FDI AND TECHNOLOGICAL SPILLOVERS IN THE SADC REGION. J Paul Dunne and Nicholas Masiyandima. School of Economics University of Cape Town

Transcription

1 FDI AND TECHNOLOGICAL SPILLOVERS IN THE SADC REGION J Paul Dunne and Nicholas Masiyandima School of Economics University of Cape Town Abstract: This paper contributes to the important current debate over whether foreign direct investment (FDI) confers benefits to host countries in terms of improved productivity and growth. It investigates the impact of FDI on technological spillovers to domestic firms for a group of African countries in the SADC region, using the World Bank Enterprise Surveys firm data. The results of the study strongly suggest the existence of positive FDI productivity spillover effects within firms, using both labour and total productivity measures. There is also evidence for positive intra-industry spillover effects, but with some individual country heterogeneity. Seven countries exhibit positive intra-industry spillovers while two countries experience negative effects. In overall terms, it appears that in most countries FDI plays an important role in facilitating technical progress. Keywords: Growth; development; technology; spillovers; productivity; FDI; SADC JEL classification: Preliminary draft. Comments welcome; Please do not quote without permission of the authors. 1

2 Introduction Research has suggested that much of the observed differences in countries per capita incomes emanate from cross country differences in firm productivity (Banerjee and Duflo, 25; Restuccia and Rogerson, 28; Hsieh and Klenow, 29; and Bartelsman, et al, 213). While endogenous growth framework identifies the technological externalities associated with R&D, innovation, human capital, foreign direct investment (FDI) and trade as constituting the major drivers for technological growth, the international transfer and diffusion of technology has played a critical role in assisting the developing economies to access more advanced technologies. This channel is crucial for development in underresourced developing countries given their relatively low levels of R&D and human capital development. For the majority of developing countries, foreign sources of technology account for the biggest proportion (9% or more) of domestic productivity growth (Keller, 29). Foreign direct investment is also beneficial in bridging the savings gaps and current account imbalances of recipient countries. The perceived importance of FDI has seen most countries offering tax holidays, tariff reductions or exemptions, and subsidies incentives to attract and retain foreign investments. Henceforth, the global stock of inward FDI increased by more than ten-fold between 199 and 212, from US$278 billion to US$ billion dollars (UNCTAD, 213). Developing countries are benefiting from an increasing share of the FDI flows, receiving 52 per cent of global FDI inflows in 212 against 42 per cent for the developed countries (UNCTAD, 213). This most probably reflects the extent to which the developing world is becoming a more attractive destination for FDI in terms of policies, improved market conditions and improvements in social and absorptive capital. The global benefits from FDI, however, suggest a contrasting and discouraging trend, with a marked decline in the growth rate of the contribution of FDI to value added, from a pre-global financial crisis average of 35% between 25 and 27 to a mere 5.5% in 212. Similarly, FDI employment contribution growth declined from 141% to 5.7% while exports contribution growth declined from 234% to.6% over the same period (UNCTAD, 213). A belief in the positive effects of FDI is not universal. Expectations of the impact of FDI vary across schools of thought. The dependency, neo-marxist, school sees FDI as benefiting the FDI source countries and the modern economy at the expense of the host and periphery (Wilhelms, 1998). The more recent market stealing hypothesis (Aitken and Harrison, 1999) perceives FDI as a potential source of crowding out of local firms in domestic markets, with consequent costs borne by domestics firms. In addition, doubt on the benefits of FDI is echoed by Carkovic and Levine (25) who are sceptical of the research methods that have identified positive links between FDI and growth, while Rodrik (1999) remarks that the evidence in support of the FDI productivity spillover effects is sobering and that the claims for the positive spillover effects of FDI are extravagant. These differing perspectives suggest the importance of further careful empirical research on the debate. The questions that come to mind are whether the evident benefits and or costs of 2

3 FDI at the macro level can be generalized to the micro level; whether the gains and losses from FDI are uniform across countries; and exactly what the nature and structure of the benefits or costs of FDI are to the firms with or interfacing with foreign equity holdings in domestic firms. This study considers these questions. It investigates the role that FDI plays in domestic technology and productivity spillovers in the SADC region. It investigates the FDI spillover effects of foreign equity holdings both within the recipient firms and across other local firms for the pooled firm cross section for the whole region; for firm cross sections in individual countries and for the small and large firms separately. As most of the countries in the region are at stages of transforming from factor-accumulation-based growth to growth based on efficiency gains (Fedderke and Romm, 26), we hypothesise that increased growth in technology in the region through potential spillovers from foreign direct investment is important for sustained improved economic performance in the region. The results of the study are optimistic on the spillover hypothesis. They point to the existence of positive spillover effects from FDI to productivity improvement among countries in the SADC region. The findings are robust to two measures of firm productivity and across different firm sizes. The picture with respect to the impact of FDI on firm productivity at country level, however, suggests a non-uniform impact of the spillover effects. These results are uniquely informative. The facts and arguments emanating from the study are peculiar not only from a developing country perspective but more importantly for Africa where studies on international technology diffusion and transfer are still scanty. The next section of the paper considers FDI technological spillover channels and linkages, with section 3 presenting the theoretical framework of the study. Section 4 considers the estimation methods, presents the results of the study and discusses their implications. Finally, Section 5 provides some conclusions. 2. The FDI Technological Spillover Link Several channels have been advanced as possible ways through which FDI transmit technology between source and host countries. The channels are either horizontal and or vertical. Horizontal spillovers occur at intra-industry level. They include reverse engineering, imitation, labour turnovers and demonstrations. Vertical spillovers occur through backward and forward linkages between local firms and their foreign affiliates. The relationship is often of the supplier-buyer type of relationship in intermediate inputs and final goods. Findly (1978) and Liu (28) model spillovers realized through some form of firm-specific capital that firms accumulate building on the public information about new and better production methods availed by FDI. In their view, FDI increases the marginal profitability of investing in firm specific capital by local firms. In the process there is greater potential created for productivity improvement in domestic plants when firms allocate more managerial time to the production of this form of capital. Similarly, Walz (1997) suggests that the presence of MNCs results in knowledge spillovers to the domestic R&D sector and contribute to enhanced productivity. 3

