Africa's Manufacturing Puzzle: Evidence from Tanzanian and Ethiopian Firms

Abstract

Recent growth accelerations in Africa are characterized by declining shares of the labor force employed in agriculture, increasing labor productivity in agriculture, and declining labor productivity in modern sectors such as manufacturing. To shed light on this puzzle, this study disaggregates firms in the manufacturing sector by average size, using two newly created firm-level panels covering Tanzania (2008–2016) and Ethiopia (1996–2017). The analysis identifies a dichotomy between larger firms with superior productivity performance that do not expand employment and small firms that absorb employment but do not experience much productivity growth. Large, more productive firms use highly capital-intensive techniques, in line with global technology trends but significantly greater than what would be expected based on these countries’ income levels or relative factor endowments.

1. Africa's Anomalous Growth

Many economies in Sub-Saharan Africa were growing rapidly in the first two decades of the new millenium, faster than at any time since independence. Superficially, these economies were going through the classic structural change scenario: employment shares in agriculture were falling, and urban populations and occupations were expanding. Yet African growth accelerations were anomalous when viewed from the perspective of comparative development patterns. Aggregate labor productivity growth within manufacturing and other modern sectors has been disappointing. In fact, those countries where growth-promoting structural change was significant (Ethiopia, Malawi, Senegal, and Tanzania, especially) experienced negative to zero labor productivity growth within their nonagricultural sectors (Diao et al. 2019). This is especially puzzling in the case of manufacturing, the canonical modern sector. And it is the case not just for resource-dependent countries, but also for others such as Ethiopia that have made some progress in industrialization and attracting foreign investment in manufacturing from China and elsewhere (Abebe, McMillan, and Serafinelli 2022).

These facts are not consistent with a process of growth driven by productive improvements in manufacturing and other modern sectors (a supply-side model of growth). Diao et al. (2019) argued that a demand-side story presents a more plausible account. An increase in demand for urban products—whether due to transfers from abroad, public expenditures, or income gains in agriculture—could explain the observed structural-change patterns. It would also explain why productivity in the more modern parts of the economy lagged or declined: increased demand for modern-sector output could be met only through an expansion of less productive firms and activities at the margin. This perspective therefore made the authors skeptical about the sustainability of these growth accelerations.

This paper begins by confirming the negative correlation between growth-promoting economy-wide structural change, on the one hand, and labor productivity growth within nonagricultural sectors, on the other, for an expanded set of African countries. This analysis is based on the GGDC/UNU-WIDER Economic Transformation Database (ETD), which goes until 2018 and covers 18 African countries, compared to the earlier 10 Sector Database produced by the GGDC that only covered 11 African countries through 2011 (de Vries et al. 2021). The paper then focuses specifically on manufacturing in two countries, Ethiopia and Tanzania. To establish a benchmark, the study first compares employment trajectories in Tanzania and Ethiopia to two earlier East Asian industrializers, Taiwan and Vietnam. In all cases the analysis splits manufacturing employment into firms with 10 or more employees (formal) and the remainder (small and/or informal) by combining UNIDO's Indstat2 data with the ETD data. The contrast is stark. The share of formal sector manufacturing employment took off during the growth accelerations in Taiwan and Vietnam. In Tanzania and Ethiopia, it is the share of employment in small and informal firms that has expanded during the period of growth acceleration.¹

Why isn't the share of formal manufacturing employment expanding more rapidly in Tanzania and Ethiopia? To understand this, the study takes a closer look at the manufacturing sectors in both countries. The core of the analysis rests on two newly created panels of manufacturing firms, one for Tanzania covering 2008–2016 and one for Ethiopia covering 1996–2017. In both cases, the panel covers firms with 10 employees or more. But in the case of Ethiopia, it is possible to supplement the analysis with nationally representative surveys of small-scale manufacturing firms employing fewer than 10 workers, which are available for 2002, 2006, 2008, 2011, and 2014. With these data, it is possible to take a finer-grained look at employment and productivity patterns within smaller manufacturing firms.

These findings shed light on the nature and sources of manufacturing underperformance. In both countries, there is a sharp dichotomy between larger firms that exhibit superior productivity performance but do not expand employment much, and small firms that absorb employment but experience limited productivity growth. The problem lies not in the productivity performance of the larger firms, which is more than adequate, but in their inability to generate employment opportunities. The labor-absorbing firms, by contrast, are the smaller ones with significantly worse productive trajectories. The firm-size threshold where the productivity penalty kicks in seems to be different in the 2 countries: compared to larger firms, productivity growth is lower among firms with fewer than 50 employees in Tanzania and fewer than 10 employees in Ethiopia. But in both cases, the growing dualism within manufacturing—with the bulk of new employment in manufacturing being absorbed in the smaller and informal firms—depresses the productivity of manufacturing in aggregate.²

The bulk of the paper is devoted to demonstrating these new empirical findings and exploring their generality and robustness to various data considerations. Section 5 of the paper discusses a possible explanation for its findings, having to do with global changes in manufacturing technologies and patterns of specialization. The study suggests that the growing dualism that is observed may be linked to large firms’ decisions to adopt newer, significantly more capital-intensive technologies, to the detriment of their employment levels, in order to compete with foreign firms.³ Consistent with this hypothesis, the study shows that the larger, most productive firms in Ethiopia and Tanzania exhibit levels of capital intensity that approach (or exceed) those observed in the Czech Republic, a country that is around twenty times richer. The study sketches a model that qualitatively generates the outcomes that are observed when the global technology frontier moves out in a capital-biased direction.⁴

The outline of the paper is as follows. Section 2 provides a macro-overview of structural transformation in Africa. Using updated data, the study confirms the trends that are discussed in Diao et al. (2019). Tanzania and Ethiopia are compared to two East Asian cases, Taiwan and Vietnam, to highlight the anomalous expansion of small and informal manufacturing employment in the former cases. Section 3 describes the data, namely, the newly constructed firm-level panels for Tanzania and Ethiopia, and their construction. Since these panel data have not been used previously, the study also compares the firm-level characteristics of Tanzanian and Ethiopian firms to what has been previously reported in the literature for firms in developing countries. Section 4 presents the results of firm-level regressions where the analysis relates productivity and employment growth rates to firm characteristics such as size, ownership, and export orientation. The study presents a number of robustness checks and also presents comparable results at the industry level. This section documents the dichotomy between the large and small firms noted above. Section 5 documents the capital-intensity of Tanzanian and Ethiopian manufacturing firms, across different firm categories, taking the Czech Republic as the main comparative benchmark. The study also presents and discusses the interpretation of this result. Section 6 offers concluding remarks.

