Journal of International Development J. Int. Dev. 12, 343±369 (2000)

PRODUCTIVITY SPILLOVERS FROM TECHNOLOGY TRANSFER TO INDIAN MANUFACTURING FIRMS
VINISH KATHURIA* Gujarat Institute of Development Research (GIDR), Gota, Ahmedabad, India

Abstract: The present paper employs techniques from stochastic production frontier and panel data literature to test a spillover hypothesis for large sized ®rms that `presence of foreign-owned ®rms and foreign technical capital stock in a sector leads to reduced dispersion in eciency in the sector and fall is higher for the ®rms that invest in R&D activities'. Dispersion being a relative concept, it may still fall if both the leading foreign ®rm and domestic ®rms show fall in technical eciency over the period and the fall for the leader is higher and vice versa. Given the focus of the study, where concern is for the learning by the domestic ®rms, the study tries to get around with the problem partially, by testing the hypothesis for those local ®rms that have shown productivity improvement over the period. Results suggest that foreign-owned ®rms are close to the frontier in 13 of the total 26 sectors studied. Spillovers result for these 13 sectors indicate that there exist negative spillovers from the presence of foreign ®rms in the sector, but available foreign technical capital stock has a positive impact. Interesting di€erences emerge when the sample is bifurcated into scienti®c and non-scienti®c subgroups. Results for the scienti®c subgroup indicate that the indirect gains or spillovers are not automatic consequence of foreign ®rm's presence, but they depend to a large extent on the e€orts of local ®rms to invest in learning or R&D activities so as to decodify the spilled knowledge. On the other hand, the evidence of spillovers to non-scienti®c nonFDI ®rms is not very strong. Copyright # 2000 John Wiley & Sons, Ltd.

1

INTRODUCTION

The role of multinational corporations (MNCs) in international technology transfer has been a matter of debate for many years. However, in the past decade or so, the

* Correspondence to: Vinish Kathuria, Gujarat Institute of Development Research (GIDR), Nr. Gota Char Rasta, P. O. High Court, Gota, Ahmedabad 380 060, India. E-mail: gidr@ad1.vsnl.net.in

Copyright # 2000 John Wiley & Sons, Ltd.

344

V. Kathuria

proponents of foreign direct investment (FDI)1 seem to have gained upper hand as several developing countries have liberalized their policies to encourage the ¯ow of FDI. The proponents often argue that technology (knowledge) being a public good, the MNCs or technology recipient ®rms cannot capture all the quasi-rents due to their productive activities. Thus, a large share of the host countries' bene®ts from FDI may come in the form of `external e€ects' or `spillovers' of knowledge to domestically owned ®rms that are competitors, suppliers, customers or to those ®rms which have some point of economic contact with MNCs. In fact, one of the reasons often cited to invite foreign ®rms (MNCs) is the prospect of bene®ts accruing to the host country in the form of `productivity spillovers' besides access to modern technology. The simplest example of such a spillover is the case where a local ®rm improves its productivity by imitating the technology used by MNC aliates operating in the local market (demonstration e€ect). However, the most widely acknowledged spillover bene®t of MNCs is the `competitive e€ect' that occurs with the entry of a foreignowned ®rm (Dunning, 1993; Caves, 1974). This forces local ®rms either to use existing technology and resources more eciently or to search for alternate ecient technologies to stay in competition. Indirect bene®ts may also be realized from nonspeci®c human capital investment made by MNCs that are ultimately utilized by the domestically owned ®rms as a result of labour migration. Lastly, the improved managerial and manufacturing practices employed by MNCs like JIT (Just-in-Time), QA (Quality Assurance), QC (Quality Circles), etc., which are pre-requisites for e€ective and ecient use of the newer technology may spill over to the rest of the economy (technology di€usion). In the past two decades, a signi®cant amount of empirical work has been devoted to quantify and estimate the spillovers. The results of these studies are contradictory in terms of overall size and signi®cance of spillovers. The studies by Caves (1974) on Australia, Globerman (1979) on Canada, Blomstrom and Persson (1983) on Mexico and Nadiri (1991) on several OECD countries have found either positive or weak positive spillover impacts of foreign presence on the productivity of local ®rms. On the other hand, there are studies on Britain (Perez, 1998), Uruguay (Kokko et al., 1996), Morocco (Haddad and Harrison, 1993), Venezuela (Aitken and Harrison, 1993), and several European countries (Cantwell, 1989) that have found either negative or no impact of foreign presence on the productivity of domestic ®rms. One possible explanation for the contradictory ®ndings is the di€erence in methodologies and data used in the respective studies. However, it has also been argued that the host country characteristics may have in¯uenced the incidence of spillovers. For instance, in some countries, tougher competition might have forced local ®rms to learn quicker, while in others it is the level and type of technology brought by MNCs that would have facilitated the spillovers or the government might have devised suitable policies to extract the maximum bene®ts from MNCs (Kokko, 1994; de Mello, 1997). Another explanation for contradictory ®ndings comes from the literature on technology gap i.e., there may be ®rm speci®c di€erences in the ability of domestic ®rms to absorb spillovers. The literature on technology gap has also spawned from two opposing strands. Lapan and Bardhan (1973) argue that spillovers are negatively related to the complexity of MNC technology and the width of technology gap.
1

In the present paper, the terms FDI and MNCs have been used interchangeably. J. Int. Dev. 12, 343±369 (2000)

Copyright # 2000 John Wiley & Sons, Ltd.

Technology Transfer and Productivity Spillovers

345

Findlay (1978) and Wang and Blomstrom (1992), on the other hand, hypothesize that spillovers grow with the width of technology gap. Both arguments, however, stress that a threshold level of technology gap is a pre-requisite to absorb the spillovers.2 The empirical literature on spillovers, in general, has corroborated the argument provided by Lapan and Bardhan only. Cantwell (1989), Aitken and Harrison (1991), Haddad and Harrison (1993), Kokko (1994), Kokko et al. (1996) and Perez (1998) in their respective studies have found that spillovers have occurred mainly in those sectors, where the technology gap between the aliates of MNCs and local ®rms is relatively small. The only study that has found that spillovers have grown with the size of the technology gap between foreign and local ®rms is by Blomstrom and Wol€ (1994) on Mexico. While the previous studies give sucient insights on the nature and direction of spillovers, there are some major limitations of these studies. Most of the previous studies, except two, have used cross-section data. Since the process of spillovers essentially involves a process of learning or imitation by a local ®rm from a more knowledgeable foreign or technology recipient ®rm, it takes time to absorb or internalize this learning. Hence, it is dicult to capture the full spillover e€ects using cross-section data. Further, most of the studies are aggregate in nature. Results can be misleading if there is large variability in the size and technological capabilities of local ®rms. Availability of ®rm level data (as in the present case) facilitates to examine domestic and foreign ®rms' behaviour separately. The third major limitation of most of the studies is that they have concentrated on the spillovers associated with MNCs only. As licensing (arm's length of technology-transfer) has been a more widely practised channel of technology transfer for the developing countries since the late 60s, the results of previous studies might be under-stating the true spillover bene®ts.3 In addition, most of the previous studies (except two) have tested the spillover hypothesis on labour or capital productivity, which are only partial measures. E€ect on multi-factor productivity is a better indicator of spillovers from technology transfer. Lastly, in most of the previous studies, spillovers have been assumed as a consequence of presence of foreign ®rms only. However, it is well proven that knowledge (or technology) being tacit, requires investment even by the technology recipient ®rm to decodify it. The level of such investment increases multifold for other ®rms if they wish to bene®t from the spilled knowledge. This implies that the models used in previous studies may be mis-speci®ed, thereby giving con¯icting results. The present study, which aims to ®nd out whether domestic ®rms in the Indian industry have bene®ted from either forms of disembodied technology transfer, i.e. FDI or arm's length transaction, contributes to the existing literature by accounting for all the drawbacks enumerated above. The paper employs techniques from stochastic production frontier and panel data literature and uses data for 14 years from 1975±76 and 1988±89 covering 368 large-sized ®rms belonging to 26 three-digit manufacturing industries for the analysis. The data coverage of the present paper is
2 There is a third line of argument also which suggests that spillovers would be maximum in both the cases whether the technology gap is small or large (Catch-upable gap) (Mody, 1989). A small gap implies that domestic ®rms would be in a position to compete with the foreign ®rms (demonstration e€ect or competition e€ect) and a large technology gap implies that domestic ®rms would gain a lot from other forms of spillovers such as mobility of labour, entrepreneurial culture etc. (technology transfer e€ect). 3 Further, the arm's length transaction being a direct measure of technology transfer, can yield larger bene®ts than the presence of foreign ®rms where the spillover bene®ts are mostly indirect.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

346

V. Kathuria

that of pre-liberalization period when Indian industry was highly regulated in terms of industrial and technology policy. An in-depth analysis of the spillovers in preliberalization phase may be useful both as a benchmark for evaluating the impact of new policies as well as for providing insights to ®ne-tune the evolving policy instruments. The paper is organized as follows. Section 2 brie¯y discusses the model to be estimated to test the spillover hypothesis. Measurement of the variables and the data sources used in the study are discussed in Section 3. Section 4 compares the relative performance and conduct/behaviour of local and foreign owned ®rms followed by statistical results of the models in Section 5. The paper ends by drawing some conclusions in Section 6 and discusses the scope for further research in the area. 2 MODEL SPECIFICATION