4 For Wang (199), foreign direct investment plays the role of accelerating investment in domestic human capital. Superior production methods introduced through FDI force domestic labour to be retrained to match with the new technologies. Fosfuri et al. (22) present models of technology spillovers through labor turnover. Glass and Saggi (1998) suggest that the presence of foreign firms in the same sector helps to lower the cost of imitation for the domestic firms. Alternatively, MNCs force local firms to use their resources more efficiently or to search for new technologies (Blomstrom and Kokko, 1998). Jarvocik (24) suggests that the MNCs set specific quality requirements on their local firm affiliates and in the process force them to invest in better technologies. The buyer-customer relationships between local firms and mother MNCs also often comes with training, technical assistance and managerial capacity building which benefit the domestic firms. The international transfers and diffusion of technology is not without costs. The MNCs have an incentive to block the flow of capital knowledge to local competitors through protection of their intellectual property, production and trade secrecy and paying higher wages to prevent labor turnover or investing in sectors where the domestic firms have limited imitative advantages (Jarvocik, 24). In models in which FDI is a source of firm-specific capital, the accumulation of the firm specific capital requires allocation of managerial time and effort towards investments in firm specific capital by local firms and in the process firms lose potential output. Aitken and Harrison (1999) put forward a market stealing hypothesis associated with FDI crowding out the local firms in markets and increase their costs. Liu (28), however, warned that the analysis of the impact of FDI on productivity should separate the effect at impact and the effect in the long run. He argues that FDI has negative level effect at impact and positive rate effect in the long-run. Empirical literature on foreign direct investment spillover effects has produced mixed results. In early studies, Caves (1974) finds positive and significant spillovers in the Australian manufacturing sector, while a study by Germidis (1977) for a sample of 65 multinational subsidiaries in 12 developing countries finds no evidence of technology transfer from foreign to local firms. Rhee and Belot (1989) in a study on Mauritius and Bangladesh find that the entry of foreign firms in the textile industry is the driver for the growth of domestically owned textile firms. Studies by Haskel et al. (22) and Keller and Yeaple (23) confirm spillover effects on firm productivity for the United Kingdom and the United States, respectively. Using cross-sectional enterprise-level data for Hungary, Schoors and Tol (21), get evidence in support of a positive FDI spillover effect through backward linkages for the Hungarian firms. Helpman, et al (24) in a study of international technology spillovers find that the effects of FDI on productivity are greater than the effects of export. Yasar and Paul (28) using a propensity matched score approach concludes that exports and FDI dominate imports in transmitting improved technology in Turkey. Negative spillover effects from FDI on firm productivity have been identified for Morocco (Haddad and Harrison, 1993). Carkovic and Levine (25) in a study of the OECD countries failed to confirm the FDI-growth positive link and argue that most studies on FDI-growth linkage are flawed with methodological issues that include endogeneity and simultaneity 4

5 problems, measurement errors and failure to account for country specific effects. Aitken and Harrison (1999) in a study of Venezuelan firms find negative effect of sector FDI domestic firm productivity. In a study of Chinese manufacturing firms, Liu (28), however, argues and shows that the crowding out effect of FDI is a short term impact concept. He uses firm panels that span for long period and find that foreign direct investment has a positive effect on firm productivity in the long term. Javorcik (24) argues that studies that find negative intra-industry spillover effects of FDI on domestic firm productivity are looking for the spillover effects in the wrong place. He argues that the spillover effects should be in vertical relations between local firms and their MNCs affiliates. He finds the existence of positive backward productivity effects among a group of Lithuanian firms and conclude that local firms benefit more when they are suppliers to the MNCs than when they a buyers from them. In view of the inconclusive debate on the impact of FDI on local productivity and growth, one can argue that the differences in countries conditions, the nature of the data and the research methods used are a potential cause for the observed differences. Countries that are marred with distortionary interventions, institutions and policies are likely to benefit less from the potential technological flows from FDI than freer ones where market driven incentives and institutions drive business. Equally important, is the issue of social and FDI absorptive capital in the form of good financial intermediation, human capital and good infrastructure availability which are critical for countries to harness maximum benefits from FDI (Chen, 27). 3. Research Methods 3.1 The Technological Spillover Framework This study uses the endogenous growth model framework of firm-specific human capital accumulation (Ehrlich and Lutter, 1989; Ehrlich, et al, 1994 and Liu, 28). The model envisages firm-specific capital as one of the firm s production inputs. The firm specific capital endogenously depends on optimum decisions by the firm to allocate managerial time between its production and firm output production. Thus the firm s production function is specified as follows: Yijt At Bt Lt Kt [ HtMt ] (1) Where Yijt is output for firm i in industry j. At represents exogenous technology. The exogenous technology is common to all firms; in FDI. Lt and Ktare labour and capital; Bt is the state of technology that is embedded Ht is the stock of firm specific capital. Ht is unique to the firm and it depends on the effort, resources and time that the firm devotes to R&D, imitation and learning from observing techniques employed by the MNCs. It is positively related to the amount of technical information the firm has. M t is the amount of time the firm devotes to current output production. The firm is assumed to have one unit of production 5