2. Duality in the Manufacturing Sector: Macroeconomic Evidence

In previous work, a puzzling pattern of growth in African countries was documented using data from the Groningen Growth and Development Centers’ (GGDC) 10 Sector Database (Diao et al. 2019). That study found that rapid aggregate labor productivity growth in Africa had been accompanied by weak-to-negative labor productivity growth in these countries’ nonagricultural sectors. In this section, the present study updates those results using the Economic Transformation Database (ETD) (de Vries et al. 2021), an updated and more expansive version of the GGDC 10 Sector Database. The significant advantage of the ETD is that it covers an additional 7 African countries, bringing the total to 18, and also extends to 2018; the present study does not include Botswana or Mauritius because their growth accelerations took place earlier. The updated analysis confirms the earlier results. The study also presents aggregate evidence from the manufacturing sectors in Ethiopia, Tanzania, Taiwan, and Vietnam that is consistent with the idea that the growth patterns in Africa are demand driven while those in Asia were supply driven.

Before updating the growth decomposition and plotting employment trends, the study gauges the accuracy of the total manufacturing employment numbers for Ethiopia and Tanzania reported in the ETD. It was found that for Tanzania the manufacturing employment numbers corresponded to what had been reported in the censuses of 2002 and 2012. For Ethiopia, the analysis combined manufacturing employment reported in the National Labor Force Survey (LFS), Urban Employment and Unemployment (UEU) surveys, and the Living Standards Measurement Study—Integrated Surveys on Agriculture. When comparing the study's own estimates to those in the ETD a similar trend in employment is found but considerably lower levels of employment. Since higher levels of employment would bias the study's labor productivity estimates downward, the study chooses to use its own more conservative estimates of total manufacturing employment for this section of the paper. This process is described in detail in “Adjusting GGDC Economic Transformation Database (ETD) Ethiopia Manufacturing Employment Estimates” in section S1 of the supplementary online appendix.

Recent Patterns of Growth

As in its previous work, the study decomposes economy-wide labor productivity growth into its between (structural change) and within components; an extensive discussion of the decomposition is provided in Diao et al. (2019). To home in on Africa's modern sectors, within-sector labor productivity growth is further decomposed into agriculture and an employment-weighted average of the manufacturing, trade services, business services, construction, and transport sectors. Figure 1 replicates the patterns shown in Diao et al. (2019) using the updated ETD numbers for African countries and this study's adjusted estimates for Ethiopia. The bars in fig. 1 are coded according to how much of labor-productivity growth comes from structural change (in grid) and how much comes from within-sector labor productivity-growth in agriculture (in diagonal lines) and in nonagriculture (in black). The main difference between fig. 1 and the previous results is that including the additional African countries and extending the period to 2018 shrinks the contribution of within-nonagricultural-sector productivity growth to close to zero.

Figure 1 shows that in Africa, prior to the post-2000s growth acceleration, average annual labor productivity growth was around 1.5 percentage points, 0.9 percentage points of which is from structural change while the within-sector nonagricultural component of labor productivity growth is 0.04 percentage points. After the growth acceleration, structural change contributed significantly to growth in Africa—the average annual labor-productivity growth rate rises to 2.55 percent in 2001–2018, and structural change contributed 1.86 percentage points of the total. This is not surprising since it is expected that the payoff to structural change would be greatest in poor countries. The contribution of within-sector labor productivity growth in the nonagricultural sector is −0.01. This is troubling in the sense that were this pattern to continue, labor-productivity growth would eventually peter out.

Figure 2a is a scatter plot of the relationship between within-sector productivity growth (in the nonagricultural sector only, horizontal axis) and the labor-productivity growth that arises as a result of structural change (vertical axis) for selected subsectors (including manufacturing) in African countries. The pattern revealed by fig. 2a is a negative correlation between these two components of overall growth. The five nonagricultural sectors—manufacturing, construction, trade, transport, and business services—are denoted by different shapes in the legend; manufacturing is represented by a shaded circle. Turning to the manufacturing sector in Tanzania, fig. 2a shows that the contribution of the manufacturing sector to within-sector labor productivity growth is 0.1 percentage points while its contribution to growth from structural change is about 0.3 percentage points. The trend is very similar in Ethiopia, where the contribution of the manufacturing sector to within-sector labor productivity growth is slightly over 0.1 percentage points and its contribution to growth from structural change is just over 0.3 percentage points.

For comparative purposes, fig. 2b shows the same correlation for 7 Asian countries during the first 10 years of their initial growth accelerations.⁵ In contrast to the African countries, the Asian countries exhibit a positive correlation between the within-sector and structural change components of labor productivity growth for each specific nonagricultural sector. In all seven countries, the manufacturing sector contributed positively to within-sector labor productivity growth and the growth that comes from structural change. For example, the manufacturing sector in Thailand contributed around 0.55 percentage points to growth from structural change; this is very similar to the numbers for Ethiopia and Tanzania. But unlike Ethiopia and Tanzania, within-sector labor productivity growth in Thailand's manufacturing sector was about 1 percentage point.