2.1 Productivity Spillovers The externalities or technological spillovers can be estimated only through indirect means, as by their nature they do not have any market value. An important indirect measure to estimate spillovers is the e€ect on the dispersion in productivity levels, i.e. the di€erence in the productivity level between the most ecient ®rm (MEF) and the average ®rm in the industry (Stewart and Ghani, 1991). In a static sense, if positive spillovers have occurred because of technology transfer then the sectors with a larger foreign presence or a larger stock of disembodied technology import should show a smaller dispersion in productivity. However, in a dynamic setting if the domestic ®rms are bene®ting from the spilled knowledge then the dispersion in productivity between technology recipient ®rm and other ®rms in the sector should fall over time. Thus, the study tests the hypothesis that: `the presence of foreign owned ®rms and disembodied technology import in the industry reduces the dispersion of productivity level over time and the fall in dispersion is more for the ®rms that invest in learning or R&D activities'.4 It is to be noted that the hypothesis assumes that MNCs or technology recipient ®rms are at or closer to the frontier and domestic ®rms have a lot to learn from them. The testing of above hypothesis involves two major steps. In the ®rst step, ®rmspeci®c time-variant technical eciency indices are calculated for each industry separately using eciency frontier and panel data literature. These technical eciency indices are then compared with the best-practised ®rm in the industry. The comparison yields the relative eciency of the ®rm with respect to the best-practised ®rm in the industry. The second step involves checking whether the relative technical eciency has increased over time and if the change can be attributed to the presence of foreign ®rms and technology import in the sector.

4 Depending upon the impact, spillovers can be broadly classi®ed into `intra-industry' and `inter-industry'. It is evident from the hypothesis that the focus of present study is on intra-industry spillovers.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers 2.2 Stochastic Production Frontier (Eciency Frontier)

347

In the eciency frontier literature, the production function f(x) de®nes the maximum possible output a ®rm can produce given input bundle, x, representing the bestpractice or frontier technology in the industry (Farrell, 1957). A ®rm could be technically or productively inecient, if it employees a larger bundle of inputs than the minimum required to obtain actual output. Once the best-practice or frontier production is estimated, an eciency index for a ®rm can be derived from the deviation of its actual output from the frontier. For example, as shown in Figure 1, a ®rm, which is producing a level of output Q, with observed input coecients represented by D is technical inecient compared to the frontier ®rm at A. The technical eciency of the ®rm relative to the ecient frontier would be OA/DD (Figure 1). From a ®rm's point of view, eciency index indicates how far a given ®rm can increase its output without absorbing further resources.

Figure 1.

Measurement of technical eciency.

Most studies on eciency frontiers have been cross-sectional requiring strong distributional assumptions about the error term in order to separate technical eciency from random noise. The studies have also assumed that technical eciency is independent of factor inputs. The assumption of independence of factor inputs can be a potential source of error in the estimation. For instance, if a ®rm knows its level of technical ineciency, it would try to change its input choices accordingly. Availability of panel data (repeated observations for an individual over time) facilitates estimation of consistent and robust ®rm speci®c technical eciency indices obviating the need for both these assumptions (see Schmidt and Sickles, 1984 for details). But the model by Schmidt and Sickles assumes time-invariant technical ineciency (i.e. the eciency of a ®rm is ®xed over time). The assumption of this time-invariant ®rm speci®c eciency is very strong and depending upon the data may
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

348

V. Kathuria

prove highly unrealistic. For instance, in a technologically dynamic or growing industry Ð like machine tools or software industry, the ®rms' technical eciency would continue to change depending upon their e€orts in investment in human and physical capital etc. The assumption of time-invariant ®rm speci®c eciency was ®rst relaxed by Cornwell et al. (1990).5 In the present study, Cornwell et al. (1990) model has been used to calculate a time-invariant ®rm speci®c technical eciency. Model We begin with a production function with value-added Y, a function of two inputs, labour (L) and capital (K) Yijt ˆ Aijt F…Lijt Y Kijt † …1†

where Aijt is the time-varying productivity level of ®rm i, belonging to industry j. For an industry j, the empirical model can be written as yijt ˆ aj ‡ b xkijt ‡ vijt À uijt Y i ˆ 1Y 2Y F F F N and t ˆ 1Y 2Y F F F T for industry j …2† where i indexes ®rms and t indexes time periods, yijt is the output in logarithms ( for ®rm i in industry j at time t) and xkijt is a vector of k inputs (i.e. labour and capital) in logarithms. vijt is the usual normally distributed statistical noise accounting for random disturbances, measurement errors and minor omitted variables, uijt (uijt 5 0) represents technical ineciency of the ®rm which is not only di€erent across ®rms but also time variant. The technical ineciency uijt implies that the output of any ®rm i in an industry j must lie on or below the frontier, aj ‡ bH xkijt ‡ vijt . The above model can be written as yijt ˆ aijt ‡ b xkijt ‡ vijt
H H

…3†

where aijt …ˆ aj À uijt ) is the eciency level of a ®rm which is not only di€erent across the ®rms but also changing over time. This change in eciency level over time of the ®rm can be because of accumulated learning, the e€ect of competition, capacity utilization or any other factor. The feature can be realized by introducing a ¯exible (e.g. quadratic) function of time into the production function, with coecients varying over ®rms (Cornwell et al., 1990). The ¯exible function can be thought of as representing eciency growth, at a rate that varies over ®rms and it implies that level of eciency for each ®rm varies over time,6 i.e. aijt ˆ yij1 ‡ yij2 *t ‡ yij3 *t ˆ Wt yij where Wt ˆ …1Y tY t † and yij ˆ …yij1 Y yij2 Y yij3 †
H 2 H 2 H

…4†

5 The study by Haddad and Harrison (1993), which uses panel data from 1985 to 1989, employs the Schmidt and Sickles (1984) model assuming that productivity of the ®rm is time-invariant. Given the short period of the study, the assumption may not be too imposing but availability of longer time period as in the present case facilitates to get away with the assumption. 6 From equation (4), it is clear that the intercept for each ®rm is quadratic in time, but the form of the quadratic varies over ®rm. The form of quadratic itself depends on the coecients yijs. This also implies that neither the maximal ( frontier) intercept nor the pattern of technical ineciency for a given ®rm may necessarily be quadratic.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers

349

Cornwell et al. (1990) have suggested that without imposing a structure on the error term and without assuming uncorrelatedness of error term with the regressors, applying least squares on the deviation form model (i.e. within estimation) would give ” ” ” consistent and ecient estimates of bH . After obtaining bH , the residuals (Yijt À bH Xkijt) can be used to derive ®rm-speci®c time-variant technical eciency using equation (4). ” ” ” ” If a1jt , a2jt , a3jt , . . . , aNjt are the ®rm speci®c, time variant technical-eciency indices, then the most ecient ®rm (MEF) in the industry j for the year t would be ” ” ” ” a a ajt ˆ max…” ijt † ˆ max…” 1jt Y a2jt Y a3jt Y F F F Y aNjt † …5†

The dispersion in productivity or relative technical eciency of ith ®rm in time t would be given by ” ” a uijt ˆ …” ijt À ajt † …6†

If the best ®rm is treated as 100 per cent ecient, the relative technical eciency would be ” ” pijt ˆ exp…uijt † …7†

This is nothing but Farrell's measure of technical eciency given by OA/OD (see Figure 1). Thus, the change in dispersion in productivity between the initial period (i.e. period 1) and the terminal period (T) would be ” ” ” Dpij ˆ pijT À pij1 …8†

To test for the hypothesis that presence of a foreign-owned ®rm (denoted by Spil1) or technology import by a ®rm in the sector (denoted by Spil2) would result in fall in dispersion of productivity over time, the model would be ” Dpij ˆ f……Spil1†j Y …Spil2†j † …9†

The argument that spillovers captured by each domestically owned ®rm in an industry is directly proportional to the amount of subsidiary production/sales in the industry and available foreign technical capital stock in the sector embodies the notion that such spillovers have a public-good nature, i.e. consumption by one domestically owned ®rm does not reduce the amount available for other ®rms. Though knowledge (or technology) may be a public good to a certain extent, it would still need research activities/investment by the ®rm to decodify it and use for its purpose (Wang and Blomstrom, 1992). Thus, it is very much likely that a ®rm that engages in R&D activities would bene®t more from these knowledge spillovers. In order to account for it, two interaction terms with R&D, i.e. Spil1  R&D and Spil2  R&D are employed to capture the possible e€ect of role of R&D in enhancing the ability to absorb the spillovers. Thus, the model to be estimated becomes ” Dpij ˆ f…Spil1j Y Spil2j Y …Spil1j  R&Dij †Y …Spil2j  R&Dij †† …10†

The estimation of the above equation may not give consistent estimates as there are several other factors that also a€ect the dispersion level. Caves and Barton (1990)
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

350

V. Kathuria

argue that there are four sets of circumstances, the absence of which results in technical-inecient ®rms in an industry. These circumstances are related to lack of competitive pressure in the industry, incomplete level of di€usion of innovations, incomplete contracts ( principal-agent problem) within a ®rm and distortionary regulatory framework in the industry (Caves and Barton, 1990). External competition in the form of presence of foreign ®rms (or threat of entry) (Spil1) not only substitutes for the lack of competition in the industry but also forces the members of coalition (i.e. the factors of production) in the ®rm to be more ecient. As many important innovations take place outside the country, the foreign subsidiaries due to their multimarket presence also act as an important source of di€usion of these innovations. The di€usion of innovations is also facilitated by the foreign disembodied technology import in the sector (i.e. the technology brought in through arm's length transactions, Spil2), thus leading to reduced dispersion level of the ®rms. Besides these, there are other ®rm speci®c factors like ®rm's own R&D e€orts, its outward orientation or its human capital formation etc., which a€ect the dispersion level of a ®rm. The remaining section brie¯y discusses the various ®rm speci®c controlling variables that have been used in the study.