6 time. 1 Mt is, therefore, the amount of time left for the production of firm-specific capital. The production of firm specific capital is specified as: rht[ 1 Mt Gt (2) H ] The production of H t positively depends on three factors - its current stock; the amount of time the firm devotes to its new accumulation ( 1 M t ) and the technical and managerial information the firm has ( G ). The information input in the production function ( G ) is either internal to the firm or is external and domiciled in the public domain. When MNCs bring some FDI and are producing alongside the domestic firms, they release information on new and advanced techniques into the public domain. This is done through demonstrations, worker turn over or simply through observing and imitating. Hence G increases with the level of FDI in a country. r is an efficiency parameter of the firm specific capital production function. The parameter < <1 indicates whether there are diminishing, constant or increasing returns in the production of firm specific capital. represents the intensity of spillovers. If 1 there are increasing returns from FDI spillovers in the production of the firm specific capital. If, there are no spillovers. The firm s optimization problem involves choosing the amount of labour, capital and time devoted to producing firm specific capital that maximize its value or its expected value. The Hamiltonian for the firm s value optimization problem can be stated as: V ] t pat Bt Lt Kt [ HtMt ] w' Lt c' Kt rht[1 Mt Gt (3) Where w' and c ' are the prices of labour and capital, respectively and p is the price of the firm s product. We assume the firm is producing and hiring factors under competitive environments. The log of total factor productivity from the production function is: tfp A B M H (4) And the steady state growth of firm specific capital at optimum time allocation is given by: ˆ * r[1 (5) H M t ] G t Equations 4 and 5 imply that the choice of M has two opposing effects on TFP. A reduction in the amount of time allocated to production M has a negative scale effect on current firm productivity. At the same time it increases the productivity of the firm specific capital and hence increases TFP. The resultant net effect on firm productivity can, therefore, be positive or negative. 6

7 Under conditions of perfect competition, the firm s optimization solution is where the allocation of production time is at a point where the marginal profitability of time allocated to producing firm specific capital is balanced with the marginal profitability of time allocated to output production. When MNCs bring some FDI, this increases the marginal profitability of time allocated to production of firm specific capital and thus create scope for higher firm productivity as firms invest more in the production of firm specific capital. 3.2 Empirical Model, Data and Definition of Variables Empirical Model Our estimated model emanating from the envisaged FDI productivity spillover framework borrows from Aitken and Harrison (1999) and Liu (213). It is defined by equation (6) as follows: fp ij 1FDI firmij 2FDI sec j 3FDI firm*sec ij 4Xij 5I 6T 7R 8C it (6) Where fpij is a proxy for the FDI spillover effects on firm i in sector j. The study utilizes the logarithms of labour productivity and firm total factor productivity for fp ij. FDI firmij is firm FDI measured as foreign equity ownership in the firm. FDI sec j is sector FDI measured as a weighted total of firm level FDI for firms in sector j. FDI firm secij * is an interaction term of firm and sector FDI meant to capture the effect of sector FDI on firms that have foreign ownership. The matrix X ij captures other determinants of firm productivity identified in the literature. The determinants that are identified from our data include firm size, age, infrastructure obstacles, corruption, credit constraints and availability, human capital, industry regulations, access to land, legal institutions and macroeconomic and political (in)stability measures. T, I, R and C are the year, industry, region and country dummies meant to isolate the impact of FDI on firm productivity. The time dummy takes account of the differences in the years the country surveys were done. All measures of FDI are computed at firm or sector levels in each country separately. Weights used in computing them are based on country quantities. it is a random shock to firm productivity assumed to be exogenous and independent of other productivity covariates. From the spillover hypothesis 1, 2 and 3 are expected to be positive. Our preferred measure of the spillover effects is the labour productivity moment. The total factor (TFP) productivity measure could have been a better measure. In our case, however, the TFP suffers from the well-known simultaneity bias in its estimation given the cross section data at our disposal. We, however, use industry dummies in estimating the TFP from firm production functions to minimize the bias. The correlation between LP and TFP has also been found to be high and positive in the literature on firm productivity. In our case the 7

