDck is the R&D stock in sector k of partnercountry c; Pkj is an element of the patent information flow matrix P, and it captures the flow of sector k’s R&D efforts to sector j; and Scj is the technological congruence between sector j of Indonesia and the same sector of its partner country c. Scj is derived by comparing the input coefficient vectors for sector j in the two countries. It takes a value of 1 if sector j is perfectly similar in the two countries, and zero in the event of complete dissimilarity between them (see Appendix B of Jacob and Los, 2005). Given the fact that the R&D data we use are available only at a level of aggregation of 2, 3 and, in a few cases, 4-digit (ISIC Rev 2), IRDj in the above equation corresponds to these levels. To generate IRD at the 5-digit level, we constructed similarity indices between the two sets of classifications, using their respective input coefficients vectors.

5. Results 5.a Results for frontier and inefficiency estimation We begin by documenting some summary statistics of the variables used in the SFA model in 17 5-digit-industry, plant-level samples. Table 1a shows the means and standard deviations across plants of the levels of the variables. Table 1b shows similar statistics of the average annual growth rates of the variables. The highest average labour productivity

12

level and growth rate is found for “drugs” (35222). The lowest average values for these two variables are found for “tobacco” (31410) and “clay tiles” (36422), respectively. The former industry is also found to have the lowest average energy intensity, while “plywood” (33113) has the highest. Plants in “garments” (32210) recorded the fastest average growth in energy intensity, while those in “crumb rubber” (35523) recorded the slowest growth. Plants in “paints” (35210) could benefit from the biggest pool of relevant R&D done in foreign countries. The smallest pool is found for “sawmills” (33111). In other variables too, differences are noticeable. For example, the access-tospillover variable has a maximum average value of nearly 0.70 (“drugs”, 35222) and a minimum average value of less than 0.01 (“crumb rubber”, 35523). Foreign ownership is absent in eight of the 17 industries. Average foreign ownership grew fastest in “garments” (32210), while it declined in “macaroni” (31171) and “crumb rubber” (35523).12 Another noticeable feature in both panels of Table 1 are the high standard deviations reported for many of the variables across plants in the industries considered. Quite often, the standard deviation is larger than the mean value. This phenomenon reflects the highly dual structure of the Indonesian manufacturing sector, described in detail by, among others, Hill (1996). Table 2 reports the SFA estimation results. For brevity, we do not report the estimation results for the plant-dummies included in the inefficiency function. The results for the frontier production function show that the coefficients of both energy intensity, e-l, and the international R&D stock representing knowledge spillovers, ird, have a positive sign in most industries. The estimated coefficients of energy intensities for the slopes of the productivity frontiers are generally fairly small, however, and even statistically insignificant at the 10% level for 4 industries. This implies that it does not pay off very much for plants just to invest more, as is suggested by advocates of accumulationist theories. Consequently, accumulation alone cannot be considered as an important source of productivity growth in Indonesian manufacturing.13

12

13

In Table 1, the standard deviations for the variable IRD always equal zero. This is due to the fact that we defined the international R&D stocks as industry-level variables (see Section 4). It should be noted, however, that this result might be partly due to the fact that new equipment is often more energy-saving than more outdated machinery. Consequently, rises in energy consumption might overestimate rises in capital intensity.

13

Table 1a. Summary Statistics of 17 5-digit ISIC Industries: Levels (means, standard deviations in brackets) Industry (ISIC)

plants

Val/ lab

Egy/ lab

IRD

age

foreign

access

lqual

size

31171

22

31179

27 15

31420

34

32114

68

32121

26

32130

20

32210

74

33111

13

33113

20

33211

17

34200

45

35210

13

35222

46

35523

20

35606

29

36422

18

1.221 (1.026) 5.761 (5.073) 0.364 (0.661) 0.801 (0.983) 30.881 (32.014) 2.996 (2.109) 8.846 (6.301) 1.144 (1.248) 650.563 (600.405) 3928.307 (4241.68) 6.223 (4.811) 3.802 (5.118) 3.332 (3.062) 4.854 (5.229) 329.395 (243.739) 7.526 (5.669) 2.255 (1.673)