2.3 Factors A€ecting Dispersion 2.3.1 Foreign equity participation (FEqty) and disembodied technology import (KS) A foreign owned ®rm or a ®rm having foreign collaboration would be more ecient as it has an access to a host of tangible and intangible assets of the parent ®rm Ð e.g. technical know-how, marketing and managerial skills, reputation etc. Alternatively, a ®rm that goes for technical collaboration gets access to better technology thus making it more ecient. In fact, the magnitude of spillovers (i.e. technology di€usion from technology recipient ®rms to domestic ®rms) is likely to depend on the magnitude of productivity advantages exhibited by these ®rms. The present study uses extent of foreign equity participation (FEqty) and stock of foreign technology import (KS) as controlling variables to analyse the impact of foreign and technical collaborations. 2.3.2 Capital-vintage e€ect (AGE) When an innovation is introduced, it is generally adopted by newer plants, while other units tend to delay as it may be costly to scrap products of not so recent-vintage (Caves and Barton, 1990). This implies that the eciency level of ®rms' decreases with their age (AGE). The present study therefore, includes age as a variable to study the `vintage-e€ect' on productivity dispersion. 2.3.3 Embodied technical import (CGImp) Technology can also be acquired in embodied form, i.e. in the form of capital goods import. Till 1990, capital import was the most preferred form of technology acquisition in India as there were relatively more restrictions on other modes of technology import (e.g. on FDI or arm's length technology transfer). A ®rm having a larger stock of embodied technology is likely to be more ecient. The present study uses capital goods import intensity (CGImp) as a variable to account for embodied technology import.
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers

351

2.3.4 International orientation (EXP) Exposure to international trade is an important force imposing competition on domestic producers. This is due to the fact that world markets bring local producers into competition with a large, shifting and unfamiliar group of foreign rivals. International orientation has also an indirect e€ect on the eciency of the ®rm. The choice set of foreign technologies available to an export-oriented ®rm would be larger leading to better choice in technological solutions. This would ®nally result in improved eciency of the ®rm. The present study uses export intensity (EXP) as a variable to see the e€ect of ®rm's outward orientation in explaining the change in eciency level. 2.3.5 R&D activities (R&D) The evidence suggest that in order to use an imported technology eciently, the importing ®rm is generally required to adopt the technology to local conditions with own R&D e€orts (Desai, 1980; Wang and Blomstrom, 1992). Moreover, R&D activity is also a source of technological change provided the motive is not to take advantage of the ®scal incentives like accelerated depreciation or tax rebate or technology import at concessional duty rates etc. Most of the studies, have used R&D intensity as a variable explaining ®rm's productivity and have found the positive role of own R&D in enhancing capabilities. The present study calculates R&D stock (R&D) of each ®rm based on past R&D expenditures to see its impact on change in relative eciency.7 2.3.6 Human capital (SKIL) There are a number of incremental innovations that are not codi®able. Dosi (1988) also argues that these certain aspects of technical change which are not traded as such, are `privately appropriable' under certain conditions. A complex proprietary technology of a multinational ®rm or learning embedded in the skilled labour are two such examples. Thus, a ®rm that has higher content of skilled manpower (SKIL) is likely to become more ecient over time and hence, would be able to close the eciency gap. 2.3.7 Catch-up variable (TINI) The customary approach in the literature on technology transfer and di€usion is that the di€usion of technology (or knowledge) follows a logistic curve. This implies that the ®rms, which are at a lower level of eciency, can gain the most in terms of productivity enhancement (see for example, Abramovitz, 1986; Dollar and Wol€, 1988). Alternatively, a ®rm that is highly technically inecient (but economically viable) may not experience signi®cant productivity gains for a number of reasons. An ecient ®rm may have a higher threshold for its perception of threats as well as opportunities and hence a lower level of aspirations. Similarly, the options available for (re)combining into new and more reproductive con®guration may itself be limited
7 As per the statutory requirement in India, ®rms in India are required to report their R&D expenditure only if it exceeds one per cent of total sales. Thus some of the ®rms in our sample might be incurring R&D expenditures but would be classi®ed as having no R&D. The sample data also shows that of the 192 ®rms which have recognized R&D units only 115 ®rms have ever spent on R&D during the study period. While interpreting the results, the above caveats need to be kept in consideration.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

352

V. Kathuria

to inecient ®rms (Caves and Barton, 1990).8 Thus, it is dicult to predict which of these factors in¯uence more. In order to test for this `catch-up' hypothesis, an initial level of technical eciency (TINI) has been used in the analysis. Besides these factors, the openness of an industry to foreign competition can also have substantial impact in reducing the dispersion level.9 However, as the present study covers a period in which India's FDI policy was more selective in nature and very few foreign investment proposals of the form of equity participation were approved during the period, rendering this channel of improving eciency virtually ine€ective. Various Reserve Bank of India (RBI) bulletins giving industry-wise distribution of foreign controlled rupee companies (FCRCs) reveal that the number of foreign controlled rupees companies during the study period have remained constant. As per the bulletins in 1980±81 there were 224 FCR companies in the engineering sector and in 1988±89 the number of companies were reduced to 217. Thus the ®nal model to be estimated for analysing spillover e€ects is ” Dpij ˆ f…Spil1j Y Spil2j Y …Spil1j  R&Dij †Y …Spil2j  R&Dij †Y FEqtyij Y KSij Y AGEij Y EXPij Y CGImpij Y R&Dij Y SKILij Y TINIij1 † …11†

The productivity spillovers as per the concept has signi®cance mainly for domestic ®rms, i.e. the ®rms that have not gone for any foreign equity participation or have very little foreign equity participation (de®ned as non-FDI ®rms). Thus, the above model is estimated for the non-FDI ®rms. The exact de®nition of non-FDI ®rms is given in the next section. The model is estimated in two stages. In stage one, the stochastic production frontier for each industry is calculated employing the methodology given by Cornwell et al. (1990). The second stage involves estimation of equation (11) to see the impact of spillovers and other ®rm speci®c variables in a€ecting the dispersion in eciency. The ordinary least squares (OLS) techniques with correction for heteroscedasticity is used in the second stage. All the variables10 employed in the model (equation (11)) (except TINI) are the averages over the study period. 3 DATA AND VARIABLES

The estimation of equation (11) involves a great deal of data cleaning and construction. The ®rst step in the selection of ®nal sample involved excluding the industries reserved for the small-scale sector such as tobacco, leather and leather products, matches etc. Moreover, these industries have non-existent foreign equity participation. Few industries like steel, cement, etc. with no foreign ®rm are also dropped as we need information on the size of productivity gap, i.e. eciency di€erence between local ®rms and foreign aliates. Thus, the ®nal analysis involves
8 These ideas can be found in Carnegie School, Leibenstein, Nelson and Winter and others. The important works in the area are Cyert and George (1969), Bower (1970), and Nelson and Winter (1982). 9 In the short-term, entry of foreign ®rms may lead to increase in the dispersion but over time when local ®rms learn from the ecient foreign ®rms or compete with the foreign ®rms, the dispersion may fall. 10 As spillovers or technology import or R&D investment by a ®rm does not yield results immediately, it takes time for the ®rm to decodify and generate the knowledge/technology. These variables have been averaged with one-year lag.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers

