Enhancing productive capabilities through intra-regional trade and cross-border investments in Southern Africa

ABSTRACT

Regional integration is an important factor for enabling knowledge flows between economies and enhancing the capacity of firms within the integrated block to benefit from local knowledge spillovers. This study analyses data on economic interactions between Botswana and its technologically more advanced southern neighbour, South Africa, to examine the extent to which knowledge flows facilitated by geographical proximity translate into fostering technological learning and productivity of manufacturing firms. Industry- and firm-level data on bilateral capital goods trade and investments over the period 1991–2013 are used to assess the technological learning of the manufacturing sector in Botswana. This study also applies the Hunt, J, & Tybout, J (1999. Does Promoting High-Tech Products Spur Development? FEEM Working Paper REG 42. Milan: Fondazione Eni EnricoMattei) technological sophistication framework to analyse the role played by regional trade and investment flows between the economies of South Africa and Botswana in the skills intensification of manufacturing firms. Skills intensity decomposition reveals that Botswana’s manufacturing technical intensity has been positively influenced by the extent of capital goods trade and investment linkages with South African economy.

1. Introduction

The tacit nature of technological knowledge often renders face-to-face interactions a necessity for technology to diffuse because knowledge circulates best locally (Kesidou & Szirmai, 2008). It is through those interactions that tacit knowledge can be translated into explicit, usable new knowledge. Geographic proximity is therefore important for technological learning since it facilitates direct interactions and knowledge diffusion. Due to its large diamond deposits and its geographical proximity to South Africa, Botswana is well poised to benefit from mastering and applying technologies already in use in its southern neighbour to support successful technological catch-up and long term-growth.

The argument that Botswana’s economy can benefit from its interactions with South Africa is rooted in the various theories of international knowledge flows and their implications for economic performance. Whereas neoclassical growth models have assumed technological knowledge to be freely available and transferable across national borders, economic historians such as Gerschenkron (1962) and Abramovitz (1986) have observed that large technological disparities could persist in the absence of mechanisms to promote technological catch up. In line with Friedrich List (1841), they pointed out the need for government support in building capabilities and guiding resources to enable technologically less advanced countries to absorb as well as exploit external and internal sources of knowledge to increase their own productivity. The recognition that innovation and learning capacity has to be built early in the development process has been shown to be crucial for developing countries to be able to catch up with earlier industrialisers (OECD, 2012; Lundvall & Lema, 2014).

Botswana’s government has prioritised the development of the manufacturing sector with a view to curbing the country’s dependence on mineral exports. Imports of embodied technology for the manufacturing sector is therefore part of technological learning process. One of the expected outcomes of this learning is the intensification of technical skills and an increase in manufacturing productivity. The industrial sector in Botswana is however narrower than that of South Africa. It is argued that foreign technologies primarily contribute to productivity growth by increasing the technical intensity of the manufacturing sector while widening the range of its productive activities. This could spur new activities which can increase the diversity of export. To support this view, this study extends the technological sophistication framework used by Hunt & Tybout (1999) by combining it with the capital goods trade model developed by Mutz & Ziesemer (2008). This model is used to explore the extent to which Botswana’s firms have utilised imported embodied technologies to improve their productivity. The resulting shift in skills intensity in various industrial branches of the manufacturing sector is then categorised in terms of source of learning (i.e. skills intensification in existing industries versus shift towards new industrial activities). In Hunt & Tybout (1999), the product technological sophistication model starts from what they designate as the Lucas/Krugman/Stokey/Young (LKSY) view suggesting that growth is accomplished by concentrating resources in industries whose production processes induce learning and knowledge spillovers. That model was developed to relate firm- and plant-level technological sophistication of the manufactured products as well as the productivity growth rate associated with each product to the knowledge flow accruing to the analysed industries. For so doing, Hunt & Tybout (1999) distinguish between two mechanisms through which increasing sophistication can take place, namely the continual shifting of resources towards high-end products as predicted by the LKSY product spectrum models and a general increase in the intensity of skilled input use among all types of products. The latter is referred to as human capital deepening that provides an engine for growth in models that do not distinguish a spectrum of products in terms of their potential to generate learning.

By relating the quantitative measures of interactions to the measures of technological learning achieved by the various manufacturing industries in Botswana, this study enables us to get a closer look at their ability to adopt, internalise, and diffuse new technologies, contributing to the country’s productivity growth. The remainder of the paper is organised as follows: the next section presents the technological learning process in a context of National System of Innovation (NSI) and explains how it is used as a framework to analyse knowledge flows and technological manufacturing productivity growth in the technology importing country. Thereafter, section three provides a snapshot of existing knowledge flows through capital goods imports and cross-border investments. The fourth section outlines the theoretical model linking business interactions to the manufacturing productivity and skills intensification in the country that is sourcing embodied technologies from abroad. It also presents the data used to evaluate how trade in capital goods and investment flows between South Africa and Botswana influence manufacturing productivity growth and shifts in skills intensity among importing firms. The findings are discussed in the last part of section four while the final section summarises and concludes.