5.137 (0.000) 8.536 (0.000) 7.030 (0.000) 7.532 (0.000) 36.293 (0.000) 29.681 (0.000) 50.235 (0.000) 35.502 (0.000) 1.907 (0.000) 3.684 (0.000) 4.654 (0.000) 98.361 (0.000) 321.818 (0.000) 211.064 (0.000) 16.434 (0.000) 14.390 (0.000) 38.812 (0.000)

17.182 (7.624) 19.759 (15.421) 25.967 (10.542) 30.059 (14.033) 20.794 (9.159) 24.846 (12.699) 16.950 (9.202) 16.716 (10.996) 14.346 (5.757) 10.400 (3.144) 17.618 (10.588) 24.122 (16.939) 25.500 (17.239) 20.261 (10.031) 16.850 (3.843) 14.638 (4.998) 22.722 (15.973)

0.006 (0.027) 0.033 (0.134)

31410

3.782 (4.925) 6.502 (6.495) 3.241 (3.521) 8.472 (8.500) 7.836 (7.151) 5.069 (2.867) 9.469 (5.964) 6.632 (6.830) 20.489 (12.767) 20.194 (8.183) 9.099 (4.172) 11.421 (8.181) 27.484 (30.403) 28.679 (22.037) 17.594 (16.738) 7.259 (5.550) 3.335 (1.675)

0.023 (0.044) 0.074 (0.165) 0.001 (0.004) 0.048 (0.105) 0.062 (0.136) 0.074 (0.144) 0.084 (0.214) 0.130 (0.199) 0.005 (0.016) 0.024 (0.035) 0.056 (0.081) 0.113 (0.133) 0.354 (0.233) 0.686 (0.243) 0.000 (0.000) 0.587 (0.242) 0.024 (0.069)

0.087 (0.069) 0.177 (0.160) 0.071 (0.063) 0.099 (0.065) 0.117 (0.070) 0.106 (0.067) 0.144 (0.083) 0.094 (0.066) 0.179 (0.071) 0.152 (0.060) 0.159 (0.060) 0.260 (0.121) 0.293 (0.153) 0.399 (0.195) 0.162 (0.067) 0.139 (0.058) 0.069 (0.071)

151.996 (160.860) 308.573 (407.210) 182.561 (120.635) 734.493 (984.163) 484.556 (550.715) 265.920 (260.919) 340.538 (350.749) 501.825 (680.247) 452.494 (577.791) 1445.056 (766.116) 319.110 (396.681) 169.704 (116.685) 162.660 (104.254) 233.133 (133.355) 239.948 (85.367) 332.158 (400.633) 135.398 (73.559)

0.032 (0.151)

0.032 (0.132) 0.065 (0.197) 0.045 (0.139)

0.165 (0.262) 0.247 (0.349) 0.242 (0.430)

Notes: (i) Val/Lab – value added per employee; Egy/Lab – energy use per employee; Size - number of employees; IRD – international R&D stock in real 1990 PPP US dollars (millions). Value added and energy are in thousands of real 1990 PPP US dollars. (ii) See main text for the definitions of the other variables. (iii) See appendix table A.1 for industry definitions.

14

Table 1b. Summary Statistics of 17 5-digit ISIC Industries: Average annual growth rates (means, standard deviations in brackets) Industry (ISIC)