353

use of annual report data for 368 large sized ®rms having a paid-up capital of more than Rs. 10 millions11 for 14 years from 1975±76 to 1988±89 belonging to 26 threedigit level manufacturing industries. The data has been provided by Institute for Studies in Industrial Development (ISID), Delhi, India. Of the 368 ®rms, 150 ®rms12 (i.e. nearly 41 per cent) have a foreign equity participation of 25 per cent or more during the study period. In terms of the representation, the foreign ®rms included in the sample account for nearly 66 per cent of the total foreign ®rms covered by RBI in its several studies. However, they account for more than 90 per cent of the total foreign ®rms' sales during the study period (RBI 1984, 1988, 1991, 1992). In terms of representation of the population, the only source that has ever carried out a census of operating units in India is RBI. As per the census done by RBI in 1976, there existed 356 large-sized non-®nance companies in operation in India in 1971±72.13 The list also included ®rms belonging to trading, textiles and tea plantation. The present study uses 368 large-sized ®rms belonging to 26 sectors (excluding trading, tea plantation, textiles, SSI reserved sectors etc.), that account for over 90 per cent of the paid-up capital and market share of the large sized ®rms for the year 1975±76. The sample ®rms cover a broad range of manufacturing sector: 8.4 per cent of the observations come from automobile, 14.7 per cent from electrical machinery, 20.7 per cent from non-electrical machinery, 14.4 per cent from metals, 24.2 per cent from chemicals, 7.9 per cent from drugs and pharmaceuticals, 5.4 per cent from paper and paper products and 4.3 per cent from rubber and plastic products. Table 1 gives the industry-wise coverage of foreign ®rms and their market share. It can be seen from the table that except in ®ve sectors, the foreign presence in number is less than 50 per cent. All the variables used in the stochastic production frontier and in testing the spillover hypothesis are measured in constant 1975±76 prices. The procedures used to obtain output and input data and the measurement of foreign technical capital and R&D stock are discussed at a greater length in the Appendix. Gross value-added has been used as a measure of output. Labour input has been obtained by dividing total wage bill of a ®rm by a three-digit level wage rate (as calculated from various issues of Annual Survey of Industries, ASI) corresponding to the industry to which the ®rm belongs. A net capital stock series is generated as a measure of capital input of the ®rm. The capital stock as reported in the annual report is at their purchase prices (i.e. the historical cost of the capital). As discussed in detail in Appendix, this reported capital stock is re-calculated to bring to constant 1975±76 prices. The annual report of a ®rm gives the lump-sum ®gures on disembodied technology import, i.e. expenditure on foreign patents, royalties, technical and consultancy fees etc. These ®gures have been used to construct an index of disembodied foreign technical capital stocks (KS1) using a perpetual inventory method (see Appendix).
11 This de®nition of large size is ocially used by the Reserve Bank of India to categorize the ®rms into small, medium and large sized. In the present sample, two ®rms had paid-up capital slightly below Rs. 10 million, but have been considered as the large-sized ®rms. 12 In case of three ®rms the equity participation fell below 25 per cent during the study period but to avoid possible loss of degrees of freedom, these ®rms have been retained in FDI group only. 13 These 356 ®rms (10.6 per cent) accounted for over 60 per cent of the market share and 66 per cent of the total paid-up capital for the year 1971±72. The other two categories Ð small and medium sized ®rms though formed 32.4 and 57 per cent of the total operating units, accounted for 1.7 and 38 per cent market share respectively. In terms of paid-up capital, they accounted for 1.2 and 30 per cent of the total paid-up capital (RBI, 1978).

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

354

V. Kathuria
Table 1. Distribution of sample. Final sample (No.) (2)
10 21 11 5 38 9 5 62 7 5 12 18 11 7 9 12 6 25 29 10 20 5 4 15 5 7 368

Id

Industry (1)

Foreign ®rms (No.) (3)
3 11 2 4 15 4 1 40 2 1 1 2 4 1 3 2 2 9 19 5 12 2 2 1 1 1 150

(%) (4)
30.00 52.38 18.18 80.00 39.47 44.44 20.00 64.52 28.57 20.00 8.33 11.11 36.36 14.29 33.33 16.67 33.33 36.00 65.52 50.00 60.00 40.00 50.00 6.67 20.00 14.29 40.76

Sales share (%) (5)
20.32 72.29 36.13 92.23 61.79 45.54 10.39 70.26 17.98 7.74 3.23 9.38 54.64 19.95 31.19 12.80 72.82 50.50 84.48 47.62 77.60 52.98 24.23 4.65 1.35 25.13 38.73

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 Total

Automobiles Automotive components Electric cables Dry cells and batteries Electrical machinery Machine tools Textile machinery Non-electrical machinery Steel tubes and pipes Steel wire ropes Steel forgings Foundries & engineering works Metal products Chemical fertilizers Dyes and dyestu€s Man-made ®bres Plastic raw materials Basic industrial chemicals Drugs and pharmaceuticals Paints and varnishes Soaps and toiletries Tyres/tubes Rubber products Paper & pulp Paper products Plastic products

Similarly, R&D stock variable (R&D) has been constructed based on the past expenditure incurred by the ®rm on R&D activities (see Appendix). The construction of various ®rm-speci®c variables used in the study is discussed below. 3.1 Firm-Speci®c Variables14 3.1.1 De®nition of a foreign-owned ®rm Given the objective of present study to measure the spillovers from the presence of foreign-owned ®rms, the categorization of a ®rm as foreign-owned (or domestic) ®rm is an important issue. In India, the Reserve Bank of India (RBI) de®nes foreign controlled rupee companies (FCRCs) as joint stock companies registered in India in which 25 per cent or more of equity capital is held abroad by a foreign company or its nominee or 40 per cent of the equity is held outside India. The present study also uses
14 It is to be noted that wherever industry sales has been used in the calculation of variables, e.g. spillover variables, etc., it is industry sales of the sample ®rms not the total industry sales.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers
Table 2. Variable
” Dpij Spil1 Spil2 Size Interact1 Interact2 FEqty KS CGImp SKIL AGE R&D EXP TINI

355

Description of spillover and ®rm speci®c variables.

Description of variable
Change in ®rm's relative technical eciency in the industry over the period, i.e. ” ” (pijT À pij1). Spillover variable 1; share of sales of foreign ®rms to total industry sales. Spillover variable 2; foreign technical capital stock of rest of the ®rms in the industry to the total industry sales. Size of the ®rm; net ®xed assets (NFA) of the ®rm to the largest ®rm in the industry. Interaction term 1 ˆ Spil1  R&D. Interaction term 2 ˆ Spil2  R&D. FEqty ˆ 1, if the foreign equity participation in the ®rm is 25 per cent or more during the study period; and 0, otherwise. Own foreign technical capital stock of the ®rm (KS1) as a ratio of total capital stock of the ®rm. Capital goods import intensity; Ratio of capital goods import to the annual sales turnover of the ®rm. Skill levels of the ®rm; Ratio of salary and wages of high-income employees* to the total salary and wages of the ®rm. Capital vintage e€ect; proxied by ratio of accumulated depreciation to the value of total plant, machinery and equipment. R&D stock of the ®rm. Export intensity; exports as a ratio of total sales turnover. Initial technical eciency; Technical eciency of a ®rm relative to the most ecient ®rm's eciency in the ®rst year of the study.

*The Companies act of India makes it mandatory for the ®rms to report in their annual reports the number of employees who get Rs. 3000 per month or higher, and the total salaries of these employees. These employees are termed as high-income employees (HIE). The cut-o€ ®gures, however, keeps on changing time-to-time. In the present case, there is no change in the cut-o€ ®gures during the study period.

the same cut-o€ to de®ne a foreign-owned ®rm. A ®rm is de®ned as non-FDI or domestic if it has zero or less than 25 per cent of foreign equity participation. Spillover variable has been constructed in the literature in di€erent ways using R&D expenditure or patents registered by other ®rms (see Haksar, 1995, for a brief review). The present study uses two spillover variables. The ®rst spillover variable (Spil1) i.e. the spillovers due to the presence of foreign ®rms in the sector is constructed as a ratio of sales share of foreign-owned ®rms to the total industry sales. The other spillover variable, Spil2 is constructed as foreign disembodied technical capital stock of rest of the ®rms in the sector to the total industry sales. Table 2 gives a brief description of measurement of various spillover and ®rm speci®c variables used in the study. All variables are in rupees (Rs.) terms and are proxies/surrogates for the variables in question. The use of proxies has been necessitated as the variables like spillovers or own technological e€orts or learning from disembodied technology import etc. are not easily quanti®able. The descriptive statistics of the two spillover variables, Spil1 and Spil2 for non-FDI ®rms are given in Table 3. 4 COMPARISON OF FDI AND NON-FDI FIRMS

The notion of (technological) spillovers assumes that the foreign ®rms or foreign technology recipient ®rms are at the frontier or closer to it thereby, paving signi®cant
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

356

V. Kathuria
Table 3. Variable (%) Descriptive statistics of spillover variables. Std. Deviation (2)
27.332 0.703

Mean (1)
40.707 0.764

Minimum (3)
1.35 0.00

Maximum (4)
92.23 3.22

N (5)
218 218

1 2

Spil1 Spil2

Table 4. Id

Performance comparison of two groups of ®rms. RTE (%) FDI (1) Non-FDI (2)
59.18 57.34 20.93 16.24 35.87 31.72 37.21 34.43 76.29 51.37 37.75 18.70 4.80 32.14 31.36 45.02 49.20 34.06 45.26 53.58 31.40 65.44 72.24 11.45 67.45 31.03

Leadera (3)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

73.90 58.04 0.42 45.23 50.66 51.65 37.34 43.77 57.39 49.07 14.78 26.75 26.74 100.00b 50.19 35.09 32.04 31.70 34.63 59.40 34.01 61.02 85.71 0.04 40.86 100.00b

F F N F F F N F N N N N F F F N N N N F F N F N N F

Notes: Leader ®rm is based on average productivity level of the group as well as leading ®rm in the industry. a F, FDI; N, non-FDI ®rm. b Implies there is only one foreign ®rm in the sector and is the leader.