Summarizing, in Asia well-performing nonagricultural sectors have contributed to economy-wide productivity growth both by drawing labor from lower-productivity sectors and by experiencing rapid productivity improvement. In Africa sectors outside of agriculture have not performed well and have only contributed to economy-wide labor productivity growth through structural change. To gain a better understanding of the underlying causes of these patterns, the analysis will next disaggregate manufacturing employment data into employment in relatively large formal firms and employment in smaller typically informal firms. Because the study does not have micro data that covers the universe of small firms, it is not possible to precisely determine the formality status of the smaller firms and so label these firm small/informal.

Explaining Patterns of Growth: Manufacturing Employment Trends

The patterns described in Africa are especially puzzling when it comes to manufacturing, the canonical “modern” sector. In the researchers’ previous work, they hypothesized that these patterns might be explained by differences in the sources of structural change. A simple model was developed to highlight the differences between demand- and supply-driven structural change. Supply-driven structural change in this model was captured by a positive productivity shock to the modern sector (in this case manufacturing) allowing it to draw labor from other, less-productive sectors of the economy. This case was associated with the East Asian example. By contrast, demand-driven structural change was a result of positive aggregate demand shocks, possibly as a result of public investment, external transfers, or increases in rural incomes. The study associated this with the African case, where in practice the aggregate demand shock was likely the result of a combination of all three.

These underlying differences in the roots of structural change produce differential growth rates in both formal- and informal-sector manufacturing employment trends across the two continents. To the extent that structural change is supply driven one would expect to see an expansion of modern-sector (or formal) activity in the manufacturing sector. Demand-driven structural change on the other hand is likely to be accompanied by the entry of less-productive or informal manufacturing firms at the margin.

To explore this hypothesis and assess trends in the manufacturing sector, the analysis combines the sector-level employment data for manufacturing from the Economic Transformation Database (ETD)—using the adjusted estimates for aggregate Ethiopian manufacturing employment—with manufacturing employment data from the United Nations Industrial Organization (UNIDO) Indstat2 database. The ETD manufacturing employment data largely follow the methodology of the GGDC's previous sector databases and is primarily based on population census data covering both the formal and informal sectors (de Vries et al. 2021). By contrast, Indstat2 records manufacturing employment data for formal-sector firms in the manufacturing sector (UNIDO 2020). Although country statistics sometimes vary in terms of the size of establishments covered, typically Indstat2 covers firms with 10 or more employees.

As a data check, the study compares the UNIDO data that purportedly capture employment in manufacturing firms with 10 or more employees in Ethiopia, Tanzania, and Vietnam to the study's own estimates using firm-level census data from each of these three countries (the study does not have firm-level census data for Taiwan). Figure S1.3 in the supplementary online appendix shows that the trends in employment using firm-level data are nearly identical to the trends using the UNIDO data.⁶

For each of Tanzania, Ethiopia, Taiwan, and Vietnam, the study computes small and informal-sector employment in the manufacturing sector as the difference between total employment (ETD) and formal-sector employment (Indstat2). The analysis then plots the levels of employment in the small and informal sector versus the formal sector, as well as total manufacturing employment for each country over time. The top panel of fig. 3 shows the results of this exercise for Ethiopia and Tanzania. The most striking trend common to both countries is a high and increasing share of small and informal manufacturing employment.⁷ The beginning of the rise in small and informal-sector employment coincides with the beginning of the growth acceleration in both Tanzania and Ethiopia. McMillan and Zeufack (2022) conduct the same exercise for 10 additional countries and find similar patterns for the low-income African countries. This dominant role of the informal sector in employment is also reflected in household-level data (Fox et al. 2013).

The bottom panel of fig. 3 shows starkly different patterns for Taiwan and Vietnam. Unlike the African cases, the share of formal-sector manufacturing employment in Taiwan and Vietnam is considerably larger than the employment share of small and informal-sector firms. Moreover, the figure displays the enormous shift in the composition of manufacturing employment, which took place over a relatively short period of time in Vietnam; this pattern does not appear in Taiwan because it industrialized earlier. Both Taiwan and Vietnam were largely agrarian societies before the start of their growth accelerations (Huang 1993; McCaig and Pavcnik 2017). Although not shown, increases in formal-sector manufacturing employment in these two countries coincided with gradual declines in agricultural employment shares.⁸ These stylized facts are consistent with the idea that structural change in these two countries was a result of positive supply shocks to the manufacturing sector.

The evidence presented so far seems to indicate that formal-sector manufacturing firms in Tanzania and Ethiopia have not absorbed large numbers of workers. The opposite is true for Taiwan and Vietnam. This evidence is consistent with the hypothesis that structural change in African countries—expansion of manufacturing and other urban “modern” sectors—has been demand driven.

Tanzania and Ethiopia have both made industrialization cornerstones of their growth plans. And Ethiopia has been heralded as the China of Africa in numerous news outlets. Its low wages and generous incentives for foreign investors have been viewed as omens of successful industrialization. The languishing of formal manufacturing sectors in these countries raises numerous questions. Why is formal manufacturing-sector employment lagging? Are formal manufacturing firms in Tanzania and Ethiopia performing poorly? To answer these questions, the study turns to firm-level analyses of the manufacturing sector in each of these two countries using newly constructed longitudinal data for the period 2008–2016 for Tanzania and 1996–2017 for Ethiopia. The analysis is preceded by a description of the data.

3. Describing the Firm-Level Data

Analyses of firm performance in the manufacturing sector in low-income Africa have been hampered by data limitations. This paper makes use of two newly created panels of manufacturing firms, one for Tanzania covering 2008–2016 and one for Ethiopia covering 1996–2017.⁹ This section begins by describing these datasets, their construction, cleaning, and shortcomings and reports summary statistics in table 1. As this paper is the first to use the Annual Survey of Industrial Production (ASIP) data in panel format¹⁰ and the first to use the Large and Medium Scale Manufacturing Industries Survey (LMSM) extended panel, and then compare employment and labor productivity in levels by firm type to what researchers have reported in the literature about firms in developing countries.