2. Technological learning as an analytical framework for knowledge flows

International technological catch-up theories stress the importance of knowledge as a weightless production factor that can diffuse across national borders through various channels, including trade and investment flows that diffuse embodied and disembodied technological knowledge as well as management knowhow and practices (Barba Navaretti & Soloaga, 2001). Grounded in the endogenous growth literature, they emphasise two mechanisms of international technological knowledge transfer: transmission of ideas that can be traded independently from goods and the trade in intermediate inputs and capital goods that incorporate new ideas, known as ‘the lab-equipment model’ (Rivera-Batiz & Romer, 1991). These theoretical insights are confirmed by substantial empirical evidence such as Coe & Helpman (1995) and Keller (1997, 2002).¹

The amount and degree of sophistication of technology that developing countries can adopt and efficiently utilise, depend among others, on their supply of technical and managerial skills and their interactions with technologically more advanced countries (Benhabib & Spiegel, 2005; Criscuolo & Narula, 2008). For many, if not most developing countries, catching up technologically depends on the extent to which they are able to learn and position their systems of innovation to best take advantage of knowledge flows from developed as well as from other developing countries. Countries that succeed in transforming scientific knowledge and technical innovation into profitable economic productivity are the ones that become economically successful (Feinson, 2003). Such economies enjoy technological benefits because they possess a complex, integrated system of human capital, infrastructures, and institutions for translating new knowledge and innovation into economically viable new products and processes.

Such systems, now known as ‘national innovation systems (NIS) or national systems of innovation (NSI)’, have increasingly been recognised as both a supplement and an alternative analytical framework to recognised macroeconomic perspectives on development. The application of NSI as a framework of analysis must be articulated around the various functions that national innovation systems perform. This means that countries, industrial sectors and firms assess their success in effecting technological change by evaluating the overall performance of the functions of their innovation systems. However, as a consequence of contextual and institutional differences between the innovation system of the developed and developing countries, it has been argued that developing countries need their own specific approach to NSI (Edquist, 2001; Juma et al., 2001).

One of the arguments in favour of specific approach to NSI in developing countries is that it cannot be aligned simply with neoclassical theories of growth in developing countries as indicated by Lundvall (1997). This has led Edquist (2001) to propose the concept of systems of innovation for development (SID), which has a number of key differences with the NIS approach taken in developed countries:

• Product innovations are more important than process innovations because of the effect on product structure;

• Small, incremental innovations are more important and more attainable than radical ones;

• Absorption (diffusion) of existing technologies is more important than the development of innovations that are new to the world;

• Innovations in low- and medium-technologies are more attainable than those in high-technology systems or technology frontier.

Using this approach, the most important attribute of NSI is to stimulate technological learning. Although insufficient on its own, active learning is a necessary condition to achieving long-term sustainable development. This explains why development scholars have emphasised the building of absorptive capacity by developing nations or their ability to acquire, learn, and implement the technologies and associated practices already in use in developed countries (Dahlman & Nelson, 1995). The promotion of learning and national absorptive capacity through various NSI components is thus indispensable for long-term industrial and economic development.

According to the OECD (1987), the interplay of incentives and technological capabilities within an institutional framework gives rise to sustained economic growth. Lall (1992) uses three broad headings: physical investment, human capital, and technological effort to categorise technological capabilities at the national level. Incentives comprise three broad categories: macroeconomic stability, competition, and factor market conditions. Finally, institutions reflect the rules of the game that emerge from how markets function and facilitate transactions, interactions, and learning. Institutions act to alter capabilities and change incentives. They can also modify behaviour by changing expectations and attitudes. Following Lall (1992), this study uses this approach involving the interplay of incentives, capabilities and institutions to explore numerous factors influencing the technological learning that occur because of trade and investment interactions between South Africa and Botswana. This study uses this technological learning approach to probe the role of regional integration with South Africa in facilitating knowledge flows to Botswana. These knowledge flows are analysed by examining two transmission mechanisms, namely import of capital goods and cross-border equity investments.

In the next section we cast a glance on some of the elements of the systems of innovations in both South Africa and Botswana in order to gain some insights into how they relate to interactions and technology flows between the two countries.

3. South Africa-Botswana proximity and knowledge flows

With a PPP-adjusted gross domestic product (GDP) estimated at US$ 723.5 billion for 2015 (US$ 13,300 per capita), South Africa remains the economic and technological powerhouse of the continent (second only to Nigeria in terms of total GDP). It is also the knowledge hub of the continent, with world-class academic institutions as well as cutting-edge technological leadership in some domains. The country leads the continent in industrial output (40 per cent of total output) and mineral production (45 per cent). For its part, Botswana’s economy is still heavily dependent on raw diamond exports and faces many challenges in its efforts to move to high value-added activities in other industries in spite of consistent efforts aimed at economic diversification. Like other industrialising nations, Botswana has increasingly recognised that successful economic development is narrowly linked to the capacity to acquire, absorb, apply, and disseminate modern technologies within its economy. However, since most technological innovation takes place through deliberate research and development (R&D), and world R&D activities are highly concentrated in a small number of industrialised countries, Botswana will have to depend on technological knowledge developed outside its borders. It is important to bear in mind that the engineering and management skills required to acquire the capacity to optimise resources and assimilate the acquired technology are quite complex. Various kinds of high-quality training are needed to provide the personnel of the recipient firms the skills, knowledge, and expertise applicable to particular products and processes. Botswana’s absorptive capacity therefore requires further strengthening. The economic and technological leadership within South Africa however presents a considerable growth opportunity.²