scope for local ®rms to learn from them. In order to check whether FDI ®rms are indeed leaders or not, the MEF in each sector is calculated using equation (5). Table 4 tabulates the average technical eciency relative to the MEF for each sector in the ®rst year of the study. Columns 1 and 2 of the table give the average relative technical eciency (RTE) for FDI and non-FDI ®rms respectively (see equation (6)). The detailed analysis of
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers

357

results of equations (5) and (6) indicates that in ®ve sectors (i.e. Id 4, 5, 6, 14 and 26) FDI ®rms are not only the leading ®rms but also they are on the whole closer to the frontier. In ®ve other sectors (i.e. Id 1, 2, 8, 20, 23) though FDI ®rm is not the MEF, but on an average foreign ®rm's eciency level is higher. Examining the results closely (i.e. the position of each ®rm in the sector), it is found that there are three more sectors (sectors 13, 15 and 21) where average RTE of FDI ®rms is higher. Hence of the 26 industries, in 13 sectors foreign ®rms are closer to the eciency frontier (column 3). Thus, it is only in these 13 sectors one can expect productivity spillovers for domestic ®rms. Table 5 reports the relative performance of the two groups of ®rms. The indicators used to compare the relative performance are: export intensity (exports as a percentage of total sales), extent of vertical integration (ratio of value added to total sales), capital goods import intensity (capital goods imports as a percentage of total sales), employee compensation behaviour (share of salaries and wages of high income employees (HIEs) to total salaries and wages), technical eciency (TE) relative to the MEF and level of technological capabilities as proxied by R&D and foreign technological capital stock. Column 1 of Table 5 reports relative performance indicators as a ratio of FDI ®rms to non-FDI ®rms covering all the industries. Figures in parentheses are relative performance indicators of the two groups of ®rms for those 13 industries for which one can expect spillovers.

Table 5. Indicator
1 2 3 4 5 6 7 8 Export/sales GVA/sales HIE/tot.S&W CGImp/Sales Size KS R&D Tech. eciency

Behaviour comparison of two groups of ®rms. FDI/non-FDI (1)
0.956 (0.736) 1.235* (1.171*) 1.902* (1.645*) 1.097 (1.174) 1.023 (1.074) 1.561 (1.251) 1.544 (2.356**) 1.234* (1.327*)

Size controlled ratio (2)
0.935 (0.685) 1.208 (1.090) 1.86 (1.532) 1.073 (1.093) 1.000 (1.000) 1.527 (1.165) 1.510 (2.193) 1.207 (1.236)

Note: *, **Indicates that the di€erence in means is statistically signi®cant at the 1 per cent level and 10 per cent level, respectively (two-tail t-test).

Comparison of the two groups show that on average foreign-owned ®rms pay higher salaries (row 3), are more vertically integrated (row 2), have higher stock of R&D capital (row 7) and are closer to the MEF (row 8). However, in terms of export intensity (row 1), embodied and disembodied technology import (rows 4 and 6), the two groups do not di€er statistically. It is well established that foreign ®rms in general, are larger in size. Thus, the superior performance of foreign owned ®rms may be due to their size. The data however, does not indicate that foreign owned ®rms are signi®cantly larger than non-FDI ®rms (row 5). After accounting for this partial size di€erence, the indicators are recalculated to get the weighted means Ð with weights given by average size. Performance indicators as expected change only marginally (column 2).
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

358 5

V. Kathuria STATISTICAL RESULTS

5.1 Spillover E€ects of Technology Transfer on Dispersion of Eciency 5.1.1 Non-FDI ®rms (sectors with FDI ®rm as leader) The results of estimation of equation (11) examining the productivity spillovers impact of technology transfer are reported in Table 6. Column 1 gives the results for the non-FDI ®rms belonging to those 13 sectors for which one can expect spillovers, i.e. foreign ®rms are closer to the frontier. Before going into the details of the results, it needs to be mentioned that the two spillover variables (or interaction terms) as used in the analysis capture two di€erent mode of learning from two di€erent, but not entirely mutually exclusive source; hence have some degree of overlap. Thus, the variables are neither complementary in nature nor perfectly substitute each other. Further, in some of the estimations, only one of the spillover variable (or interaction term) is reported, this is necessitated due to the severe multicollinearity between the spillover variables (and/or interaction terms). Rows 1 to 4 of Table 6 report the coecients for spillover variables. The negative and highly signi®cant sign of the spillover variable, Spil1 (row 1) suggests that larger the sales share of foreign ®rms in the sector negative would be the spillovers for the local ®rms. One possible explanation for the negative spillovers could be that if foreign ®rms and domestic ®rms are competing in the same market then presence of foreign ®rms may have forced local ®rms to operate at higher level of their average cost curves. The sign and signi®cance level of other spillover variable, Spil2 (row 2) indicates that ®rms experience positive spillovers from the available foreign technical capital stock in the sector. This implies that higher is the technology transferred through arm's length transactions, larger is the knowledge spillovers to the domestic ®rms, resulting in fall in eciency gap. Though the interaction term (Interact 1) to see the possible complementarities between spilled knowledge and ®rm's R&D e€orts has come with the expected sign, is not statistically di€erent from zero.15 Rows 5 to 9 give the coecients of various ®rm speci®c variables that can have impact in reducing the eciency gap of a ®rm. From the sign and the signi®cance level, it appears that the ®rm's embodied technical import (CGImp) has been a signi®cant factor facilitating the ®rm to reduce the dispersion level over time (row 6). However other ®rm speci®c variables like export intensity, disembodied technology import, capital vintage e€ect and the skill content of the ®rm do not seem to have any impact. Row 10 of the table gives the coecient of initial level of technical eciency (TINI). The negative and highly signi®cant sign of the variable suggests that a ®rm, which is away from the frontier, would experience less eciency gain. As argued earlier, the pursuit of innovations and productivity raising opportunities for a laggard ®rm would depend on the action taken by rival ®rm and the perceived threats and opportunities they reveal. Given the level of disembodied technology import and R&D activities which is not only lower for this group of ®rms but also relatively less varying across ®rms, precludes the possibility of perceived threat or productivity raising investments by the other ®rms. The restriction on entry of foreign ®rms as imposed by the policy regime during the study period, may have further lessened the possible threat to (inecient) ®rms.
15 As interaction term (Interact 1) is not signi®cant, the model has been re-estimated with R&D as an explanatory variable instead of Interact 1. The coecient of R&D is however, found positive and highly signi®cant (results not reported here). Both R&D and Interact 1 could not be used together because of high collinearity.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers
Table 6. Variable Estimation results for productivity spillovers (FDI ®rm as leader). Non-FDI ®rms coecienta (1)
À0.516*** (À4.671) 6.567** (2.693) 0.657 Â 10 À4 (0.856) ±b À0.622 (À1.581) 2.057*** (3.14) À2.521* (À1.499) À0.117 (À0.934) À0.503*** (À5.784) 35.3*** (4.249) 0.361 8.449 108

359

Non-FDI ®rms showing productivity growth coecienta (2)
À0.545*** (À4.151) 6.847*** (2.598) 0.177 Â 10 À4 (0.207) ±b À0.248 (À0.693) 1.506** (2.333) À12.038* (À1.829) 0.619*** (3.027) À0.252** (À2.404) 39.763*** (3.133) 0.274 4.247 68

1 2 3 4 5 6 7 8 9 10 11 12 13 a bInteract

Spil1 Spil2 Interact1 ˆ Spil1  R&D Interact2 ˆ Spil2  R&D KS CGImp AGE EXP TINI Constant Adj. R-sq. F N

Figures in parentheses are t-ratios and ***, **, * are signi®cance levels at 1, 5 and 10 per cent respectively. 2 is found to be highly correlated with Interact 1, hence has been dropped from the estimations. Note: SKIL variable does not attain signi®cance in any variant of the model, hence has not been 16 reported.