Data Description and Summary Statistics

The study created the Tanzanian panel dataset for the period 2008–2016 from repeated years of the Annual Survey of Industrial Production (ASIP), conducted by the National Bureau of Statistics (NBS) of Tanzania.¹¹ The Tanzanian government has conducted the Annual Survey of Industrial Production (ASIP) since the early 2000s and has published the ASIP analytical and statistical reports routinely since 2008 (NBS 2010a, 2010b, 2012, 2016a, 2016b, 2018a, 2018b). ASIP is meant to cover all industrial establishments in the country that employ 10 persons or more. This includes mining, manufacturing, and utility sector firms. The study limits the analysis to the sample of manufacturing firms. NBS assigned consistent firm identifiers across the years 2008–2016, which were used directly to create a firm-level panel. There were some changes to the questionnaire over time, but key variables are consistently reported.

To check the coverage of the ASIP data, the study compared estimates of the number of manufacturing firms (broken up by employment size class) in the ASIP to what is recorded in Tanzania's Central Registry of Establishments (CRE). The study also compared estimates of total employment to the weighted estimates from Tanzania's National Panel Survey (NPS) and found similar levels of aggregate employment. The 2013 year of the ASIP was technically a census of all sizes of manufacturing firms and has sampling weights; the analysis uses these NBS-provided weights in the analyses for the year 2013. More details on these checks can be found in “ASIP Census Coverage” in section S1 of the supplementary online appendix.

The study created the Ethiopian panel dataset for the period 1996–2017 from repeated years of the Large and Medium Scale Manufacturing Industries Survey (LMSM), conducted by the Central Statistical Agency (CSA) of Ethiopia (CSA 2001, 2004, 2008, 2011a, 2011b, 2011c). Like ASIP, the LMSM survey is meant to cover all manufacturing establishments in the country that employ 10 persons or more and that use power-driven machinery. While the LMSM survey does not cover mining, it does include some utilities firms. The study's analysis is limited to the sample of manufacturing firms. Other researchers have worked with a version of the panel up to 2013; this paper is the first the authors are aware of to use an extended version of the panel connecting the older years up to 2017. The Ethiopian government assigned consistent panel identifiers from 1996 to 2011; the panel identifiers used to connect the later years of the survey were developed by Abebe, McMillan, and Serafinelli (2022) and a team of researchers based at the Ethiopian Development Research Institute and Oxford University. More details on the process of creating consistent panel identifiers are provided in “Ethiopia Large and Medium-Scale Enterprise Survey” in section S1 of the supplementary online appendix.

Importantly, both datasets are enumerated at the establishment level. Thus, in some cases the firms that are being studied are part of a group of firms operated by a parent company; the study's within-plant estimates of employment growth do not capture the opening of new plants by a parent company. However, the number of multi-plant firms in Tanzania and Ethiopia is very low; the study uses information from a technology transfer module administered with the ASIP in 2016 and the LMSM in 2016/17 to measure this: just 6 percent of firms in Tanzania and 5 percent of firms in Ethiopia report being part of a larger firm. Furthermore, activity as a result of new plant openings by parent firms will be measured in the industry-level analysis.

For much of the analysis, the article splits firms by firm size. It defines firm size based on average employment over the lifetime of the firm. This way of defining firm size is akin to correcting for mean reversion by averaging employment over two periods when studying annual average employment growth as described in Haltiwanger, Jarmin, and Miranda (2013). The study tests an alternate definition of firm size in which size is defined according to the average level of employment in the first two years that a firm is observed in the panel. The key results are consistent between the two alternate definitions of firm size and are available upon request.

To select firm size groups, the analysis first splits firms into the size groups 10–19, 20–49, 50–99, 100–499, and 500 + workers, and examines growth rates in employment and labor productivity within each of these size groups. The results of this analysis indicate that in Tanzania there is little to no productivity growth in the two smallest firm-size categories and significant labor productivity growth in the larger firm-size categories; within-firm employment growth is an imprecisely measured zero across all size categories. By contrast, in Ethiopia, within-firm employment growth is largest in the smaller firm-size categories including the 50–99 category, while productivity growth is roughly similar across size groups. Thus, for ease of comparison across countries the study defines small firms as those with between 10 and 49 employees and large firms as those with 50 or more employees. For Ethiopia, the analysis tests the robustness of the results to redefining small firms as those with between 10 and 99 employees. (These results are presented in “Ethiopia with 100+ Employment Cut-Off” section S2 of the supplementary online appendix.) And for both countries, the study shows in the robustness tests that these size groups are robust to controlling for firm age.¹²

Employment in Tanzania includes all workers who are hired on more than a month-long basis—this should include some temporary and seasonal workers, but likely not all. The ASIP data do not allow for further disaggregation into permanent vs. short-term. The LMSM data for Ethiopia, however, do include permanent and temporary/seasonal workers separately. The study limits the bulk of its analysis to permanent workers because at the firm level the reporting of seasonal/temporary workers is highly variable. CSA's manual instructs enumerators to get the person/months equivalent of seasonal employment by multiplying the number of months worked by the number of workers separately for each cohort of seasonal/temporary workers, then summing the products before dividing by 12; only the final calculation is reported. By comparison, the World Bank Enterprise Survey (WBES) data for Ethiopia (covering the year 2014) do not ask enumerators to do any calculations and instead collects the number of temporary workers and the average number of months worked by temporary workers. As a check on the number of full-time equivalent temporary workers in the LMSM, the section titled “Ethiopia Tests with Seasonal and Temporary Workers” in section S2 in the supplementary online appendix compares the seasonal/temporary workers reported in the LMSM to the number calculated from the WBES. This exercise reveals a lower average number of full-time equivalent seasonal/temporary workers per firm in the WBES—37 workers—compared to the LMSM data, which have an average of 74 seasonal/temporary workers hired per firm; this difference is not accounted for by a greater share of firms in the LSMS hiring seasonal/temporary workers, as in the WBES 69 percent of firms report hiring temporary/seasonal workers compared to 42 percent in the LMSM. Nevertheless, the study conducts additional robustness checks using the LMSM data and including the reported seasonal/temporary workers in the measure of employment; these results are reported in “Ethiopia Tests with Seasonal and Temporary Workers” in section S2 of the supplementary online appendix.