The existing interactions between the two economies and the knowledge flows they represent already indicate the growth and diversification advantages of this proximity. Interactions with South Africa account for more than 60 per cent of total foreign direct investment stock in Botswana and more than 70 per cent of the needed capital equipment imports (Grobbelaar & Sotetsi, 2005; UNCTAD, 2015). Regional cooperation programmes play a non-negligible role in facilitating these interactions. The most important is the Southern African Customs Union (SACU), in which free and unimpeded trade takes place among members. SACU consists of Botswana, Lesotho, Namibia, South Africa and Swaziland.³

Botswana sources the lion’s share of its imports from its SACU partners, and mainly from South Africa. Table 1 shows the US $ value of capital equipment import from South Africa into Botswana over the period 1995–2013. The corresponding technological knowledge embodied in the imported equipment is part of the knowledge flows between the two countries.

Table 1. Value of equipment imports from South Africa into Botswana 1995–2013 (in thousands of US$)^a.

Year	1995	1997	1998	2001	2002	2003	2004	2005	2007	2008	2009	2010	2011	2012	2013
Type of machinery/equipt
Engines	316	389	380	296	179	128	220	242	947	221	115	162	333	233	422
Rotating motors	6421	7685	7787	6891	5726	7091	7823	9357	11305	19788	16247	16119	20161	20691	15137
Other power generating equipt	1314	1563	1592	1430	1367	1953	6302	7413	7787	12634	9218	10313	14204	12239	14333
Civil engineering and plant equipt	21363	24240	26268	28054	32436	66136	56462	50382	93424	111023	61059	63742	124591	189202	97059
Textile and Leather machinery	2505	2963	3075	2924	2137	3914	2848	3241	2070	2085	2316	1602	2505	2731	1996
Paper mills & pulp machineries	420	457	501	550	1194	535	685	201	662	429	996	516	281	525	360
Printing and book-binding machinery	2400	2798	2889	2709	3579	5426	5774	3688	7033	14084	16922	7385	10913	8920	4007
Food processing machinery (excl domestic)	2353	2703	2904	3035	2934	2774	8919	2271	3367	4243	3721	6307	5245	2948	3959
Other machinery, n.e.s.	12946	14684	16005	17337	18291	18403	18623	25532	36465	55542	32617	42182	75864	41728	41598
Machine tools removing materials	1540	1616	2046	2963	1704	1019	5804	1151	2403	4707	4320	3513	3375	4678	3098
Machine tools for working metals	300	336	384	460	253	275	814	389	1808	3074	1266	1265	2038	1255	776
Parts and accessories of machines n.e.s.	215	266	253	177	210	255	1456	189	725	448	318	1275	2162	1032	668
Metal working machinery, excl machine tools	1709	1966	2146	2333	1467	2620	2718	2779	3496	2902	2361	2837	7012	2540	1887
Mechanical handling equipt	8997	10233	10926	11239	15618	19359	22403	13415	18772	30826	14981	24211	37772	26138	18148
Other non-electr. machinery	11933	15999	13423	4019	3255	5281	5955	6158	7976	8913	7419	9088	11051	11749	8941
Non-electric parts of machinery	1577	1866	1954	1908	1045	1957	1953	1710	2070	2360	2446	2347	3008	2042	1975
Parts of machinery of groups 751–753	10787	12824	13042	11673	11598	14997	14243	11710	20908	17429	15615	11637	14747	17886	13248
Total	87098	102588	105574	97998	102992	152124	163002	139828	221217	290708	191938	204502	335261	346538	227611

^aBefore 2000, the trade of the member countries of Southern African Customs Union (SACU) was not reported separately, but for SACU as a whole. Therefore, for the period 1995 to1999, intra-trade figures of these countries are UNCTAD estimates.

Source: UNCTAD Statistics.

The Southern Africa Development Community (SADC)’s Regional Indicative Strategic Plan, recognises the importance of science and technology in stimulating economic development and regional economic integration. At the same time, the African Cooperation Unit within the Department of Science and Technology (DST) is also responsible for engaging with SADC partners to develop and strengthen national systems of innovation to provide scientific and technological solutions for sustainable socioeconomic development.⁴ Over the past few years, the DST has engaged in a number of projects in policy and capacity development to achieve these objectives. Such interactions, although they undeniably contribute to facilitating or speeding up technological knowledge flows, fall beyond the scope of the current analysis.