The above results suggest that there exists positive spillovers from the available foreign technical capital stock in the sector but presence of foreign ®rms has a negative impact on the dispersion in eciency level. However, one can argue that the dispersion being a relative concept, it may still fall if both the leading FDI ®rm and domestic ®rms show fall in technical eciency over the period and the fall for the leader is higher resulting in fall in dispersion.17 In that case, it would be a probable case for negative spillovers. Similarly if both the ®rms experience eciency gain over time, but the gain is more for the frontier ®rm. In that case the gap would appear to have increased but the domestic ®rms in fact, may have learnt from the spilled knowledge. Given the focus of the study, where concern is for the learning by the
16 The variable `SKIL' comes out insigni®cant in all the variants of the model. This may imply that the variable represents more of employee compensation behaviour of MNCs rather than the `skill intensity'. A detailed information about the education level of HIEs can shed more light on the skill intensity. 17 For instance, if the frontier ®rm, F has a productivity score of 10 and domestic ®rm has a productive score of 8 in the ®rst year of the study then the dispersion would be 2 (ˆ10 À 8). If at the end of terminal year the score changes to 8 and 7 respectively then the dispersion would be 1 (ˆ8 À 7). Though the dispersion (i.e. the gap) has fallen from 2 to 1 but it would be a case for negative spillovers as the productivity of domestic ®rms has reduced.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

360

V. Kathuria

domestic ®rms, the study tries to get around with the problem partially, by testing the hypothesis for those local ®rms that have shown productivity improvement over the period. In such a situation, the hypothesis to be tested also becomes conditional that for the sectors where foreign ®rms are at the frontier and domestic ®rms have shown improvement in eciency level, has the spillovers from technology transfer resulted in fall in dispersion.18 Results also show that of the 13 sectors having 108 non-FDI ®rms (where one could expect spillovers), it is only 68 local ®rms that have experienced eciency-growth over the period. Thus, in the next stage equation (11) has been estimated for these 68 non-FDI ®rms only. The results of estimations are given in column 2 of Table 6. 5.1.2 Non-FDI ®rms showing productivity growth in sectors with FDI ®rm as leader The spillover results however, do not change as the sign and signi®cance level of both the spillover variables remain same (rows 1 and 2). The interaction term (Interact 1) again comes out to be positive but is still not signi®cantly di€erent from zero in statistical terms (row 4). The insigni®cance of interaction term also raises the validity of our hypothesis of complementarity between R&D and spilled knowledge. Among the ®rm speci®c variables export intensity has come out to be positive and highly signi®cant suggesting the positive role played by competitors in the export markets leading to fall in dispersion (row 6). The embodied technology import (CGImp) has again come out to be positive and signi®cant. The highly signi®cant negative sign of initial level of technical eciency (TINI) lends support to Caves and Barton argument. The capital vintage e€ect (AGE) has a signi®cantly negative sign indicating that the dispersion in eciency of a ®rm having capital of older vintage has increased. Based on above results one can say that the presence of foreign ®rms has a negative spillover impact on the dispersion of eciency of local ®rms whereas available foreign technical capital stock results in learning by the local ®rms leading to reduced dispersion level. One of the important channels of spillovers or di€usion of knowledge is through employee mobility. Given the strategy of MNCs to pay higher than their local counterparts (row 4, Table 5), employee mobility from MNCs to nonMNCs may be very little, thereby closing this channel of technology di€usion for local ®rms.19 Dunning (1981, pp. 282±3) has also suggested that one of the reason for MNCs paying higher is the need to reduce labour turnover. Another possible explanation for negative spillovers from the presence of foreign ®rms can be the characteristic of the sample ®rms as the sample includes all the ®rms from all the sectors irrespective of their technological and production requirements. There are sectors like drugs and pharmaceuticals, chemicals, etc. where a strong technological base is needed in the form of greater skill content and R&D activities not only to carry out the production but also to facilitate any spillovers. Alternatively, there are sectors like paper, metal products, etc. requiring less capabilities on the part of the
18 It is to be noted that the sample still includes the cases where frontier ®rm has grown faster than the local ®rms. But as the de®nition of frontier ®rm is not very tight (see Section 4) and in some cases the position of frontier ®rm itself is very ¯uctuating, such cases could not be taken out. 19 There is another viewpoint that indirectly dismisses any possibility of `competitive' spillovers to domestic ®rms from the presence of foreign ®rms. According to this view, domestic and foreign ®rms belong to two di€erent strategic groups and catering to two di€erent market segments (Kumar, 1994). Firms within a particular strategic group recognize their mutual interdependence more than those between the strategic groups and hence, learn more from the ®rms in that group only.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers
Table 7. Variable Testing for spillovers for scienti®c and non-scienti®c groups. Scienti®c ®rms coecienta (1)
À0.636*** (À3.074) 11.379** (2.236) 0.104 Â 10 À3** (1.974) ±b 0.416 (0.360) À12.374 (À1.649) À12.682 (À0.846) 0.788 (0.936) 0.0716 (0.367) 23.338 (0.918) 0.475 4.436 32

361

Non-scienti®c ®rms coecienta (2)
À0.107 (À0.739) ±c 0.679 Â 10 À5 (0.0787) ±b À1.887** (À2.453) À0.459 (À0.639) À29.324** (À2.369) 0.0008** (2.118) À0.185 (À1.476) 48.072 (3.184) 0.112 5.463 36

1 2 3 4 5 6 7 8 9 10 11 12 13

Spil1 Spil2 Interact1 ˆ Spil1  R&D Interact2 ˆ Spil2  R&D KS CGImp AGE EXP TINI Constant Adj. R-sq. F N

Note: see notes to Table 6. cSpil2 is found to be highly correlated with Spil1, hence has been dropped from the estimations.

®rm to absorb the spillovers. Thus, to gain some insights into the nature and e€ect of spillovers, i.e. which sectors entail larger bene®t, the sample has been bifurcated into two broad subgroups: `scienti®c' and `non-scienti®c'. The `scienti®c' subgroup consists of ®rms in the drugs and pharmaceuticals, chemicals, electronics industries, etc. and `non-scienti®c' subgroup comprises ®rms in the automobiles, non-electrical machinery, metal products, etc.20 The bifurcation of the sample also facilitates to see whether any `technology di€usion' has taken place or it is more of `competitive e€ect (or demonstration e€ect)' that has resulted in spillovers. This is due to the fact that in `non-scienti®c' sectors where the technological innovation is `product-centred', ®rms may learn by reverse engineering (demonstration e€ect) or the level of technological requirement is itself not very high making ®rms to be more in direct competition with the foreign-owned ®rms. On the other hand, in scienti®c sub-sectors, where the innovations are `process oriented', the learning for the local ®rms would be mainly through technology di€usion. It is also very much likely that a weak patent regime in India may have enhanced the imitation potential of foreign technologies in the scienti®c sectors. Table 7 reports the estimation results for both sub-groups of ®rms. Column 1 reports the results for `scienti®c' sub-group and column 2 reports the results for `non-speci®c' sub-group.
20 Basant and Fikkert (1996), Griliches and Mairesse (1984), etc., also have used the same groupings in their analysis.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

362

V. Kathuria

5.1.3 Non-FDI scienti®c and non-scienti®c ®rms showing productivity growth Rows 1 to 4 of Table 7 give the coecients of spillover variables. From the coecients and their signi®cance level, it appears that the incidence of spillovers do not change for scienti®c subgroup. However, the interaction term becomes positively signi®cant. This suggest that presence of foreign ®rms itself may have negative impact but the ®rms which are engaged in research activities seem to have bene®ted from the spilled knowledge. Thus, indicating the possible complementarities between spilled knowledge and the research e€ort of the ®rm for the sectors, where technological and production requirement are complex. Based on the coecient it seems that for a 10 per ” cent increase in the sales share of foreign ®rms in the sector at the mean of Dpij , a ®rm having average sectoral level of R&D stock would be nearly 7.0 per cent closer compared with the ®rm which does not engage in R&D activities. However, the results change signi®cantly for non-scienti®c subgroup. The spillover variable loses signi®cance for non-speci®c non-FDI ®rms. The interaction term is still insigni®cant. This implies that for the sectors, where the production and technological requirements are not high, it is the outward orientation (EXP) and the capital of recent vintage (AGE) that facilitate in reducing the dispersion level rather than the spillovers from the presence of foreign-owned ®rms in the sector. Thus, the results do not provide evidence on knowledge spillovers from foreign owned ®rms to non-FDI ®rms. As argued earlier, given the strategy of MNCs to pay higher than their local counterparts the employee mobility from MNCs to domestic ®rms may be very little, thereby leaving limited scope for local ®rms to absorb the spillovers. The data also shows that foreign-owned ®rms in non-scienti®c sub-sector pay signi®cantly higher (nearly 1.7 times higher) than their domestic counterparts, thereby virtually closing this channel of productivity spillovers to local ®rms. The most counter-intuitive result is the negative and highly signi®cant sign of KS variable implying that a ®rm, which has larger foreign disembodied technical capital stock would have fallen away from the frontier. One possible explanation for negative sign can be lack of capabilities of the local ®rms and possible high costs associated to absorb the technology. It is well proven that technology transfer has costs beyond the payments for blueprints, designs or concepts or machines that embody the innovation. Even if a simple technology is being transferred through licensing agreements, the transfer would not be costless. It would involve signi®cant amount of learning and capabilities on the part of recipient ®rm in order to fully absorb the technology (Teece, 1986). A larger stock of foreign technical capital may also imply that the ®rm is relying on excessive technology import. In either case, large stock would not result in higher eciency level unless ®rm is able to decodify the technology by continuous R&D. The data shows that of the 27 non-FDI ®rms that have gone for technology purchase agreements during the study period only nine ®rms have spent on R&D activities to decodify the purchased technology.21

21 Furthermore, if the motive for technology import is to capture new markets by developing new products then the impact of technology import would be on pro®tability of the ®rm and not on the eciency level. Given the diversi®cation strategy of Indian ®rms pursued in eighties and empirically established in several studies (Pandit and Sidharthan, 1994; Desai, 1988), the negative role of technology import in in¯uencing the eciency level cannot be ruled out. This indicates that the purchase of disembodied technology may have been motivated by the diversi®cation strategy rather than to improve the eciency.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers 5.2 Sensitivity Analysis