plants

31171

22

Val/ lab

Egy/ lab

IRD

foreign

access

lqual

0.191 0.243 0.056 -0.038 -0.271 0.062 (0.160) (0.178) (0.000) * (0.296) (0.078) 31179 27 0.209 0.304 0.056 0.044 0.102 0.100 (0.211) (0.371) (0.000) (0.062) (0.583) (0.156) 31410 15 0.409 0.550 0.056 4.327 0.108 (0.365) (1.034) (0.000) * (0.165) 31420 34 0.195 0.978 0.056 -0.232 0.401 (0.228) (2.950) (0.000) (0.488) (1.703) 32114 68 0.262 0.208 -0.028 -0.025 0.218 0.069 (0.208) (0.287) (0.000) (0.044) (1.538) (0.110) 32121 26 0.219 0.270 -0.028 0.236 0.114 (0.207) (0.225) (0.000) (2.084) (0.142) 32130 20 0.366 0.195 -0.028 -0.088 0.090 (0.513) (0.212) (0.000) (0.327) (0.086) 32210 74 0.286 2.417 -0.028 0.099 0.469 0.175 (0.260) (17.706) (0.000) (0.164) (2.822) (0.492) 33111 13 0.314 0.179 0.032 0.003 -0.735 0.108 (0.210) (0.330) (0.000) (0.005) (0.442) (0.126) 33113 20 0.196 0.326 0.032 0.019 0.920 0.202 (0.156) (0.410) (0.000) (0.027) (3.295) (0.298) 33211 17 0.263 0.128 0.032 1.033 0.037 (0.377) (0.152) (0.000) (4.609) (0.084) 34200 45 0.214 0.274 -0.047 -0.010 0.089 (0.178) (0.256) (0.000) (0.764) (0.126) 35210 13 0.233 0.350 0.039 0.025 0.169 0.058 (0.226) (0.321) (0.000) (0.026) (0.882) (0.091) 35222 46 0.432 0.247 0.061 0.005 -0.014 0.039 (0.608) (0.216) (0.000) (0.036) (0.304) (0.079) 35523 20 0.314 0.100 0.030 -0.022 -0.656 0.069 (0.153) (0.174) (0.000) (0.039) * (0.101) 35606 29 0.268 0.310 0.030 -0.157 0.066 (0.202) (0.340) (0.000) (0.336) (0.097) 36422 18 0.174 0.233 0.019 -0.212 0.117 (0.091) (0.246) (0.000) (0.292) (0.192) * The growth rate of the variable under consideration is positive in only one observation.

size 0.043 (0.063) 0.066 (0.071) 0.067 (0.103) 0.030 (0.057) 0.051 (0.069) 0.051 (0.064) 0.077 (0.083) 0.074 (0.114) 0.051 (0.051) 0.108 (0.092) 0.120 (0.075) 0.034 (0.065) 0.047 (0.053) 0.042 (0.063) 0.016 (0.033) 0.108 (0.098) 0.038 (0.046)

The sensitivity of the frontier to increases in foreign R&D appears to be prominent. Apparently, the best-practice plants in Indonesia reaped substantial fruits from technological spillovers from abroad. The coefficient for this variable was statistically significant for 10 of the 17 industries studied, at the 10% level. Our main interest lies in understanding the factors that cause deviations from the best-practice technology. The estimate for the variance parameter ? (gamma in Table 2) that corresponds to the estimated share of the inefficiency term in the variance of the composite error term has a positive sign in all industries, and is significant in most industries (12 industries). This should be considered as evidence for the idea that inefficiency effects contribute substantially to the variety of labour productivity levels indicated by positive standard deviations for this variable.

15

Table 2. Stochastic Frontier Estimates for 17 5-digit (ISIC. Rev.2) Industriesa Industry (Isic Rev.2) Constant Egy/Lab Ird Constant Age Foreign Access Lqual Size Gamma Plants Observations (contd) (contd)

31171

31179

-5.132 (1.765)* 0.121 (0.039)* 0.885 (0.114)*

-11.661 (19.375) 0.111 (0.040)* 1.413 (0.225)*

1.701 (0.406)* -0.251 (0.120)* -0.312 (0.293) 0.651 (0.330)* -0.688 (0.231)* -0.292 (0.093)* 0.126 (0.049)* 22 264

3.924 (18.938) 0.219 (0.144) -1.409 (0.364)* 0.049 (0.442) -1.445 (0.338)* 0.261 (0.116)* 0.601 (1.917) 27 324

31410

31420

32114

Production function 5.484 -3.450 27.665 (5.993) (0.998)* (5.495)* 0.047 -0.056 0.169 (0.025)* (0.086) (0.029)* 0.244 0.852 -1.050 (0.382) (0.089)* (0.314)* (Mean) Inefficiency function 3.049 -0.220 4.296 (1.132)* (0.972) (0.786)* -0.769 -0.028 -0.370 (0.426)* (0.333) (0.180)* 4.408 (11.460) 7.028 0.311 -0.039 (4.324) (0.999) (0.149) -1.967 0.119 -0.903 (1.499) (0.997) (0.356)* -0.535 -1.097 0.119 (0.183)* (0.214)* (0.074) 0.765 0.542 0.492 (0.107)* (0.220)* (0.537) 15 34 68 180 408 816