4. Economic proximity and productivity growth

4.1 Rationale and methodology

Capital goods imports embodying technological knowledge contribute to knowledge flows, productivity growth and technological rent spillovers into the importing country as has been demonstrated among others by Mazumdar (1999), Eaton and Kortum (2001) and Keller (2004). Cross-border equity investments contribute to technological knowledge flows by bringing capital, proprietary technology, technical skills training and advanced management practices to the host country, thereby increasing productivity and competitiveness of domestic firms (Djankov and Hoekman 1998; Aitken and Harrison 1992). Finally, licensing transfers disembodied technology in form of blueprints, designs, and production processes that enhance existing production methods and techniques in the host country. In addition to these market-transaction knowledge flows, various technological spillovers arise from demonstration effects (Akamatsu 1962), linkages, or allocative efficiency (Caves 1974). Applying the above argument on Botswana’s manufacturing sector, we can thus argue that manufacturing productivity growth is positively affected by the volumes of imports of capital goods, equity investments and the value of licensing agreements with technologically advanced countries in general and South Africa in particular.

Panel data of a sample of 340 manufacturing firms were used to analyse the effects of these knowledge flows. The intensity of the above three indicators of interaction has been contextualised in relation to the manufacturing productivity growth and technical skills upgrading. As already pointed out in the introduction, technological sophistication of firms occurs through two main mechanisms: continual shifting of resources towards high- end products as predicted by the LKSY product spectrum models and a general increase in the intensity of skilled input use among all types of production. In order to distinguish between these two types of increases in the sophistication of production, the model decomposes the growth rate of manufacturing-wide technological sophistication, e, between two subsequent periods, t−1 and t as the sum of two components:$\frac{\mathrm{\Delta }{e}_{jt}}{e_{jt}}=\frac{{\sum }_{j=1}^J\mathrm{\Delta }{e}_{jt}\overline{{\theta }_j}}{e_{jt}}+\frac{{\sum }_{j=1}^J\mathrm{\Delta }{\theta }_{jt}\overline{e_j}}{e_{jt}}$Here, e is the total number of technicians in manufacturing, expressed as a share of total manufacturing employment. Subscripts j and t indicate the industry and time period, respectively, θ_j is the jth industry’s share in manufacturing-wide employment, an overbar indicates the simple average over two time periods, and Δ is the difference operator for the period t−1 to t. The same expression can be used, mutatis mutandis, to decompose changes in manufacturing-wide technician wages as a share of some manufacturing-wide normalising variable (total wages, expenditures, or production). The first term on the right-hand side captures the change in manufacturing-wide technological sophistication due to within-industry deepening of technical intensity, and the second term represents the reallocation of technically skilled workers across industries. If the second term is positive, then the technician-intensive industries are growing relatively rapidly, indicating the type of resource reallocation consistent with LKSY-type productivity growth. In contrast, if all of the change in aggregate technical intensity comes from intra-industry deepening, there is no evidence of this type of broad resource reallocation. Nonetheless, it may still be the case that within particular 3-digit or 4-digit industries, resources are being shifted toward high-end products, in which case further disaggregation is needed to detect the LKSY growth mechanism.

4.2 Estimation model

Mutz and Ziesemer base their model on a modified version of a two-gap growth model with imported inputs as introduced by Bardhan & Lewis (1970). That model, advanced by Khan & Knight (1988), emphasises that for developing countries, imported inputs paid for by export are the major mechanism of growth in the relation between export and growth in the short run. In the Mutz & Ziesemer (2008) model, the importation of capital goods and the elasticity of export demand can explain the growth behaviour of developing countries. The simplifying assumption made of no domestic production of capital goods is a fair approximation for many least developed countries and is thus suitable to analyse the case of Botswana.

For the purpose of this quantitative analysis, the relationship between capital inputs used in production and the subsequent manufacturing value added over the considered period (1985–2013) is assumed to be defined through a production function of the Cobb–Douglas type, with constant returns to scale with respect to labour and capital inputs.(1) $Y_t=A_t{K}_t^{\alpha }{L}_t^{\beta }{\xi }_t$(1) where $K_t$ is the capital stock used in the production period t,$L_t$ is the labour input in efficiency units, while $A_t$ is the level of productivity or technology factor in the same period and ${\xi }_t$ a stochastic factor that measurement and observational errors. α and β are the usual input elasticities.

Since we assume no capital is produced domestically in the importing country, the available capital stock in each production period is determined by the accumulation of capital through imports and depreciation using the perpetual inventory method:(2) $K_t=(1-\delta )K_{t-1}+I_{t-1}$(2) where $K_t$ is the capital stock at the beginning of period t, $K_{t-1}$ is the capital stock at the beginning of the preceding period, $I_{t-1}$ it the capital imports during period t−1, whereas δ is the depreciation rate.