363

As the explanatory variables used in the model (Section 6) are the averages over the period (see Section 2), it can be argued that the period for which the variables have been averaged is fairly long and can introduce bias in the results. In order to see whether the results change if we use alternate de®nitions of independent variables, the model has been re-run with the independent variables averaged over the ®rst nine and six years.22 The exercise thus facilitates to verify the robustness (i.e. the sensitivity analysis) of the results. The sign and signi®cance of spillover variables for the two subgroups are reported in Table 8. From Table 8, it appears that the results obtained to test for spillover hypothesis are fairly robust and are not sensitive to alternate de®nitions of independent variables. Only in one case, the interaction term changes the signi®cance level (see row 3, column 6). On the basis of these results one can say that there exist spillover bene®ts, but mainly to the scienti®c non-FDI ®rms that invest in R&D activities to decodify the spilled knowledge. This implies that spillovers are not automatic consequence of foreign ®rm's presence and a threshold level of investment is needed for the ®rms to decodify the spilled knowledge.
Table 8. Variable Sensitivity analysis (scienti®c and non-scienti®c non-FDI ®rms only). Scienti®c ®rms Average over the period (1)
1 2 3 4 Spil1 Spil2 Interact1 ˆ Spil1  R&D Interact2 ˆ Spil2  R&D À(s) ‡(s) ‡(s) *

Non-scienti®c ®rms Average over the period (4)
± ” ‡ *

Average over 9 years (2)
À(s) ‡(s) ‡(s) *

Average over 6 years (3)
À(s) ‡(s) ‡(s) *

Average over 9 years (5)
± ” ‡ *

Average over 6 years (6)
± ” ‡(s) *

Note: s in parentheses implies that the variable is signi®cant at minimum 10 per cent level. *Interact2 is found to be highly correlated with Interact1 and hence has been dropped from the estimations. ” À Spil2 being highly correlated with Spil1, has been dropped from the estimations.

6

CONCLUDING REMARKS

This paper employs techniques from stochastic production frontier and panel data literature to test a spillover hypothesis for large sized ®rms that `presence of foreignowned ®rms and foreign technical capital stock in a sector leads to reduced dispersion in eciency in the sector and fall is higher for the ®rms that invest in R&D activities'. Results suggests that foreign-owned ®rms are not at or close to the frontier in all the sectors. It is only in 13 sectors foreign ®rms reveal higher eciency and are closer to the frontier. Spillover results for these 13 sectors indicate that there exist negative spillovers from the presence of foreign ®rms in the sector, but available foreign technical capital stock has a positive impact. Interesting di€erences emerge when the sample is
22

The choice of averaging over nine and six years is however, arbitrary. J. Int. Dev. 12, 343±369 (2000)

Copyright # 2000 John Wiley & Sons, Ltd.

364

V. Kathuria

bifurcated into scienti®c and non-scienti®c subgroups. The results indicate that there are positive spillovers due to presence of foreign ®rms in the sector, but mainly to the ®rms belonging to scienti®c subgroup provided the ®rms invest in signi®cant R&D activities. Thus suggesting the complementarities between R&D and spillovers. The available foreign technical capital stock in the sector also has a positive e€ect on the eciency level of scienti®c domestic ®rms. However, for non-scienti®c subgroup the results, in general, do not indicate the presence of spillovers from either source. From the results, one can conclude that the indirect gains or spillovers are not found to be automatic consequence of foreign ®rm's presence but they depend to a large extent on the e€orts of local ®rms to invest in learning or R&D activities so as to decodify the spilled knowledge. On the other hand, the evidence of spillovers to nonscienti®c non-FDI ®rms is not very strong. An insigni®cant coecient on foreign share for the non-scienti®c subgroup, in fact, does not preclude the possibility that some technology transfer from joint ventures (JVs) to domestic ®rms does occur. For instance, foreign investment could be associated with declining or no productivity gain on aggregate and at the same time convey substantial bene®ts to those few plants located nearby. Whether trained workers leave the JV to work at nearby domestic ®rms, or whether the JV demonstrates a product, process or market previously unknown to domestic owners, the bene®ts are likely to be ®rst received by neighbouring domestic ®rms before di€using to other distant domestic ®rms. The unavailability of plant level data, however restrains us to check this aspect. As the present study covers that period of India's industrial policy regime, which was highly regulated, one can argue that the restrictions in the form of licensing of import of technology, import of goods or capacity expansion etc., might have hindered the non-scienti®c local ®rms to gain bene®ts from spilled knowledge. Moreover, the level of ( foreign) competition itself was limited due to severe controls on the FDI and entry of foreign ®rms. Thus, it would be interesting to see how the nature of spillovers changes in the two regime ( pre- and post-liberalization). In order to get more insights into spillovers, it is necessary to do analysis for post-liberalization period. ACKNOWLEDGEMENTS I am extremely thankful to Institute for Studies in Industrial Development (ISID), Delhi, India for permitting me to use the data. A substantial part of this paper was drafted during my Ph.D. Internship at United Nations University/Institute for New Technologies (UNU-INTECH), Maastricht, The Netherlands. I acknowledge with thanks the ®nancial and other support I received from UNU-INTECH. An earlier version of the paper was presented at International conference on Economics at Middle East Technical University (METU), Ankara, Turkey, during 18±20 September 1997. The author would like to thank Dr Subir Gokarn, Dr Subrata Sarkar and Dr Veena Mishra for their useful comments and suggestions. I also thank the three anonymous referees for their valuable comments. The usual disclaimer applies. REFERENCES
Abramovitz, M. (1986). `Catching up, forging ahead and falling behind', Journal of Economic History, 46, 385±406.
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers

365

Aitken, B. and Harrison, A. (1994). `Do domestic ®rms bene®t from FDI? Evidence from panel data'. World Bank Policy Research Working Paper, No. 1248, February. Annual Survey of Industries: Summary Results for the Factory Sector, Central Statistical Organisation, Ministry of Planning, New Delhi: Government of India. Basant, R. and Fikkert, B. (1996). `The e€ects of R&D, foreign technology purchase and technology spillovers on productivity in Indian ®rms', The Review of Economics and Statistics, 78, 187±199. Blomstrom, M. and Wol€, E. (1994). `Multinational corporations and productivity convergence in Mexico'. In Baumol, W., Nelson, R. and Wol€, E. (eds) Convergence of Productivity: Cross National Studies and Historical Evidence. Oxford: Oxford University Press. Blomstrom, M. and Persson, H. (1983). `Foreign investment and spillover eciency in an underdeveloped economy: evidence from the Mexican manufacturing industry', World Development, 11, 493±501. Bower, J. L. (1970). `Planning within the ®rm', American Economic Review, 60, 186±194. Cantwell, J. (1989). Technological Innovation and Multinational Corporations. Oxford: Basil Blackwell. Caves, R. E. and Barton, D. R. (1990). Eciency in U.S. Manufacturing Industries. Cambridge: MIT Press. Caves, R. E. (1974). `Multinational corporations, competition and productivity in hostcountry markets', Economica, 41, 176±193. Chandhok, H. L. and the Policy Group (1990). Indian Data Base: The Economy (Annual Time Series Data, Vol. 1 & 2) New Delhi: Living Media India Ltd. Cornwell, C., Schmidt, P. and Sickles, R. C. (1990). `Production frontier with cross-section and time-series variation in eciency levels', Journal of Econometrics, 46, 185±200. Cyert, R. E. and George, K. D. (1969). `Competition, growth and eciency', Economic Journal, 79, 43±61. de Mello, L. R. (1997). `Foreign direct investment in developing countries and growth: A selective survey', Journal of Development Studies, 34, 1±34. Department of Science and Technology of India ( published annually), Research and Development Statistics. New Delhi. Desai, A. V. (1980). `The origin and direction of industrial R&D in India', Research Policy, 9, 74±96. Desai, A. V. (1988). `Technology acquisition and application: interpretations of the Indian experience'. In Lucas, R. E. B. and Papanek, G. F. (eds) The Indian Economy: Recent Development and Future Prospects. Delhi: Oxford University Press. Dollar, D. and Wol€, E. (1988). `Convergence of industry labour productivity among industrial countries 1963±1982', The Review of Economics and Statistics, 70, 549±558. Dosi, G. (1988). `Sources, procedures and microeconomic e€ects of innovation', Journal of Economic Literature, 26, 1120±1171. Dunning, J. G. (1993). Multinational Enterprises, and the Global Economy. Reading: AddisonWesley. Farrell, M. J. (1957). `The measurement of productive eciency', Journal of the Royal Statistical Society, 120, 253±282. Findlay, R. (1978). `Relative backwardness, direct foreign investment, and the transfer of technology: a simple dynamic model', Quarterly Journal of Economics, 92, 1±16. Globerman, S. (1979). `Foreign direct investment and `spillover' eciency bene®ts in Canadian manufacturing industries', Canadian Journal of Economics, 12, 42±56.
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