32121

32130

32210

33111

33.378 (6.428)* 0.139 (0.037)* -1.403 (0.376)*

54.907 (11.056)* 0.122 (0.053)* -2.543 (0.533)*

38.426 (3.900)* 0.090 (0.022)* -1.626 (0.229)*

-1.746 (2.131) 0.043 (0.052) 0.869 (0.150)*

1.933 (1.286) 0.105 (0.232)

1.649 (1.373) 0.011 (0.201)

0.022 (0.179) -0.581 (0.496) -0.021 (0.100) 0.821 (0.082)* 26 312

0.164 (0.485) 0.184 (0.685) 0.194 (0.114)* 0.206 (0.811) 20 240

3.313 (0.718)* -0.037 (0.083) -0.630 (0.255)* -0.258 (0.111)* -0.220 (0.376) 0.061 (0.056) 0.342 (0.447) 74 888

1.627 (0.691)* -0.048 (0.171) 0.860 (0.785) -0.623 (1.037) 0.020 (0.862) -0.410 (0.223)* 1.000 (0.000)* 13 156

Table 2. Stochastic Frontier Estimates for 17 5-digit (ISIC. Rev.2) Industriesa Industry (Isic Rev.2) Constant Egy/Lab Ird Constant Age Foreign Access Lqual Size Gamma

33113

33211

14.732 (8.177)* 0.180 (0.039)* -0.466 (0.537)

-12.599 (1.025)* 0.062 (0.043) 1.391 (0.071)*

1.607 (0.582)* -0.221 (0.227) 2.767 (3.456) -2.181 (2.197) -2.717 (1.055)* 0.039 (0.185) 0.442 (0.179)* 20 240

-0.298 (0.818) 0.088 (0.240) 0.946 (0.947) 0.581 (1.202) 0.240 (0.278) 0.224 (0.128)* 17 204

34200

35210

35222

35523

35606

36422

24.417 -20.476 -14.688 (1.031)* (4.554)* (1.464)* 0.072 0.122 0.006 (0.035)* (0.049)* (0.031) -0.818 1.580 1.332 (0.056)* (0.237)* (0.077)* (Mean) Inefficiency function -1.135 2.671 2.051 (0.426)* (1.360)* (0.600)* 0.117 0.200 -0.305 (0.123) (0.260) (0.132)* -1.103 -0.122 (0.422)* (0.391) -0.142 -0.007 -0.215 (0.257) (0.228) (0.116)* -0.670 0.431 0.652 (0.401)* (0.673) (0.342)* 0.088 0.499 0.477 (0.085) (0.205)* (0.120)* 0.249 0.354 0.092 (0.044)* (0.095)* (0.021)* 45 13 46 540 156 552

-27.614 (4.848)* 0.147 (0.045)* 2.126 (0.288)*

-20.899 (1.521)* 0.165 (0.039)* 1.765 (0.090)*

-40.764 (15.799)* 0.144 (0.040)* 2.832 (0.850)*

0.360 (0.210)* -0.137 (0.026)* 0.111 (0.428) 0.153 (1.029) -0.456 (0.217)* 0.010 (0.127) 0.000 (0.000)* 20 240

1.520 (0.629)* -0.304 (0.180)*

3.564 (4.923) -0.505 (0.211)*

0.119 (0.136) -1.713 (0.764)* 0.238 (0.143)* 0.096 (0.027)* 29 348

-0.974 (0.418)* -0.696 (0.764) -0.319 (0.144)* 0.456 (0.875) 18 216

Plants Observations Notes: (i) a Standard errors are in parentheses; * significant at 10%. (i) Egy/Lab – e / l. (ii) See main text for the definitions of the other variables. (iii) See appendix table A.1 for industry definitions.