Labor productivity is measured as value-added per worker; value added is total sales minus raw material costs, and employment is measured as number of permanent workers.¹³ In addition to comparing large and small firms, firm types also include exporter and ownership status. The variables for exporter, foreign, and public are all time-variant and depend on the firm's reporting in the current year. A firm is defined as an exporter if it exports any of its production, defined as foreign if it reports foreign or joint venture ownership, and similarly defined as public if it reports public ownership. More details on variable creation and cleaning are in “Variable Definitions and Cleaning” in section S1 of the supplementary online appendix. Summary statistics for the final samples are provided in table 1.

The study's ASIP sample covers 3,582 unique firms in Tanzania over 2008–2016, and firms are observed in the panel on average 2.4 years. Firms in Tanzania have 84 workers on average, average value added is USD 1.7 million (TZS 3.7 billion), and the average value of the capital stock is USD 1.2 million (TZS 2.7 billion). Large firms account for 26 percent of the sample. 17 percent are foreign, 10 percent are exporters, and 4 percent are public.

The LMSM sample covers 9,626 unique firms in Ethiopia over 1996–2017, and firms are observed in the data on average 3.1 years. The average number of workers is 87, while average value added is USD 885,000 (birr 19 million) and the average value of capital stock is USD 714,000 (birr 16 million). Large firms comprise 29 percent of the sample , 5 percent are foreign-owned, 5 percent are exporting, and 7 percent are public.

Employment and Labor Productivity Levels Estimates

As this paper is the first to use the ASIP data in panel format and the first to use the LMSM extended panel, it is useful to check that it gets reasonable outcomes when comparing the employment and productivity outcomes by firm size, ownership status, and exporter status. This is tested through a set of pooled cross-sectional regressions looking at differences in levels of value-added per worker and employment across the years in the panel according to firm characteristics. The estimating equation is:

Where i references firm, |$s\ $|industry, |$r\ $|region, and t year; |${y}_{ist}$| is the firm's value added per worker or employment. The study examines the relationship between levels of |${y}_{ist}$| and firm size (large indicates firms with 50 or more employees), exporter, foreign, and public status. The study also includes dummies |${\theta }_t\ $|for each year covered by the panel (reference year 2008 for Tanzania and 1996 for Ethiopia) and includes industry and region dummies |${\mu }_s$| and |${\sigma }_r$|⁠. The results are presented in table 2. After controlling for industry, region, and year, the results are remarkably similar between the two countries—exporters and foreign firms have higher employment and labor productivity on average than their non-exporting, and domestic counterparts. These results are consistent with the evidence presented in Harrison and Rodriguez-Clare (2010), Abebe, McMillan, and Serafinelli (2022) and Verhoogen (2023).

4. Duality in the Manufacturing Sector: Firm-Level Evidence

This section presents estimates of employment and labor productivity growth within firms and then within industries, with breakdowns by average firm size. In the case of Ethiopia, the analysis further examines employment and labor productivity growth in mechanized manufacturing firms with less than 10 employees using the Small-Scale Industries (SSI) data.

The penultimate subsection contains robustness tests. The study addresses concerns in the literature about biases associated with estimating the relationship between firm size and growth without accounting for firm age (Haltiwanger, Jarmin, and Miranda 2013; Martin, Nataraj, and Harrison 2017) and also discusses the issue of coverage of firms in Ethiopia's industrial parks, which began to attract more significant investment around 2016 (McMillan and Zeufack 2022). Additional exercises are presented in the supplementary online appendix, including 1) the study's main firm-level growth results including all interaction terms simultaneously, 2) an alternate employment size cut-off for Ethiopia, and 3) testing a measure of employment for Ethiopia that includes both permanent and seasonal workers. The article concludes with a summary of the results from this section.

Employment and Labor Productivity Growth: Within-Firm Estimates

To describe employment and labor productivity growth and how they vary depending on firm attributes such as size and ownership, the analysis begins by estimating a set of regressions with firm fixed effects in which the log of employment and value added per worker are regressed on a linear time trend. Again, firm size is measured based on average employment over the entire period observed. Estimating growth in this way gives an estimate of long-run within-firm growth by size class which incorporates transitory shocks. An important issue that is not tackled in this paper is the extent to which small firms transition to large firms and the conditions under which this happens. Subsequent sections explore the extent to which these results are robust to entry, exit and firm age. Thus, the estimating equation is:

where i references firm, t year, and f are firm fixed effects; |${y}_{it}$| is the firm's value-added per worker or employment. The coefficients in this regression may be interpreted as within-firm annual average growth in value-added per worker and employment; these coefficients are presented graphically in fig. 4. In addition to looking at differences between growth in large and small firms, the analysis also runs iterations of equation 2 that compare exporters and non-exporters, foreign and domestic firms, and public and private firms. Unlike the firm-size variable, these alternate firm characteristics are allowed to vary over time. These results are also plotted in fig. 4.¹⁴

The left-hand side of fig. 4 shows the results for labor productivity (top panel) and employment (bottom panel) growth in Tanzania. Labor productivity growth is significantly positive only in large (>50 employees), exporting, foreign, and private firms, with large and exporting firms displaying the fastest rates of growth (around 8 percent per year for large firms). Meanwhile, within-firm employment growth is close to zero across all firm types. The absence of employment growth in the larger firms that experience positive productivity growth is especially striking. A potential explanation for this could be the fact that the data are at the plant level. However, as noted, the number of multiplant firms in Tanzania and Ethiopia is negligible.¹⁵ In any case, the addition of new plants by parent firms will be picked up in the article's industry analysis.