Taking the natural logarithm of Equation (1) we obtain:(3) $\text{ln}{Y}_t=\text{ln}{A}_t+\alpha \text{ln}{K}_t+\beta \text{ln}{L}_t+{\epsilon }_t$(3) Differencing the expression in Equation (3) between period t and t−1 gives the expression for output growth(4) $\text{ln( }{Y}_t/Y_{t-1})=\mathrm{l}\mathrm{n}(A_t/A_{t-1})+\alpha \mathrm{l}\mathrm{n}(K_t/K_{t-1})+\beta \mathrm{l}\mathrm{n}(L_t/L_{t-1})+{\epsilon }_t-{\epsilon }_{t-1}$(4) The expression for productivity growth is therefore obtained by subtracting the input share weighted growth in capital and labour form the output growth:(5) $\text{ln}\left(A_t/A_{t-1}\right)=\mathrm{l}\mathrm{n}\left(Y_t/Y_{t-1}\right)-\left(\alpha \mathrm{l}\mathrm{n}\right(K_t/K_{t-1})+\beta \mathrm{l}\mathrm{n}(L_t/L_{t-1}\left)\right)$(5) From the collected data on labour input in the manufacturing sector, the constructed capital stock according to Equation (2) and the productivity growth data compute with Equation (5) it is now possible to relate productivity growth data to imports of embodied technology. In order to estimate the contribution of capital goods import to manufacturing productivity growth, we substitute the expression for $\alpha \text{ln}\left(K_t/K_{t-1}\right)$in Equation (5) by its value derived from Equation (2):$\text{ln}\left(K_t/K_{t-1}\right)=\text{ln}\left(\right(1-\delta +I_{t-1}/K_{t-1}\left)\right)\mathrm{t}\mathrm{o}\mathrm{o}\mathrm{b}\mathrm{t}\mathrm{a}\mathrm{i}\mathrm{n}:$ (6) $\text{ln}\left(A_t/A_{t-1}\right)=\mathrm{l}\mathrm{n}\left(Y_t/Y_{t-1}\right)-(\alpha \mathrm{l}\mathrm{n}\left(\right(1-\delta +I_{t-1}/K_{t-1}))++\beta \mathrm{l}\mathrm{n}(L_t/L_{t-1})$(6) For a given depreciation rate, this equation expresses the growth of capital stocks as dependent on the rate of capital imports relative to existing capital stocks. In this analysis, the underlying assumption is that no capital goods are produced domestically, so that the capital stock increase is dependent on the capital goods imports from abroad. In considering the contribution of imported technology to the productivity growth, we add the disembodied technologies in the form of licensing to value of capital imports.

If we further decompose the production function of Equation (1) and express labour input as consisting of distinct skilled (Ls) and unskilled labour (Luns) input factors, we can also capture the effects of skilled labour in productivity growth. Skilled labour share is defined in this study as the percentage of employees with a vocational, technical or higher education training. The resulting expression for productivity growth can then be rewritten as:(7) $\text{ln}\left(A_t/A_{t-1}\right)=b0+b1\ast \mathrm{l}\mathrm{n}\left(I_{t-1}\right)+b2\ast \mathrm{l}\mathrm{n}\left(L{s}_t\right)+b3\ast \mathrm{l}\mathrm{n}(\mathrm{L}\mathrm{u}\mathrm{n}{\text{s}}_t)+\mathrm{b}4\ast \mathrm{l}\mathrm{n}(K_{t-1})+b5\ast \mathrm{l}\mathrm{n}(L{s}_{t-1})+b6\ast \mathrm{l}\mathrm{n}(\mathrm{L}\mathrm{u}\mathrm{n}{\text{s}}_{t-1})+u_t)$(7) where s_t is the share of skilled labour in period t and u is the error term.

Skilled labour is defined here as labour input of employees with at least completed secondary or technical vocational education or other formal or informal training leading to comparable qualifications (see also Hunt & Tybout, 1999).

We can control Equation (7) by adding a dummy variable representing foreign equity investment, which represents the influence and control position of equity owned by foreigners in each industry. In those industries, such as dairy, food and beverages, management practices and organisation structure can be significantly influenced by the South African and other foreign investors.

The next step is to use firm-level data and analyse for a selection of manufacturing entities the extent of skills intensification, entry of new firms and exit of existing ones, as well as expansion of the various manufacturing entities. A distinction is made between domestically owned firms and predominantly foreign controlled firms to get a picture of the differences between transnational and indigenous firms in terms of their technological learning capabilities. The increase in numbers of active firms in each industry over the considered period, and the expansion of the scope of activities within various firms provides a testing ground for the LKSY hypothesis and enables to check whether the manufacturing productivity growth came from the entry of newly created firms or from improvement in production techniques of existing ones. In order to account for the possible diversification into more technologically advanced firms with higher shares of skilled labour and higher productivity corresponding to Hunt & Tybout’s (1999) technological sophistication, we check whether industry level productivity increased because all firms became more sophisticated, or because of intra-industry market share reallocations toward more sophisticated firms.

Applying the Hunt and Tybout analytical approach, we examine the productivity growth on the industry level in all industries classified by the BSIC as belonging to the manufacturing sector. This analysis allows us to isolate the technological deepening by industry from the shift to more technologically sophisticated firms.