8 correlation coefficient between the two is.68 and is statistically significant. We, therefore, use the TFP measure for robustness checking. We also do a descriptive analysis of the FDI productivity spillover effects using the firm data. Criticism on the use of the LPR emanates from potential differences in factor substitutions by firms that may affect labour productivity without necessarily having any implications on firm technological growth. We take care of this by controlling for the firm s capital-labour ratio in the estimated the spillover models that use the LP for fp ij. In modelling the FDI spillover effects, literature has cautioned against the possible simultaneity in causality between foreign firm ownership on one hand and firm and sector productivity on the other. FDI naturally flows into the higher productivity plants, sectors and regions. Observed positive link between productivity and FDI measures may be a result of FDI self-selecting itself into the higher productivity regions, sectors and firms. This makes it difficult to pin down the effect of FDI on productivity and results in upward biases on the FDI coefficients. Firm, industry and region fixed effects dummies are often employed to control for firm, sector and region specific productivity effects for the firm. In our case, we use sector and region dummies. The fixed effects specification of the spillover model in (6) is meant to partial out the impact of FDI on firm productivity. The use of industry and region dummies assumes that high productivity firms locate in high productivity industries and regions and that foreign investors are positively driven towards the higher productivity firms. In isolating regional effects on FDI flows, we use the average regional wage rate to proxy for regional fixed effects (see Aitken and Harrison, 1999; Rauch, 1991; and Dunne, et al, 24) The Data The study uses firm level cross section data pooled for twelve countries in the SADC region. The source for survey data are the country World Bank Enterprise Surveys (WBES). The WBES are available for Angola, Botswana, DRC, Lesotho, Malawi, Madagascar, Mauritius, Mozambique, Namibia, South Africa, Swaziland, Tanzania, Zambia and Zimbabwe. The surveys were done between 26 and 21 using the global survey approach. This enables us to pool the data across the countries. The survey for Malawi was done in 25 before the global survey approach and was not easily comparable with the rest of the countries. The survey for Lesotho misses critical variables-worker education and firm capital stocks. The two countries were, henceforth, excluded leaving twelve countries for analysis. The remaining twelve countries constitute a total of 2933 manufacturing firms. South Africa has the largest sample representation of 24% followed by Zimbabwe (13%), Mozambique (12%) and Zambia (11%). The least representations are in Swaziland and Botswana (2%), DRC and Namibia (4%), Madagascar (6%) and Mauritius (8%). By the design of the World Bank ES, these share representations are reflective of countries sizes, value addition and number of firms in the respective countries. The surveys data covers all the major two-digit manufacturing industries classified according to the International Standard Industrial Classification (ISIC), revision 3.1. Industries were assigned into broader industrial categories on the basis of similarities in the types of their 8

9 activities in order to form large industry sizes on which authentic estimations and inferences can be done. This yielded six industries tabled below: Table 1: Firm Categories by ISIC Codes Category ISIC Industry. of Firms 1 28, 29, 3 Industrial equipment and Fabricated Metals , 26 Basic Metals and non-metals , 25, 19 Chemicals, Plastics and Rubber , 16 Food , 18 Textiles and Garments , 21, 22, 36, 37, 31, Other 32, 33, Source: Constructed from World Bank Enterprise Surveys A preliminary detailed analysis of the variables distributions across countries, industries and size categories were done to establish any heterogeneities in the data and major variables across countries, industries and firm size categories. The analysis reveals that productivity heterogeneities across countries are mainly with respect to measures of central locations for which country fixed effects model will control for. The DRC and Angola are also outliers in terms of productivity variability Construction of Foreign Firm Ownership Measures Firm foreign direct investment ( FDI ) is defined as foreign equity participation in firmij domestic manufacturing firms. The variable ranges from % for no foreign ownership to 1% for firms that are wholly foreign owned. The use of cumulative levels of foreign firm ownership amounts to using FDI stocks rather than FDI flows. The computation of industry foreign direct investment ( FDI sec ) follows the methods used in Aitken and Harrison (1999) ij and Liu (28). It is measured as a weighted total of foreign firm equity ownership in the same industry at the two digit level. The intra-industry FDI foreign equity holding levels ( FDI sec ij ) capture the impact of the FDI spillover effects on firm productivity. It is measured as: employ i FDI sec ij XFDI firmij (7) i Employ j employ i gives weights on individual firms foreign ownership. It is measured as firm size Enploy j relative to total industry size. Size is defined in terms of levels of employment. Foreign firms tend to be more capital intensive than domestic firms. The use of capital base as a measure of firm size would upwardly bias the weight applied to the foreign firms against domestic firms. This justifies the use of employment levels. The measurement of regional foreign firm equity holding levels follows a similar method, except that the weights will be computed using regional employment levels while firms will be classified by region. 9

10 3.2.4 Estimating Measures of Firm Productivity (a) Labour Productivity Labour productivity is computed as real firm output (sales) divided by the number of workers employed adjusted for the human capital element. The study utilizes the availability of data on worker education in the World Bank Enterprise Surveys to compute the human capital component of labour. The measure of human capital ( HK ) used follows Caselli (25). It is measured as: i HK L e i i Q ( S i ) (8) where is the number of full-time employees for firm and is the average human capital per employee. is average years of schooling of the firm s employees. Following Caselli (25), is assumed to be piecewise linear with varying slopes depending on average worker education of the firm. The computed LP measure, however, fails to adjust for effective hours worked. Using the payroll labour would have been more reflective of effective hours worked as an alternate to number of workers employed. This information is not available in the WBES used. Foster, et al (28) and Bartelsman (213) have, however, argued that the correlation between measures that control for effective time worked and using the number of workers is high and positive. This justifies our empirical measure as an alternative to payroll labour. (b) The Total Factor Productivity Total factor productivity is computed from firms real value added output. A Cobb-Douglas production function with capital and labour is used to estimate TFP. The production function is estimated for each country separately. The estimated production functions control for firm specific effects on productivity by using industry dummies in the estimated equations. The estimated production function is thus specified as: Y ij A L K (9) i l i ij k i ij i ij Where Yij measures firm output. This is measured as sales in the WBES. Kij is physical capital employed by the firm; Lij is labour adjusted for human capital. It is the same as HKi in equation (8). i represents a vector of firm specific effects on firm output. Our model assumes that the firm specific effects are reflected in the industry specific effects ie. i j l i and k i are labour and capital elasticities, respectively. The World Bank Enterprise Surveys provide information on Y, ij, Lij Kij and j Sales are given in local currency units. Real output is obtained by deflating the local currency units measured sales to 25 using each country s GDP deflator then converted at 25 exchange rates to the United States dollar equivalence. 1