16

A negative sign for the coefficient of a variable indicates a negative impact of that variable on inefficiency. Among all the variables, changes in labour quality (LQual) variable provide the most promising explanation for changes in comparison to best practice performance. Its coefficient has a negative sign in most industries (11 out of 17), and is statistically significant in seven industries. Foreign ownership (foreign) has a significantly negative coefficient only in three of the nine industries in which the shares held by foreign firms are positive in one or more establishments. For the remaining industries, changes in the degree of foreign ownership do not appear to matter for assimilation. It might well be that a linear specification of the inefficiency effects is not most appropriate here. Explorations to use multiple-regime econometrics (to identify critical values of the degree of foreign ownership), however, are beyond the scope of this chapter, if only because such analyses have hardly been attempted in the SFA branch of econometrics. We argued before that access to spillover (access) is likely to exert a major influence on the technical efficiency of plants. However, this variable yielded a negative coefficient in only eight industries, with statistical significance limited just to three industries. One reason for this result could be the narrowness of our measure of access to spillover as it does not consider the import of capital goods. Secondly, the import intensity of intermediate inputs use is rather low in most industries as may be required to generate sufficient within-plant variations. An additional, and probably the most important, cause of very few significant results is the huge measurement errors that characterise data sets like ours. Although, we did ‘clean’ the data extensively, it is unlikely that this has removed all measurement errors. The age variable demonstrated a favourable impact on assimilation in a majority of the industries. A negative sign for its coefficient in 11 out of a total number of 17 industries appear to suggest that a plant’s ability to assimilate knowledge spillovers from similar best-practice plants increases with its experience. As argued by Klepper (2002), under competitive pressures, firms with greater experience are better positioned to enhancing their capabilities. Our period of analysis covers the export-oriented phase— hence, the more competitive phase—of industrialisation in Indonesia. We may therefore conclude that firms which have been in operation for a longer period of time have been more successful in enhancing their technological and managerial capabilities, and therefore in meeting the challenges of increased-competition. The final variable to be discussed is size, which displays considerable inter-industry variations in its influence. It had an adverse impact on assimilation in a majority of the industries (a positive coefficient in 11, with statistical significance in five industries). Of

17

the six remaining industries where its influence has been favourable, in five cases the coefficients displayed statistical significance. The discussion so far has revealed that changes in deviations from best-practice (due to plant-specific factors) are a significant determinant of a plant’s labour productivity growth. In the following section we extend these results by decomposing labour productivity growth into that resulting from the shifts in the frontier of an industry’s technology, capital deepening and efficiency gains (or losses). We provide theoreticallygrounded interpretations for the contributions of these three factors to average labour productivity growth in an industry.

5.b Results for the decomposition analysis We decomposed plant-level labour productivity growth during the period 1985-1996 in each of the 17 industries under consideration. We then calculated their industry average using the geometric average of the initial and final year employment shares of plants as the weight.14 Table 3 shows the decompositions of average labour productivity growth rates during the period 1985-1996. The first column shows the compound annual average growth in labour productivity and the second column, the period growth rate of productivity. The remaining columns represent the contribution of the four components to (period) labour productivity growth. All industries experienced a positive growth in labour productivity during this period, with important inter-industry variations as may be expected of a heterogeneous group of industries. In a majority of the industries, we see that productivity growth resulted from a combination of assimilation (unexplained) and innovation. While the former was the main contributor in most of the industries, the latter was the leading contributor in industries like paints & varnishes, plastics and rubber. Explained assimilation was a major contributor in about five industries important among which were tobacco, plywood, cigarettes and clay tiles. The contribution of creating spillover potential was very limited in all industries. This result is mainly due to the flat shapes of the estimated frontiers. Increasing capital intensity does hardly contribute to a higher potential labour productivity. In other words, learning or assimilation potentials remained more or less stagnant.

14

Prior to deriving the industry-average, the multiplicative components of the decomposition equation were transformed, by taking their logarithms, into additive components.