Each Δejt term in Equation (10) is decomposed into the effects of intra-industry changes in technical intensity, and the effect of changes in the allocation of skilled workers across firms. This exercise is basically the same as the sectoral decomposition; however, it is complicated by extra terms to deal with the entry and exit of producers over the sample period. The expression derived from the sectoral decomposition becomes:(8) $\mathrm{\Delta }{e}_{jt}={\overline{\alpha }}_j\left[\sum _{i=1}^I\mathrm{\Delta }{e}_{ijt}^c\overline{{\theta }_{ij}}+\sum _{i=1}^I\mathrm{\Delta }{\theta }_{ij}^c\overline{e_{ij}}\right]+\mathrm{\Delta }{\alpha }_j\left[{\overline{e}}_j^c-\frac{e_j^b+e_j^d}{2}\right]+(e_j^b-e_j^d)(1-{\alpha }_j)$(8) Here c, b and d indicate continuing, entering (beginning) and exiting (dying) firms, respectively, and i subscripts refer to firms belonging to a given industry, while ${\alpha }_j$ is the share of continuing firms in total employment within industry j.

The first term on the right-hand side resembles the first equation of the Hunt & Tybout (1999) model. Its components disaggregate changes in technician intensity among incumbent producers into two subcomponents: one is incumbent upgrading, and the other is shifts in market share among incumbents. The second term measures the effect of changes in the market share of incumbent firms. This term indicates that when incumbents are more intensive in skilled labour than entering and exiting firms, then reductions in the amount of turnover (increases in θj) will increase industry-wide technology intensity.

Finally, if entering firms are more technical-intensive than the exiting firms they replace, ongoing producer turnover will also increase industry-wide technology intensity. This replacement effect is described by the third term of the right hand side expression.

4.3 Data

The firm-level data used in this analysis are based on the data files compiled by the author from the records of the Enterprises and Establishments Register (EER) and Botswana Exporters and Manufacturing Association (BEMA) as of June 2014 and the survey conducted among manufacturing firms between May and August 2014. Over the considered period (1991–2013), basic manufacturing activities such as food and meat processing as well as the production of packaged or bottled beverages comprise the largest number of active firms.

Data on capital stock, capital equipment import, imports, equity ownership structure, employee skills and value added were collected on a total of 340 manufacturing entities found in operation between 1991 and 2013. Those firms were identified using the EER system. EER is a computerised database of enterprises and establishments in Botswana. It is mainly used as a sampling frame for economic surveys and contains relevant information on all business activities in the country.

Botswana’s capital import data were compiled from the records of the Botswana Central Statistics Office (CSO), a governmental department in the ministry of finance and development planning, and from the UNCTAD’s Comtrade database. The CSO records the current as well as the 1993 constant dollar value of capital imports from customs declaration documents. Goods declared at ports of entry/exit are classified according to the harmonised commodity description and coding system of Botswana, which is an adapted version of the internationally recognised harmonised commodity description and coding system. The official currency used in customs declaration documents as from the introduction of the Single Administration Document (SAD) in May 2002 is the Pula (local currency). Although goods can originally be declared in different foreign currencies, an exchange rate is given for any particular currency and this is used to finally convert that currency to the Pula. Total volumes of equipment imports from South Africa amounted to US$363.708 million for the year 2013, according to Comtrade data.

The manufacturing value added data of the selected manufacturing entities in the study sample were compiled by the author from the company records and survey data. Data on manufacturing employment labour input and capital investments, foreign equity ownership, specific shares of skilled vs unskilled labour input and technology licensing agreements were also gathered from company documents as well as surveys conducted by the author.

4.4 Results

The existence of a potential reverse causality between productivity growth and the measures of trade and investment calls for an estimation method that is robust to endogeneity. The application of Generalised Method of Moments (GMM) estimator is an efficient way to deal with this problem, even in the presence of unknown heteroskedasticity (Baum et al., 2003). The GMM estimator is constructed by exploiting the orthogonality conditions of the sample moments. The principle of its application is to create a set of estimating equations for the coefficients that make sample moments match the population moments. By exploiting the orthogonality conditions, the GMM estimator selects the best linear combination among a set of moment restrictions and produces a consistent estimator for any weighting matrix. Using GMM with heteroskedasticity and autocorrelation correction (HAC), allows for an efficient estimation even when some of the regressors are endogenous and heteroskedastic. With a judicious choice of instruments and lag length, the GMM can thus adequately tackle the endogeneity bias (Habiyaremye, 2016).

Due to the relatively short time span of our data, the estimation was carried out using the system GMM estimator to reduce the finite sample bias in standard errors. The use of System GMM estimator, derived from the estimation of a system of two simultaneous equations, was proposed by Blundell & Bond (1998) to avoid this type of bias that occurs for data with shorter time dimensions.⁵ The basic idea of system GMM is to estimate a system of equations in both first-differences and levels, where the instruments used in the levels equations are lagged first-differences of the series, as suggested by Arellano & Bover (1995). These instruments are valid under restrictions on the initial conditions. By exploiting additional moment conditions, system GMM can significantly reduce the finite sample bias (Blundell et al., 2000).