11 Capital is measured as firms reported net book values of fixed assets. The assets include equipment and machinery and motor vehicles. The net book value is given in the surveys in local currency units. They are converted to real values by deflating the local currency units measured values of the assets to 25 using each country s GDP deflator before they are converted to the US dollar equivalence at 25 exchanges. Deflating local currency unit measured variables using each country s GDP deflator and conversion to the US dollar at exchange rates prevailing in 25 tries to use a price index which is more direct and specific to the relevant variables. The USA deflator used in other studies would be more distanced from the measured variables of interest. Still, the use of a broader price index such as the GDP deflator has, however, been proved to fall short of completely pegging off firms idiosyncratic demand effects on the nominal variables (Foster, et al, 28). Once the production function is estimated, is residuals are used to measure firm productivity. Figure 1 superimposes the distributions of the individual countries firm productivity. The two measures have a positive and statistically significant correlation coefficient of.68. The distribution plots suggest that within country variability of the two moments of productivity is more or less similar across countries, except for the DRC and Angola identified as outliers. Evident differences are in respect of central locations. The differences in the average productivity measures are attributable to differences in country specific factors that affect firm productivity. These are controlled for by the country fixed effects in the estimated models. Differences in central locations of distributions could also have been partly contributed by the effect of time differences on productivity, given that the surveys for the 12 countries were done in different years. We control for this by using survey year dummies. Figure 1: Countries Labour and Total Factor Productivity Distributions LPR TFP x x Source of Data: World Bank Enterprise Surveys In all our pooled data scenario estimations of (6) we exclude Angola and the DRC given their outlying productivity distributions polarizations in comparison to the other countries. 11

12 Relative Density Relative Density 4. Analysis of Results In our descriptive analysis of the relationship between firm ownership and firm productivity, firms are classified into those that are foreign owned and those that are domestically owned. Firms are defined to be foreign owned if they have at least 1% foreign shareholding; otherwise they are defined as domestically owned. The cut-off point follows the definition of FDI by the IMF (IMF, 1993). The definition also matches most SADC countries investment Acts and investment codes. The study uses the relative distribution analysis for the comparisons. The approach is a comparison tool for two groups of data on a common variable. In our case this is firm productivity by firm ownership. Relative frequency distributions have the advantage of being scale invariant to monotonic transformations of the original measurement scale without need for parametric distribution assumptions (Morris and Handcock, 1999). For example, using output per worker versus log of output per worker in our analysis yields the same analytical conclusions. Figure 2 plots the relative distributions of the LP and TFP in the two defined firm ownership categories. The left hand side graphs give the LP and TFP densities for the two ownership categories. The right hand side plots are their respective relative distributions. For both the LP and the TFP, the broken densities are for the domestic owned firms while solid densities are for the foreign owned firms. Figure 2: Firm Foreign Equity Ownership and Firm Productivity Ref: Foreignown>=1 frgn_llp 2 frgn_llp_rd x Proportion of Reference Group frgn_tfpr 2.5 frgn_tfpr_rd x Proportion of Reference Group Data Source: World Bank Enterprise Surveys The plots suggest that foreign owned firms are more productive than domestic owned firms. Using the LP measure, the first row shows that domestic firms productivity is more likely (greater than 1) to fall within the 5 th quantile of foreign owned firms productivity 12

13 distribution and less likely (less than 1) to fall beyond the 5 th quantile. The TFP has similar indications, with domestic firms productivity more likely to fall within the 4 th quantile of foreign firms productivity distributions than above it. A similar analysis is also done for the two firm ownership categories for small firms and large firms categories separately. This follows arguments that the adoption and utilization of foreign technology from foreign ownerships and its impact on firm productivity has some costs to local firms. The large better resourced firms are likely to benefit more from foreign firm equity holdings. Another view is that countries, sectors and firms that are further away from the technology frontier of the multinational corporations yield greater benefits from the spillover effects than those closer to the frontier (Aghion and Howitt, 24). If this holds, greater spillover impact on the small firms is expected, assuming that small firms are on average further away from the technology frontier than large firms. Figure 3 in the appendix suggests a clear productivity gain for the foreign owned small firms. The relationship between foreign firm ownership and productivity for large firms is unclear and inconclusive. We, however, note that these descriptive analysis do not imply any direction of causality but simply suggest that there is a positive correlation between foreign firm ownership and firm productivity. Comparisons of productivity by firm ownership for the pooled cross sections, for example, simply suggests that either foreign investors are attracted to more productive firms or that higher foreign ownership drives firm productivity. Establishing direction of causality requires controlled estimation of the spillover model. Our estimation of the FDI technology spillover model in equation 6 using the pooled firm cross section data gives results in Table 2. The first column reports the productivity spillover effects assuming that the firm FDI and the sector FDI are the only productivity spillover channels. The second and third columns introduce the firm-sector FDI interaction term ( FDI firm* secij ) and foreign inputs (FGN_inp), respectively. The inclusion of foreign inputs follows literature on international technology transfers which argues that the use of foreign inputs by domestic firms is a potential channel of transmitting foreign technology into local production processes (Yasar and Paul, 28). The argument is that foreign inputs and intermediate goods are embedded with higher levels of technology that benefit domestic firms which use them. Failure to control for their impact may result in biased estimated impact of FDI on productivity, especially given the more likelihood of the use of foreign inputs by foreign owned firms. The alternate use of the time and country dummies in columns two and three has shown that only one of them can be used in a single model specification. This is because in some instances, the year of the survey also identifies the country. We, however, prefer the use of the country fixed effects in light of the useful information it conveys about differences in spillover across countries. Column four introduces the regional FDI (FDI_reg) and wage rate (Reg_wge). A possible source of misspecification of the spillover model in (6) emerges if the MNCs generate positive technology spillovers to plants that are located in their geographic proximity. The 13