18

Table 3. Decomposition of Productivity Growth: 1985-96 Industry (Isic Rev.2) 31171 31179 31410 31420 32114 32121 32130 32210 33111 33113 33211 34200 35210 35222 35523 35606 36422

Annual Growth (%)

Period Growth (%)

0.704 0.567 0.421 0.383 0.309 0.518 0.756 0.171 0.873 0.023 0.515 0.480 1.138 0.660 0.606 0.316 0.782

8.022 6.412 4.733 4.295 3.447 5.854 8.637 1.896 10.037 0.252 5.810 5.414 13.254 7.504 6.871 3.529 8.944

Contribution to Productivity Growth Explained Assimilation 3.776 -1.783 6.928 1.818 0.717 -0.253 -1.103 0.092 0.991 0.997 -2.623 -0.077 -1.098 0.307 0.578 0.193 4.219

Unexplained Assimilation 0.414 3.172 -2.668 0.448 1.579 3.330 4.033 0.828 6.520 -0.240 5.081 3.694 4.496 3.247 1.913 1.177 -0.268

Innovation

Potential

3.826 4.857 1.152 2.158 0.875 2.641 6.141 1.010 3.081 -0.721 3.627 1.659 8.102 3.943 4.816 2.586 5.199

0.007 0.165 -0.679 -0.129 0.277 0.136 -0.435 -0.034 -0.554 0.216 -0.274 0.139 1.753 0.006 -0.436 -0.426 -0.206

Note: See appendix table A.1 for industry definitions.

6. Conclusions and Future Research In this chapter, we proposed a Stochastic Frontier Analysis approach to study labour productivity growth in the Indonesian manufacturing sector (in the late eighties and early nineties). We focused specifically on the effects of inflows of foreign technology and the role of absorptive capacity changes on plant-level differences in productivity growth. Our main findings are that foreign R&D played a significant role in moving the productivity frontier in an upward direction. Hence, the inflow of technology was an important source of productivity growth for best-practice plants. The creation of spillover potential (by investing to use more capital-intensive technologies) did hardly contribute to labour productivity growth, since relations between capital intensity and labour productivity were almost non-existent for best-practice pants. As a consequence, shifting to higher capital intensities did hardly imply more potential spillovers to benefit from. This finding is clearly at odds with the major assumptions underlying the accumulationist theories of growth. Assimilation (movements towards the frontier) did play an important role. We could distinguish between two kinds of assimilation effects. The first type could be explained by our absorptive capacity indicators, such as labour quality and degree of foreign ownership. For many industries, we found estimation results that underline the importance of building absorptive capacity for assimilating knowledge from best-practice

19

firms that operate similar technology. In a quantitative sense, however, these effects were often dwarfed by the second kind of assimilation effects. This unexplained assimilation effects were very big and dominated the composite effect. The importance of unexplained assimilation is worrisome on the one hand, in the sense that we cannot explain much. On the other hand, it confirms our feeling that much heterogeneity of plants is not captured by survey-based datasets. Our absorptive capacity indicators are rough ones, and are subject to considerable measurement error. In our view, more in-depth case studies (like Pack, 1987, and Van Dijk, 2005) offer much better opportunities for assessing the importance of foreign technology and differences in absorptive capacity. Studies like ours can play a useful role in investigating to what extent case study results can be generalised. In that sense, next steps along the lines of the present chapter could be to probe further into the interindustry differences concerning the estimation results and to link such differences to differences in the types of technologies used. Industrial taxonomies, for example, as proposed by Pavitt (1984) might constitute an interesting and worthwhile starting point in this respect.

References Abramovitz, M. (1989), Thinking About Growth (Cambridge: Cambridge University Press). Atkinson, A.B. and Stiglitz, J.E. (1969), “A New View of Technological Change”, Economic Journal, vol. 79, pp. 573-578. Basu, S. and Weil, D.N. (1998), “Appropriate Technology and Growth”, Quarterly Journal of Economics, vol. 113, pp.1025-1054. Battese, G.E. and Coelli, T.J. (1995), “A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data”, Empirical Economics, vol. 20, pp. 325-332. Cohen, W. and Levinthal, D. (1990), “Absorptive Capacity: a New Perspective on Learning and Innovation”, Administrative Science Quarterly, vol. 35, pp. 128–152. Coelli, T.J. (1996), “A Guide to FRONTIER Version 4.1: A Computer Program for Stochastic Frontier Production and Cost Function Estimation.”, CEPA (Centre for Efficiency and Productivity Analysis) Working Paper, No.7, University of New England, Department of Econometrics, University of New England, Armidale, pp.33. Coelli, T.J., Rao, D.S.P. and Battese, G.E. (1998), An Introduction to Efficiency and Productivity Analysis (Boston: Kluwer Academic Publishers). Evenson, R. and Westphal, L. (1995) “Technological Change and Technology Strategy”, in T.N. Srinivasan and J. Behrman (eds.), Handbook of Development Economics, Vol. 3. (Amsterdam: North Holland). Hill, H. (1996), The Indonesian Economy since 1966: Southeast Asia’s Emerging Giant (Cambridge: Cambridge University Press).