In Ethiopia, labor productivity is growing on average across the entire spectrum of firms with more than 10 employees at a rate around 4.1 percent per year. Employment within small firms grows at 2 percent per year. Larger firms (>50 employees) have lower employment growth, but it is still positive, averaging a little more than 1 percent per year; this difference persists even after controlling for the positive effects of exporting, foreign, and public firms (which have significant overlap with the large firm group).¹⁶ While average employment growth is positive across all firm types, exporters, foreign firms, and public firms all have higher growth on average than their counterparts.

Entry and Exit: Within-Industry Estimates

Table 3 reports entry and exit (split by firm size) over the sample period for Tanzania and Ethiopia. For comparative purposes, the study also includes entry and exit in the Vietnamese manufacturing sector. Average annual rates of entry and exit (share of firms that enter in a given year and share of firms that exit) in Tanzania are 37 and 27 percent, respectively, while in Ethiopia it is 28 and 22 percent. In Vietnam, the rates of entry and exit are lower at 22 and 16 percent of firms per year on average. Rates of entry and exit are also higher among small firms than large firms in all three countries. Small firms enter at a rate of 41 percent per year in Tanzania, 30 percent in Ethiopia, and 23 percent in Vietnam, while they exit at rates of 29, 25, and 14 percent, respectively. Large firm entry rates are 26 percent in Tanzania, 20 percent in Ethiopia, and 13 percent in Vietnam, while exit rates are 23, 15, and 7 percent.

One concern with the Ethiopian numbers is that errors in the study's procedure for matching firm identifiers post-2011 could artificially inflate rates of entry and exit (the analysis does not have the same concern with the Tanzania panel because the firm identifiers are government-assigned). To determine whether the matching procedure inflated entry and exit, the article also reports entry and exit for the period over which the CSA did not change firm identifiers: 2000–2011. This fourth set of results in table 3 indicates slightly lower but still high rates of entry and exit; for large firms, the rates of entry and exit are comparable to those for large firms in Vietnam. This gives confidence that the entry and exit rates reflect real activity, but also reinforce the importance of looking at the aggregate trends in addition to the within-firm component.

These results make it clear that all three countries experience high rates of entry and exit, a point that section 5 will come back to. For now, a possible concern is that the within-firm estimates of employment and productivity growth may not reflect economy-wide trends in the presence of such high rates of entry and exit. The issue is especially relevant for the small firms in the sample that often appear in the data only one time.

To account for this, the study constructs sector aggregates of employment and value-added per worker using the firm-level data and regress these on time, at the 2-, 3-, and 4-digit ISIC levels alternately. Sector aggregates are estimated by taking the sum of value added and employment across all firms at the sector level. This is done for all firms, and then separately for the groups of small and large firms, again defining firm size using the average size of the firm over its lifetime. Employment and labor productivity growth are then estimated according to equation (3), where s denotes sector and t year, and y denotes value added per worker and employment, respectively, controlling for industry fixed effects |${\mu }_s$|⁠):

The analysis plots the coefficients at the 4-digit level in fig. 5, which shows that at the sector level in Tanzania labor productivity growth is positive only among large firms, averaging 7 percent per year. Employment growth is predominantly taking place among small firms (10–49 employees) at around 8.6 percent per year. These results combined with the high rates of entry relative to exit among small firms suggest that much of the employment growth in the small firms is a result of firm entry.

At the sector level in Ethiopia, industry-level labor productivity grows at 6.4 percent per year on average and is similar between the small and large firms. By contrast, industry-level employment growth is higher in small firms at about 6.3 percent per year on average, while employment growth averages 3.5 percent in large firms.

The sector-level results further inform the findings at the firm-level: in Tanzania there is employment growth in small firms due to entry, and the analysis confirms that productivity growth is primarily present in large firms. In Ethiopia, productivity and employment growth are present in both small and large firms, but there is evidence at both the firm level and sector level that employment growth is more rapid in small firms.

Small-Scale Industries

In terms of employment growth, the overall picture looks more favorable for Ethiopia than Tanzania when focusing on the sample of larger formal firms. The LMSM and ASIP surveys do not cover firms with fewer than 10 workers, and in the case of Tanzania, the only nationally representative data that cover small-scale manufacturers is the 2013 Census of Industrial Production (NBS 2016b). The study uses these data to explore the firm-size distribution in Tanzania but is unable to analyze employment and productivity growth in these smaller firms.

The CSA in Ethiopia conducts periodic nationally representative surveys of Small-Scale Manufacturing Industries (SSI). The raw data from these surveys are available for 2002, 2006, 2008, 2011, and 2014 (CSA 2003, 2006, 2010a, 2010b, 2014) and cover manufacturing establishments that use power-driven machinery and engage fewer than 10 workers; each of these datasets includes sampling weights. Summary statistics for the CIP and SSI data are provided in table 4. The measures are weighted using the sampling weights provided in the data, and the study lists both the unweighted and weighted number of observations covered. The data for Ethiopia come from 38,851 observations representing 282,128 firms. The average number of permanent workers is low at 3 workers per establishment and the average value added of these firms is USD 3,494.¹⁷ In Tanzania, firms with 10 or fewer workers in 2013 also hire 3 permanent workers on average, and their average value added is USD 2,067. While average firm size is similar across the two countries, the finding that value added (VA) is higher in the Ethiopian firms is likely due to the fact that the SSI only samples mechanized firms.