14 inclusion of the wage rate is meant to identify this spatial effect. Column five is the full model. It controls for all the possible channels of FDI spillover effects plus foreign inputs and regional FDI. The model also has dummies to control for industry, country and regional productivity effects. For interpretation purposes, we will use the full model in column five. Other columns are likely to suffer from omitted variables biases. TABLE-2 FDI Labour Productivity Spillovers - Pooled Cross Sections Impact of FDI on Firm Labour Productivity LPR Excl. DRC & Angola(1) LPR Excl. DRC & Angola(2) LPR Excl. DRC & Angola(3) LPR Excl. DRC & Angola(4) LPR Excl. DRC & Angola(5) FDI_firm.73*** 1.34*** 1.23*** 1.2***.87*** (.1) (.21) (.21) (.21) (.19) FDI_sec 1.15*** 1.67*** 1.51*** 1.4***.59* (.34) (.38) (.37) (.36) (.33) FDI_secfirm -.2*** -.2*** -.2*** -.1* (.1) (.1) (.1) (.1) FGN_inp.53***.47***.54*** (.1) (.1) (.1) FDI_reg (.24) (.23) Reg_wge 1.42*** 1.31*** (.17) (.17) Constant 9.35*** 9.28*** 9.14*** (.28) (.27) (.28) (1.35) (1.48) Observations 1,873 1,873 1,873 1,873 1,873 R-squared Time FE Country FE Industry FE Region FE Robust standard errors in parentheses *** p<.1, ** p<.5, * p<.1 //The model includes other covariates of firm productivity identified from the list discussed above, etc //FDI measures are in decimals/fractions avoid too small coefficients. The coefficient on firm foreign ownership is consistently positive and statistically significant across the five specifications of the model. The coefficient on FDI firmij shows that firms with 1% more foreign ownership have average labour productivity level which is 8.7% higher than their counterparts with lower FDI. The positive impact of FDI on plant level labour productivity represents the direct vertical within plant impact of foreign firm ownership on domestic firm productivity. This emanates from the more advanced technology, managerial skills and training that the foreign owned domestic plant receives from the mother MNC. The impact of sector foreign equity ownership ( FDI firm secij * ) on domestic firms is also consistently positive and significant across the five model specifications. This suggests that domestic plants in sectors with more foreign equity holdings are more productive than those in sectors with less foreign ownership. Results show that firms in sectors with 1% more foreign ownership are 6% more productive than their counterparts in sectors with lower 14

15 sector FDI. After controlling for the sector specific productivity effects, the positive productivity effect is largely attributable to the existence of more sector foreign equity holding. This result is in support of the FDI technological spillover hypothesis. A comparison of results in columns four and five helps to illustrate the importance of controlling for firm, sector and region specific effects in partialling out the exogenous impact of FDI on firm productivity. Without controlling for region and industry effects, the productivity differential between firms with 1% more FDI and those with less would erroneously be overstated as 12% instead of 8.7%. Equally the impact of a 1% difference in sector FDI would be overstated as 14% instead of 6%. The impact of the cross firm-sector FDI interaction term on domestic firms productivity is negative and significant at 1% level. This implies that domestic firms with foreign ownerships are negatively affected by the existence of other foreign owned firms in the same sector. The productivity of local firms with foreign ownership falls by.1% for an increase of 1% in sector FDI. This may be explainable by the fact that other foreign firms in the sector are likely to be equally or even more productive than the local joint ventures. They, therefore, impose stiffer competition than would be subjected by the domestic owned firms. Table 2 also presents the productivity spillover effects from the use of foreign inputs by domestic firms. The coefficient on percentage of foreign inputs used by domestic firms is positive and significant. Its significance is robust across the model specifications. Firms using 1 percentage points more foreign inputs are 5.4% more productive than their counterparts using more of domestic inputs. When the foreign input variable is introduced in column three, it reduces the plant level impact of firm FDI on productivity from 13.4% to 12.2% and the sector FDI from 16.7% to 15.1% for a 1% difference in foreign equity levels at plant and sector levels, respectively Spatial Productivity Spillover Effects Estimation of model (6) without the geographic effects of FDI on firm productivity fails to identify locational spillover advantages accruing to the domestic firms which are located near the multinational companies. Logically any spillover benefits from multinationals should be received first by the local firms closer to them before distant firms access them. There are lesser constraints to labour mobility within similar geographic areas than across geographic areas. If labour mobility is a major vehicle underlying technology transfers and diffusion in the FDI technology spillover hypothesis, then the spatial arguments for inclusion of the geographic factors into the spillovers model make reasonable sense. Column five of the results table 1 shows that firms in regions receiving more foreign firm ownership have lower productivity than those in regions with fewer MNCs. The negative effect is, however, statistically insignificant. In light of the fact that the model estimated in column 5 has controlled for the location-specific effects on firm productivity by way of including the regional average salary, it can be inferred that the coefficient on FDI_reg represents the effect FDI on productivity at the local level. Aitken and Harrison (1999) find an insignificant regional effect for FDI on local firm productivity before and after controlling for the location specific effects. The negative effect in our case suggests the market stealing effect in spatial 15