For the GMM estimation needed to deal with the endogeneity problem pointed out above, we thus also express the reverse causality equation with change in investment as a function of total factor productivity growth, change in skilled labour employment and change in unskilled labour employment, so as to estimate the system of simultaneous equations formed by this equation together with Equation (8) above:(9) $\text{ln}\left(I_{jt}/I_{jt-1}\right)=c0+c1\ast \mathrm{l}\mathrm{n}\left(A_{jt}/A_{jt-1}\right)+c2\ast \mathrm{l}\mathrm{n}\left(L{s}_{jt}/L{s}_{jt-1}\right)+c3\ast \mathrm{l}\mathrm{n}\left(\text{Lun}{\text{s}}_{jt}/\text{Lun}{\text{s}}_{jt-1}\right)+v_t$(9) (10) $\text{ln}\left(A_{jt}/A_{jt-1}\right)=b0+b1\ast \mathrm{l}\mathrm{n}\left(I_{jt-1}\right)+b2\ast \mathrm{l}\mathrm{n}\left(L{s}_{jt}\right)+b3\ast \mathrm{l}\mathrm{n}(\mathrm{L}\mathrm{u}\mathrm{n}{\text{s}}_{jt})+\mathrm{b}4\ast \mathrm{l}\mathrm{n}(K_{jt-1})+b5\ast \mathrm{l}\mathrm{n}(L{s}_{jt-1})+b6\ast \mathrm{l}\mathrm{n}(\mathrm{L}\mathrm{u}\mathrm{n}{\text{s}}_{jt-1})+u_t$(10) where $v$ is the error term for that reverse-causality equation.

The first step was to examine with firm-level panel data how cross-border investments as well as the imports of machineries and equipment impacted on the productivity of the recipient firms. To allow for a comparison between domestic and foreign firms, the study estimated regression coefficients separately for these two categories of firms. In total we included 86 foreign firms (most of them South African owned) and 254 domestically owned firms. This procedure was chosen in an attempt to overcome the constraints of the limited total number of firms in the Botswana manufacturing sector, which does not allow a sizable matched sample of domestic and foreign-owned firms with similar characteristics.

Table 2 presents the regression results of the system GMM estimation. The Hansen-J statistic tests for over identifying restrictions confirm the validity of the used set of instruments. All Sargan p-values of the reported statistics at the corresponding degrees of freedom are below 10%.

Table 2. Effects of skills and embodied technology imports from South Africa on manufacturing productivity in Botswana firms.

Dependent variable:
Variables	Manufacturing productivity growth
	Foreign-owned firms		Domestic firms
	Coeff	Std error	Coeff	Std err
Investment	0.1916*	(0.0951)	0.1376*	(0.0713)
Capital import	0.0869	(0.0534)	0.0705	(0.0467)
Skilled labour input	2.6342***	(0.0822)	2.3249**	(0.1167)
Unskilled labour	0.2153	(0.1419)	0.1742	(0.1187)
Investment (t−1)	0.9618***	(0.2729)	0.9312**	(0.4423)
Cap. import (t−1)	0.213***	(0.0742)	0.177***	(0.0671)
Skilled labour (t−1)	0.1152	(0.0745)	0.1639	(0.1368)
Unskilled lab (t−1)	0.2065	(0.2164)	0.3676	(0.1981)
Intercept	1.6138	(2.4916)	1.4697	(0.6908)
Regression fit	Eq.7		Eq9
Prob F-Test	0.000		0.000
Adj. R-squared	0.7436		0.6416
DW-stat	0.1843		0.2028
J-Stat	0.0021		0.0021
Number of obs.	1806		5334
n.J	3.7926		11.2014

*10% significance level.

**5% significance level.

***1% significance level.

Note: Instruments eq. 7: regressors + 11 (lagged dependent variable: 1 lag; lagged independent variables: 2 lags). Instruments eq. 9: regressors + 7 (lagged dependent variable: 1 lag; lagged independent variables 2 lags). GMM HAC standard errors in brackets.

The estimated results for foreign-owned firms display significantly positive coefficients for technology acquisition investments, skilled labour input and unskilled labour. The coefficient for investment is however only significant at the 10% level.

The coefficients for capital goods import and for investments and in the preceding period are positive and significant at 1% level. This implies that capital goods imports and technology investments have led to manufacturing productivity increase mostly in the subsequent period. For its part, an increase in the share of skilled labour input has a positive effect on productivity in the same period. In contrast skilled and unskilled labour inputs of past period are not significantly linked to productivity growth in the subsequent period. For domestically owned firms, the pattern is almost similar, even though the coefficients are generally smaller. Changes in unskilled labour input change have no significant effect on contemporaneous productivity neither among foreign firms nor among domestically owned ones.

Across the regressions (for domestic and foreign-owned firms), coefficients are thus positive and strongly significant for the lagged capital imports and the contemporaneous share of skilled labour, while the coefficient of contemporaneous capital import remained insignificant. This implies that the learning effects of embodied technology in imported capital on productivity growth are subject to time lags, while the effects of change in skilled labour input are likely to impact on productivity increase immediately. The Granger causality test confirms the precedence of capital goods import on the corresponding productivity growth. These results thus confirm existing view that capital imports in a given period influence productivity in subsequent periods.