366

V. Kathuria

Griliches, Z. and Mairesse, J. (1984). `Productivity and R&D at the ®rm level'. In Griliches, Z. (ed.) R&D, Patent and Productivity. Chicago: University of Chicago Press, pp. 339±374. Haddad, M. and Harrison, A. (1993). `Are there positive spillovers from direct foreign investment? Evidence from panel data for Morocco', Journal of Development Economics, 42, 51±74. Haksar, V. (1995). `Externalities, growth and technology transfer: application to Indian manufacturing sector Ð 1975±90', International Monetary Fund. Washington, DC (mimeo). India Investment Center (1982). Directory of Foreign Collaborations in India (1951±1980). New Delhi. Kokko, A. (1994). `Technology, market characteristics, and spillovers', Journal of Development Economics, 43, 279±293. Kokko, A., Tansini, R. and Zejan, M. C. (1996). `Local technological capability and productivity spillovers from FDI in the Uruguayan manufacturing sector', Journal of Development Studies, 32, 602±611. Kumar, N. (1994). Multinational Enterprises and Industrial Organisation: The Case of India. New Delhi: Sage Publications. Lapan, H. and Bardhan, P. (1973). `Localized technical progress and transfer of technology and economic development', Journal of Economic Theory, 6, 585±595. Mody, A. (1989). `Strategies for developing information industries'. In Cooper, C. and Kaplinsky, R. (eds) Technology and Development in the Third Industrial Revolution. London: Frank Cass. Nadiri, I. (1991). `U.S. direct investment and the production structure of the manufacturing sector in France, Germany, Japan and the U.K.', New York University and NBER (mimeo). Nelson, R. R. and Winter, S. G. (1982). An Evolutionary Theory of Economic Change. Cambridge: Harvard University Press. Pandit, B. L. and Siddharthan, N. S. (1994). `Technological acquisition and investment: lessons from recent Indian experience', N. Delhi: Institute of Economic growth (mimeo). Perez, T. (1998). Multinational Enterprises and Technological Spillovers. The Netherlands: Harwood Academic Publishers. Reserve Bank of India (1978, 1984, 1988, 1991, 1992). Reserve Bank of India Bulletin. Bombay: Reserve Bank of India. Schmidt, P. and Sickles, R. C. (1984). `Production frontiers and panel data', Journal of Business and Economic Statistics, 2, 367±374. Stewart, F. and Ghani, E. (1991). `How signi®cant are externalities for development?' World Development, 19, 569±594. Teece, D. J. (1986). `Transaction cost economics and the multinational enterprise: an assessment', Journal of Economic Behaviour and Organisation, 7, 21±45. Wang, J. Y. and Blomstrom, M. (1992). `Foreign investment and technology transfer: a simple model', European Economic Review, 36, 137±155.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers APPENDIX Output23

367

Gross value added has been taken as a measure of output.24 The output is de¯ated by a three-digit industry speci®c, wholesale output price de¯ators as obtained from the Index numbers of wholesale prices in India (Chandhok and the Policy Group, 1990). Input(s) Labour The ®rms in India generally report only their total payments for wages and salaries, not the total number of workers employed. In order to calculate person-days worked for each ®rm, these total payments are divided by the average wage rate of the industry to which each of these ®rms belongs: Person-days worked ˆ Payment for wages and salaries Average wage rate

Annual Survey of Industries (ASI) data available at the three-digit level of industrial classi®cation is used to calculate average wage rate for the relevant industry groups. The average wage rate has been de®ned as the total emoluments to the workers divided by the total person-days worked: Average wage rates …from ASI† ˆ Total emoluments to workers Total person-days worked

Capital A net capital stock series is generated as a measure of capital input of the ®rm. As the capital stock reported in the annual report is at their purchase prices (that is, the historical cost of the capital). In order to generate a capital stock series, this reported capital stock need to be brought at constant 1975±76 prices. Firms report accumulated depreciation and gross capital stock (at historical cost) in their annual reports. Using this available information for 1975±76 (the ®rst year of the data set) and in absence of the knowledge of exact age distribution of the capital assets for a particular ®rm as on 1975±76, average age (AA) of each ®rm's capital stock has been calculated using following formula. Generally, for accounting purposes it is assumed that full depreciation of capital stock takes 16 years. This implies that if we assume straightline depreciation method, then capital is depreciating at a rate of 6 per cent per annum. Thus, average age of the ®rm would be AA ˆ …AD75 ±76 aGC75 ± 76 †  16
23 All the input and output variables used in the stochastic production frontier, are in log values and have been calculated using three years moving average method. The three-year moving average has been carried out to smoothen out uneven ¯uctuations in output and input levels. 24 In case of eight ®rms, a small positive value close to zero was assigned to the observations which had their three-year moving average value for value added to be negative. This has been done to facilitate to take log values for the dependent variable.

Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)

368

V. Kathuria

Using average age, AA, a price de¯ator of capital (CDi) is constructed for each ®rm's capital stock to de¯ate from the year 1975±76 Ð AA to the year 1975±76. Thus, the capital stock of each ®rm at current prices would be NC75± 76 ˆ …GC75 ± 76 aCDi †  …1 À 0.06†
AA

Similarly, net capital stock for the 1976±77 at 1975±76 prices would be NC76 ± 77 ˆ …NC75± 76 †  …1 À 0.06† ‡ …NI76 ± 77 †aPC76 ± 77 where NI76±77 is the net investment in 1976±77 and PC76±77 is the price de¯ator for the year 1976±77. Same formula is used to get the series till the terminal year of the study, i.e. 1988±89. Imported technological capital stock (or disembodied technology import) To measure disembodied technology as purchased from foreign countries through expenditure on foreign patents, royalties, technical and consultancy fees in the form of lump-sump payments, etc. a foreign purchased technical capital stock has been calculated. Perpetual inventory method is used to construct the knowledge stocks generated from technology purchase: KS1t ˆ …1 À d†KS1tÀ1 ‡ TPtÀ1 where TP is the ®rm's expenditure on technology purchased in the form of licenses from foreign countries and d is the rate of depreciation of the technical knowledge. As knowledge obsoletes or depreciates faster, taking a cue from other studies (e.g. Basant and Fikkert, 1996) a depreciation rate of 15 per cent has been assumed. As United States is the largest seller of technology to India, to bring the technology purchase expenditure at constant 1975±76 prices, rupee±US dollar exchange rate at 1975±76 exchange rate has been used to de¯ate these expenditures. To compute knowledge stock for the initial year, following procedure is adopted. Department of Science and Technology (DST) documents suggest that payments stream for a foreign licensing contract lasts for four years. This implies that for a depreciation rate of 15 per cent, any agreement signed before 1967±68 would have become obsolete by 1975±76, the ®rst year of the study period. Based on this information, the ®rms and the year during the period 1967±68 to 1974±75 for which the sample ®rms has gone for technology transfer are identi®ed. Then, using ®rm level information for the ®rst year of the study period, that is 1975±76, the average ratio of technology purchase expenditures to sales is computed for three digit industry groups. These ratios are multiplied by each ®rm sales in 1975±76 to get a rough estimate of payment for technology per year for any year during 1967±68 to 1974±75, the years in which ®rm is known to have purchased technology as determined above. These payment streams are then discounted using above formula to generate the stock for the initial year.
Copyright # 2000 John Wiley & Sons, Ltd. J. Int. Dev. 12, 343±369 (2000)

Technology Transfer and Productivity Spillovers

369

R&D (or own technical capital) stock Information about past R&D investments of a ®rm can be used to approximate the technical capital or knowledge generated by it. Perpetual inventory method is used to construct the own technology stock of the ®rm: RDiYt ˆ …1 À d†RDiYtÀ1 ‡ RDPiYtÀ1 where RDPi,tÀ1 is the expenditure on research and development at time t À 1 and d is the rate of depreciation to technical knowledge. Following earlier studies, present study also uses a depreciation rate of 15 per cent. These R&D expenditures have been de¯ated using a weighted average of the wage and capital investment price de¯ators. As most ®rms did not do R&D prior to 1975±76, the ®rst year of our data, one can assume that the ®rms which have not reported any R&D in the ®rst year of our data have not engaged in any R&D activities in preceding periods also. Thus, the initial period R&D stock for these ®rms would be zero. But to calculate the initial year R&D stock for the ®rms which have reported R&D in the ®rst year of our data we need to know Ð the number of years since when the ®rms have been doing R&D (i.e. the age of their R&D unit), the rate of growth of R&D expenditures in such units and the rate of depreciation of R&D stock. The calculations indicate that real R&D expenditures per R&D unit (as recognized by Department of Science and Technology) in the pre1975 period has grown at about 5 per cent a year in India and the estimated average age of the R&D units recognized by DST in 1975, the initial year of out data has been 4.9 years. Based on the information one can assume that the ®rms reporting R&D expenditures in the initial year of our data has been doing R&D for 5 years prior to that year. Thus, the 1975 stock of R&D would be RDiY1975 ˆ RDP0 À …1 ‡ d†RDPÀ1 ‡ …1 À d†RDPÀ2 ‡ …1 À d†RDPÀ3 ‡ …1 À d†RDPÀ4 For a depreciation rate of 15 per cent and the assumed R&D growth of 5 per cent, the ®nal equation to estimate the initial year of R&D stock would be 4 5 4 ˆ s ……1 À 0.15†a…1 ‡ 0.5†† RDiY1975 ˆ RDP0 sˆ0 Copyright # 2000 John Wiley & Sons, Ltd.

J. Int. Dev. 12, 343±369 (2000)