16 terms. This means that domestic firms are disadvantaged by the existence of foreign owned individuals. The spillover results from using the labour productivity moment are by and large confirmed and supported by the total productivity measure as shown in tabl3 3. TABLE-3 FDI TF Productivity Spillovers on the Pooled Firms Cross Section Impact of FDI on Firm TFP TFP Excl. DRC & Angola(1) TFP Excl. DRC & Angola(2) TFP Excl. DRC & Angola(3) TFP Excl. DRC & Angola(4) FDI_firm.76***.6***.5**.46** (.19) (.18) (.22) (.2) FDI_sec 1.5*** *** 1.2*** (.34) (.32) (.34) (.33) FDI_secfirm -.1* (.6) (.5) (.7) (.6) Fgn_inp.37***.38*** (.9) (.8) (.1) (.9) FDI_reg * (.22) (.24) Reg_wge.81***.48*** (.17) (.8) Constant 8.32*** *** 2.67*** (.14) (1.48) (.19) (.36) Time FE Country FE Industry FE Region FE Robust standard errors in parentheses; *** p<.1, ** p<.5, * p< Are the Spillover Effects Transient in Small and Large Firms? Table 4 reports the FDI technological spillover estimation results for the small and large firms. Columns 1 and 2 are for the small firms and 3 and 4 are for the large firms. Columns 1 and 3 have results for the full spillover model with country, industry and region fixed effects. Columns 2 and 4 leave out the interaction term between firm FDI and sector FDI, given its poor performance. The results show that there is positive and significant technology spillover impact from firm foreign ownership ( FDI ) across the two size categories in both model firmij specifications. The impact is, however, stronger for the small firms in both model specifications. Small firms with 1% more FDI, for example, have an improvement in productivity of 9% while large firms productivity improves by 6.4% for the same difference in foreign ownership. The impact of sector FDI on productivity is also positive across the two models and for the two firm size categories. The intra-industry impact is, however, significant only for the small firms and insignificant for large firms. This suggests that the small firms also seem to benefit more from sector FDI than large firms. 16

17 TABLE-4: FDI Spillover Effects on Small and Large Firm s LPR Productivity Impact of FDI on Firm LPR by Firm Size LPR Small Firms (1) LPR Small Firms (2) LPR Large Firms (3) LPR Large Firms (4) FDI_firm.9***.55***.64*.52*** (.23) (.11) (.33) (.15) FDI_sec.5* (.27) (.33) (.73) (.62) FDI_secfirm -.1* -.4 (.1) (.1) Fgn_inp.61***.62***.3.3 (.1) (.1) (.23) (.22) FDI_reg * -1.4* (.25) (.25) (.58) (.57) Constant (1.62) (1.62) (4.4) (4.38).. Observations 1,515 1, R-squared Time FE Country FE Industry FE Region FE Robust standard errors in parentheses; *** p<.1, ** p<.5, * p<.1 //All models exclude DRC and Angola The impact of FDI spatial spillovers and spillover effects from use of foreign inputs are also stronger for small firms than large firms. Labour productivity improves by about 6% for the small firms which use 1% more of foreign inputs compared to about 3% gain in productivity for large firms. In the case of spatial spillover effects, the productivity gain for small firms is positive though insignificant. For large firms the spatial productivity effect is negative and significant at 1%. A similar pattern of results using the total factor productivity moment is displayed in table 6 in the appendix. These results suggest that small firms have larger exploitable economies from new technologies that come with FDI than larger firms. This is in line with Aghion (23) s views on which firms gain more from new innovation in relation to their location in the technology frontier. Most large firms in the region are MNCs which already have relatively more advanced technology than most small firms which are locally owned. Individual country results on the impact of foreign direct investment of technological spillovers to the local firms are shown in tables 5 and 6 in the appendix. Using the labour revenue productivity measure, the within plant productivity spillover effects from firm FDI are positive and significant in Botswana, Madagascar, Zimbabwe, Tanzania, Zambia, Namibia, Swaziland and the DRC. Mauritius and Angola exhibit negative within plant effects. The rest of the countries have positive but insignificant vertical spillover effects. 17