The differences in coefficients between domestic and foreign-owned firms imply that foreign-owned firms tend to have slightly higher rates of productivity increase in response to changes in production factors with technological learning content (foreign capital goods with embodied technological knowledge and skilled labour). This can be attributed to the fact that foreign-owned companies tend to import technologically more sophisticated capital goods and implement management and organisational practices that increase productivity faster in comparison to domestically owned firms. Those differences in coefficients may thus suggest that firms with foreign connections may be more efficient in translating the technological content of their imported technological knowledge into productivity gains. The result is that they will face a steeper learning curve, which eventually will help them overcome their potential disadvantage in local market knowledge with respect to domestic firms.

The last step was to apply the Hunt & Tybout (1999) approach as presented in Equation (8) to decompose the changes in skills intensity in a selection of industries across the manufacturing sector over the period 1995–2013. A summary of the results of this analysis is presented in Table 3. From these results, what is noticeable is that technological sophistication as measured by skilled labour intensity, although increasing over the considered period (1995–2013), was mainly due to upgrading within each of the industries rather than a replacement of less technological sophisticated by a portion of the total increase in skilled labour intensity, while the entry of new firms slightly increased the level of skills intensity in the industry in the considered period.

Table 3. Sources of change in skills intensity of selected manufacturing industries in Botswana 1995–2013.

Industry	Average annual growth incumbent firms	Upgrading skills existing firms	Entry new firms	Total change
Beverages	−.026	0.546	0.214	0.734
Textile	0.017	0.039	0.019	0.075
Chemical products	0.061	0.466	0.041	0.568
Paper products	0.003	0.037	0.043	0.083
Furniture	0.005	0.041	0.021	0.067
Glass products	−0.003	0.033	0.018	0.048
Dairy	−0.002	0.017	0.014	0.029
Metal products	0.005	0.052	0.019	0.076
Average	0.005	0.015	0.045	0.061

Aggregate growth in technical intensity is therefore driven by a deepening of technological sophistication in all industries considered in our analysis, not because some more technologically performing industries grew relative to other industries. We therefore do not find convincing evidence of a capital import driven diversification in more technologically advanced firms or a shift in new industries.

In the past, a number of studies such as Chenery & Syrquin (1986) have documented a systematic shift of production away from simple manufactured products as the development process unfolds, thereby validating the LKSY hypothesis. In the case of Botswana, however, we do not observe an inter-industry shift of resources towards more productive industries at the expense of less technical intensive industries. This is not surprising given the relatively young stage of industrialisation in Botswana, in which most of industries are still in the growth phase. Interestingly, the current analytical decompositions along the line of those made by Behrmanet al. (1997), suggest that most of the rise in the skill intensity of production is due to skill deepening within existing industries rather than shifts in the product mix towards more skill-intensive industries. This implies that trade in capital equipment as well as concomitant knowledge flows between Botswana and South Africa, have mainly contributed to increased productivity in Botswana’s manufacturing firms through their association with the deepening the skills intensity within existing industries.

5. Summary and conclusions

In this paper, the NIS-based approach was used to analyse the role played by the interaction between South Africa and Botswana in fostering technology diffusion between the two countries and contributing to Botswana’s effort to reduce its dependence on diamond exports. Using industry-level data, the study investigated whether external interactions, technological effort and technological interactions within Botswana’s innovation systems have spurred an import-induced productivity growth. The estimated results from of the bilateral capital goods trade and investment data indicate that productivity in Botswana’s manufacturing sector has been influenced by interactions with their South African neighbour, namely by imports of capital goods that embody technological knowledge. Moreover, the skills intensity decomposition of the examined manufacturing industries show that most industries have been increasing their technical skills intensity, mainly as a result of incumbent technological and skills upgrading induced by interaction and business links with South African entrepreneurs.

Finally, the analysis results indicate that industries that have more linkages with technologically more advanced South Africa stand to gain more from their interactions. These results are however not statistically strong because of the embryonic stage of Botswana’s export activities outside the diamond sector. This suggests that South Africa can play the role of leading goose in a regional flying geese pattern (Akamatsu, 1962), involving its regional neighbours in Southern Africa. The potential gains from increased intensification of economic and business ties with South Africa as a technological leader are likely to be substantial, not only for Botswana, but for other countries in southern African sub-region as well.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 Barba Navaretti & Tarr (2000) provide a detailed review of the various theoretical models that conceptualise how trade and their interactions affect knowledge diffusion, together with the corresponding empirical evidence.

2 Nonetheless, some scholars, like Alden & Soko (2005), have alleged that South African companies operate like sub-imperial agents as they care less for backward integration or growing local capabilities of indigenous companies.

3 Some authors have argued that South Africa benefits more from this Custom Union than other members and that some of the initiatives that could have deepen further regional integration in the sub-region have been frustrated by the country.

4 Its major strengths include its physical and economic infrastructure, natural mineral and metal resources, a growing manufacturing sector, and strong growth potential in the tourism, higher value-added manufacturing, and service industries.

5 Indeed, as shown by Blundell & Bond (1998), standard GMM estimation (as in Arellano & Bond, 1991) has poor finite sample properties and is also downwards biased, especially when the time dimension T is small. The bias is only sufficiently small for T = 30 or more. If difference GMM was applied instead, such biases would otherwise make its inferences unreliable, as explained by Roodman (2006).