Although the data for Ethiopia are repeated cross-sections making it impossible to conduct within-firm analyses, it is possible to conduct sector-level analyses of employment and value-added growth. The study therefore estimates growth in labor productivity and employment among small-scale firms at the sector level also using the equation 3 in “Employment and Labor Productivity Levels Estimates” in section 3. The analysis further tests iterations of these results with two alternate measures of employment, one counting owners as workers and one counting both owners and seasonal workers; the study retests both the value added per worker and employment growth regressions and gets similar results. The study also considers whether the results differ for firms that operate full time and tests two different samples where “full time” is measured as the firm operating at least 10 months and 12 months; the results are consistent in both iterations. These results are not provided but are available upon request.

The main results of estimating equation (3) for the SSI firms are reported in fig. 5 along with the industry-level results for the LMSM firms. The first thing of note is that both the employment and labor productivity growth estimates are considerably noisier than the results based on the LMSM data. This is not surprising in that the barriers to entry are significantly lower for the SSI firms, and so there is likely to be more heterogeneity in the performance of these firms. It is also not surprising inasmuch as smaller firms find it more difficult to keep accurate accounts. While imprecisely estimated, the results in fig. 5 indicate practically no productivity growth among the SSI firms but strong employment growth averaging around 17 percent per year (consistent with macro data finding that smaller firms are exhibiting rapid employment growth without accompanying productivity growth). With this, the evidence from Ethiopia becomes more similar to the evidence from Tanzania—the smallest firms in Ethiopia have rapid employment growth, but their labor productivity growth is close to zero.

A natural question arises as to whether the smaller largely informal firms produce the same goods that are produced by firms in the formal sector. To investigate this, the study disaggregated sectoral employment share by firm size for both Tanzania and Ethiopia, shown in fig. S1.7 in the supplementary online appendix. Visual inspection of the three size categories in each country suggests a significant overlap between what is produced by informal- and formal-sector firms. For example, the largest share of employment falls into the food category across all size groups. Of course, the quality of the products produced varies across firm sizes, and although the study does not have price data, some of these quality differences will be reflected in the value-added numbers. It may also be that some of what is classified as food manufacturing by informal firms would be better categorized as services. For example, grain milling falls under food and occurs across all size groups. In the informal sector a grain miller will sometimes be providing a service to a farmer who brings grain to be milled, taking it away in her own sack. By contrast, larger firms purchase grain from farmers to be milled and sold commercially in supermarkets. But other “food” activities in the informal sector fall squarely in manufacturing. For example, craft brewers make and sell traditional beer much the same as formal firms make and sell beer albeit in larger quantities. For its own purposes, the study sticks with the versions of the International Standard Industrial Classifications used by the governments of ETH and TZ and leaves this issue to future research.

Robustness Tests

Employment Growth Controlling for Firm Age

Haltiwanger et al. (2013) find that once one controls for firm age, the negative relationship between firm size and employment growth disappears and may even reverse sign. The present study explores the relationship between firm age and employment growth following the methodology proposed by Haltiwanger et al. (2013). First, annual average employment growth is computed using average employment between t and t−1 in the denominator instead of employment at the beginning or end of the period as follows:

The estimating equation for employment growth by size and age is as follows:

where |${s}_{ij}$| is a categorical variable representing the employment size of firm i in industry j, and |${a}_{ij}$| is the categorical variable for firm age. Firms are assigned to a size category—10–19, 20–49, 50–99, 100–249, 250–499, and 500 + workers—based on their average employment in period t and t−1 and are assigned to an age category (1–9 or 10+) based on its current age in each period t. The analysis includes year dummies |${T}_t$| and controls for industry at the 2-digit ISIC level with |${\mu }_j$|⁠. Following Haltiwanger et al. (2013), the study presents versions of the results from equation 5 with and without age controls as well as versions that measure exit and entry compared with versions limited to continuing firms only. All these results are presented graphically in fig. 4.

Overall, the results in fig. 6 corroborate the main results. Employment grows more rapidly in small firms even controlling for firm age, and this result is largely driven by firm entry. In Tanzania, among continuing firms there is very low employment growth on average, but it is greater among large firms. Once the analysis introduces entry and exit, however, there is a clearly declining relationship between firm size and employment growth. In Ethiopia, employment growth declines slightly with firm size among continuing firms; this relationship becomes stronger when entry and exit are introduced (though it is less strong with the inclusion of firm age controls).

Employment in Ethiopia's Industrial Parks

The CSA's LMSM data did not cover many of the firms in the recently opened industrial parks. To gauge the impact of this omission on the employment estimates the study obtained the following: (1) a complete list of industrial parks and their initial year of operation; (2) a complete list of the firms operating in these parks and the number of workers employed by these firms produced by the Ethiopian Investment Commission (unpublished) in 2019. Between 2013 and 2017 three industrial parks were established: Eastern IP in 2013, Bole Lemi in 2016, and Hawassa in 2017. Using firm names, the study matches firms in the industrial parks to firms in its panel. This process is imperfect because the study does not have the firm names for all observations in the panel data, and it is possible that the firms are included in the data even if it is not possible to identify them.

In the largest industrial park, Eastern IP, the study matched 23 out of 91 firms on the list, accounting for about 40 percent of employment in the park in 2019 or approximately 5,946 workers (missing 8,919). In Bole Lemi IP the study is able to match 7 out of 11 firms, again accounting for about 40 percent of employees in the park or approximately 6,260 workers. In Hawassa it was only able to definitively match 1 of 20 firms, which by itself accounted for around 11 percent of the park's employment in 2019. Because the list of firms operating in industrial parks records 2019 activity while the most recent year of the panel is 2016/2017, it is likely that some portion of the firms on the 2019 list began operations after the end of the period covered by the data; this is especially true for Bole Lemi and Hawassa, as they opened in 2016 and 2017 respectively.