18 Intra-industry productivity spillover effects from foreign equity holdings are positive and significant for Zimbabwe, Tanzania, Namibia and Swaziland. The effects are negative and significant for Mauritius. For South Africa, Mozambique, Botswana, Madagascar, Zambia and the DRC, the intra-industry spillover effects are negative though insignificant. The negative intra-industry spillover effects are reversed in sign but still remain insignificant for South Africa, Mozambique and Botswana using the TFP. Table 6 presents country results for the TFP moment. 5. Conclusions This paper has provided an empirical analysis of the productivity spillover effects of FDI on domestic firms in the SADC region. A number of interesting results emerge. The existence of within firm vertical productivity spillover effects is confirmed for the pooled firm cross sections; for both large and small firm categories; and for all the countries in the region except for Mauritius and Angola. This indicates that firms in the SADC region stand to benefit through within plant improvement in productivity through foreign shareholding in the firm. This finding is robust with respect to both labour and total factor productivity measures of firm productivity. The impact of intra-industry productivity spillover effects is mixed. Using the pooled firm data for all the countries under study (except for DRC and Angola), there is positive intra-industry spillover effects and the effect is robust to our two measures of firm productivity and across firm sizes. There is, however, considerable heterogeneity across countries, with negative and significant effects for Mauritius and negative though insignificant effects in South Africa, Mozambique, Botswana, Madagascar, DRC and Zambia. Thus while some countries have benefits that come with intra-industry FDI, firms in other countries tend to be adversely affected, possibly through the market stealing effects of foreign owned firms. The results also suggest that small firms in the region tend to benefit more from FDI technology spillovers than large firms on average. The differential gain in productivity for the small firms compared to large firms ranges between 2.6% to 12% for a 1% difference in firm, sector and regional FDI. This implies that small firms in the region have larger exploitable economies from using foreign technology than the larger firms. To the extent that most firms in the region are small, this suggests to the potential that FDI in the region has in boosting productivity and growth in most of the countries. Thus it can be argued that international transfer and diffusion of technology is an important and consistently robust channel through which productivity spillovers are transmitted to domestic firms in the region. 18

19 APPENDIX TABLE-5 FDI Spillover Effects at Country Level- Using the LPR LPR LPR LPR LPR LPR LPR LPR LPR LPR LPR LPR LPR RSA Mauritius Mozambiq Botswana Madagascar Zimbabwe Tanzania Zambia Namibia Swaziland Angola DRC FDI_firm *.59** 1.72***.86***.58*** 1.39* 1.24*** ** (.15) (.69) (.37) (.44) (.26) (.37) (.24) (.2) (.79) (.41) (1.28) (.97) FDI_sec ** *** 5.77** ** 5.14* (2.61) (4.44) (2.9) (2.52) (1.72) (2.45) (2.41) (2.63) (4.16) (2.78) (7.37) (4.2) FDI_reg *** *** *** * ** 3.64* (.26) (1.62) (1.7) (1.77) (.98) (1.26) (.57) (1.17) (1.38) (.89) (11.57) (1.89) Fgn_inp.61***.38.43*.6.57** ***.63*** ** 1.6 (.17) (.33) (.24) (.72) (.27) (.22) (.24) (.19) (.23) (.45) (1.23) Constant 11.66*** 13.8*** 1.2*** 8.25*** 5.3*** 7.77*** 7.38*** 9.25*** 6.92*** 8.3*** 15.77*** 11.86*** (.73) (1.41) (.85) (2.31) (1.28) (.79) (.58) (1.14) (.8) (.62) (4.75) (1.87).. Obs R-squared Industry FE Region FE Robust standard errors in parentheses *** p<.1, ** p<.5, * p<.1 //All regressions are done with same variables as in those in table 1, except in some instances were variable(s) were dropped either because there is not enough variation in the variable at country level, eg firm size; or the variable had too many missing observations. A case of the later was corruption, which is inadequately reported in some countries TABLE-6 FDI Spillover Effects at Country Level- Using the TFPR TFPR TFPR TFPR TFPR TFPR TFPR TFPR TFPR TFPR TFPR TFPR TFPR RSA Mauritius Mozambiq Botswana Madagascar Zimbabwe Tanzania Zambia Namibia Swaziland Angola DRC FDI_firm.26* ***.78***.5*** 1.42**.77** 5.4**.99 (.15) (.8) (.37) (.51) (.3) (.32) (.26) (.18) (.7) (.38) (2.27) (1.45) FDI_sec *** * (3.18) (6.3) (2.98) (5.83) (1.9) (2.6) (2.7) (2.15) (3.) (2.92) (12.7) (4.87) FDI_reg ** *** * *** 2.13 (.25) (1.76) (1.17) (2.6) (1.22) (1.12) (.6) (1.9) (1.3) (.85) (13.28) (2.13) Fgn_inp.49*** ***.56***.65** -4.21*.96 (.18) (.32) (.26) (.67) (.37) (.21) (.2) (.16) (.31) (2.19) (1.39) Constant 8.21*** 7.7*** 6.65*** *** 4.95*** 4.84*** 5.89*** 2.64** 6.27*** 2.6*** 1.35*** (.87) (1.71) (.91) (3.39) (1.44) (.74) (.68) (.97) (.99) (.5) (5.79) (2.66) Obs R-squared Industry FE Region FE Robust standard errors in parentheses *** p<.1, ** p<.5, * p<.1 //All regressions are done with same variables as in those in table 1, except in some instances were variable(s) were dropped either because there is not enough variation in the variable at country level, eg firm size; or the variable had too many missing observations. A case of the later was corruption, which is inadequately reported in some countries 19