20

Jacob, J. and Los, B. (2005), “The Impact of International Technology Spillover and Absorptive Capacity on Productivity Growth in Indonesian Manufacturing Firms”, Paper presented at EMAEE 2005 (Utrecht, May 19-21). Jorgenson, D.W. and Griliches, Z. (1967), “The Explanation of Productivity Change”, Review of Economic Studies, vol. 34, pp. 249-283. Keller, W. (2004), “International Technology Diffusion”, Journal of Economic Literature, vol. 62, pp. 752-782. Klepper, S. (2002), “The Capabilities of New Firms and the Evolution of the US Automobile Industry”, Industrial and Corporate Change, 11(4), pp. 645-666. Kumar, S. and Russell, R.R. (2002), “Technological Change, Technological Catch-Up and Capital Deepening: Relative Contributions to Growth and Convergence”, American Economic Review, vol. 92, pp. 527-549. Kumbhakar, S.C. and. Lovell, C.A.K (2000), Stochastic Frontier Analysis (Cambridge: Cambridge University Press). Lichtenberg, F.R. and van Pottelsberghe de la Potterie, B. (1998), “International R&D Spillovers: A Comment”, European Economic Review, vol. 42, no. 8, pp. 1483-1491. Los, B. and Timmer, M. (2005), “The ‘Appropriate Technology’ Explanation of Productivity Growth Differentials: An Empirical Approach”, Journal of Development Economics, vol. 77, pp. 517-531. Marsili, O. (2001), The Anatomy and Evolution of Industries: Technological Change and Industrial Dynamics (Cheltenham: Edward Elgar). Nelson R.R. and Pack, H. (1999), “The Asian Miracle and Modern Growth Theory”, Economic Journal, vol. 109, pp. 416-436. Pack, H. (1987), Productivity, Technology, and Industrial Development: A Case Study in Textiles (New York: Oxford University Press). Pavitt, K. (1984), “Sectoral Patterns of Technical Change: Towards a Taxonomy and a Theory”, Research Policy, vol. 13, pp. 343-373. Pitt, M.M. and Lee, L.F. (1981), “The Measurement and Sources of Technical Efficiency in the Indonesian Weaving Industry”, Journal of Development Economics, vol. 9, pp. 43-64. Polanyi, M. (1958) Personal Knowledge: Towards a Post-Critical Philosophy (Chicago: University of Chicago Press). Teece, D.J. (2000), Managing Intellectual Capital (Oxford; Oxford University Press). Tybout, J.R. (2000), “Manufacturing Firms in Developing Countries: How Well Do They Do and Why?”, Journal of Economic Literature, vol. 38, pp. 11-44. Van Dijk, M. (2005), Industry Evolution and Catch Up: The Case of the Indonesian Pulp and Paper Industry, unpublished PhD thesis, University of Eindhoven. Verspagen, B. (1997), “Estimating International Technology Spillovers Using Technology Flow Matrices”, Weltwirtschaftliches Archiv, vol. 133, pp. 226-248. Wang, H-J. (2003), “A Stochastic Frontier Analysis of Financing Constraints on Investment: The Case of Financial Liberalization in Taiwan”, Journal of Business and Economic Statistics, vol. 21, pp. 406-419.

21

Appendix TABLE A.1 Industrial Classification No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

Industry Macaroni, spaghetti, noodle and the like Bakery products Dried tobacco and processed tobacco Clove cigarettes Weaving mills except gunny and other sacks Made-up textile article except wearing apparels Knitting mills Wearing apparel made of textile (garments) Sawmills Plywood Furniture and fixtures mainly made of wood Printing, publishing and allied industries Paints, varnishes and lacquers Drugs and medicines Crumb rubber Plastics, bags, containers Clay tiles

Isic Revision 2 31171 31179 31410 31420 32114 32121 32130 32210 33111 33113 33211 34200 35210 35222 35523 35606 36422

22