To test whether the firms in industrial parks that the study could not match to the LMSM data affect the growth results, the analysis assigns each firm in the EIC list to a 2-digit ISIC category and to a firm-size group based on its current employment. The study then re-estimates the sector-level employment growth regressions with these firms’ employment included, assuming that each firm in the 2019 list was operational from the year its respective park opened and that its employment remained at the level reported in 2019 over time. These results can be found in table S2.2 in the supplementary online appendix. Even with the inclusion of the industrial park employment numbers (which mostly come from large firms), the sector-level growth estimates are largely unchanged.

Summary

In both Tanzania and Ethiopia large firms, exporters, and foreign firms all have significantly higher levels of labor productivity (and employment). These results are consistent with a large theoretical and empirical literature on manufacturing-firm performance.

Turning to employment and labor productivity growth, the firm and sector-level results for Tanzania suggest that the best-performing firms are not the ones that are absorbing employment. Labor productivity growth in large firms is on the order of 8 percent per year and 6 percent per year for large exporters. By contrast, labor productivity growth in firms with less than 50 employees is an imprecisely estimated zero, while employment growth in these small firms is as high as 8.6 percent per year at the sector-level.

In Ethiopia, it is found that employment growth in firms with between 10 and 49 workers is double that in large firms but labor productivity growth across the two firm sizes is roughly equal. The story becomes more similar to Tanzania once firms with less than 10 workers are considered. The analysis of firms with less than 10 employees is based on a group of mechanized firms with less than 10 employees using Ethiopia's Small Scale Industries survey. Average annual employment growth among these firms is around 17 percent while labor productivity is around zero, although both estimates are imprecise.

5. An Interpretation: Inappropriate Technologies

The low employment-generation capacity of modern, productive African firms presents a puzzle. It is not clear that it is possible to resolve the puzzle by appealing to conventional culprits. First, note that small firms are found to be less productive than large firms. Consistent with the arguments of Hsieh and Olken (2014), this suggests it is unlikely that credit or other constraints prevent small firms from expanding and growing into larger firms that employ more workers. In particular, the study follows Hsieh and Olken (2014) in analyzing the size distribution of firms and reaches similar conclusions. The study finds no evidence of a “missing middle” in the size distribution of firms in Tanzania and Ethiopia. The distribution of firm size is heavily right skewed, with a predominance of small firms and generally a smooth decline in frequency over the firm size distribution. There are no indications of a bimodal distribution; these results are available in fig. S1.6 in the supplementary online appendix. In addition, as in Hsieh and Olken (2014), the distributions of the average product of capital and labor are unimodal and do not show any discontinuity, as would be the case if labor costs (or access to finance/capital) were binding constraints on firm growth.

Second, high labor costs (relative to productivity) are often cited as constraints on employment growth in Africa (e.g., Gelb, Ramachandran, and Meyer 2020). But as table 5 shows, payroll shares in total value added in both Tanzania and Ethiopia are exceedingly low, even in the more labor-intensive sectors. Third, explanations that posit a “poor business environment” are belied by the high dynamism in Tanzania's and Ethiopia's manufacturing sectors, as captured by the analysis of entry and exit rates. As table 3 shows, entry rates during the two countries’ high-growth periods have been as high, if not higher, than the levels observed for Vietnam.

Fourth, institutional aspects of government business relationships are likely to play a role in the evolution of the manufacturing sectors in both Ethiopia and Tanzania, but it is unclear why these relationships would impact the employment growth but not the productivity growth of large firms. Moreover, as pointed out by Bourguignon and Wangwe (2018), corruption is not unique to Tanzania. Of course, corruption is also not unique to Ethiopia. For an extensive discussion of these issues in the Tanzanian context, see Bourguignon and Wangwe (2018) and Wangwe and Gray (2018), and in the Ethiopian context see Oqubay (2015).¹⁸

This section presents an alternative explanation. It argues that the broad patterns that are observed with respect to productivity and employment can be explained by capital-intensive modes of production in manufacturing that are excessive relative to the capital-labor endowment in the two African economies.

Capital-Labor Ratios in Tanzania and Ethiopia in Comparative Context

The Czech Republic was chosen as the comparator, a rich country, and a successful manufactures exporter.¹⁹ Some basic comparisons are provided in table 6, panel A. Per capita GDP in the African countries are a small fraction of the Czech level: between 3.5 and 6.6 percent. Aggregate economy-wide capital-labor ratios are similarly tiny: Ethiopia's K/L endowment is 3.7 percent of the Czech level, and Tanzania's 11.3 percent. But when looking at manufacturing specifically, the gaps become much smaller. Ethiopia is nearly at a fifth of the Czech economy and Tanzania's is close to one half. In other words, compared to the Czech economy, K/L ratios in Africa are 3–4.5 times larger in manufacturing than they are for the entire economy.

Since firm-level data are available for the African economies, it is possible to undertake a finer-grained comparison for specific firm types in panel B of table 6. Note in particular that the 10 percent most capital-intensive large firms, producing around a quarter of manufacturing value added (but employing less than 10 percent of the manufacturing workforce) have particularly high K/L ratios. In Ethiopia these large firms are at 81 percent of the K/L ratio for Czech manufacturing. Tanzania's large firms have K/L ratios that significantly exceed those for Czech manufacturing. Moreover, exporting firms are not more labor-intensive than the manufacturing average.²⁰ This may seem surprising, since one would expect exporting firms to compete on international markets in more labor-intensive segments of manufacturing. But it is consistent with the interpretation the analysis will develop below having to do with the adoption of more capital-intensive global technologies as a precondition for competitiveness. Also note that among large firms, new ones are considerably more capital-intensive than old ones.²¹ This too is consistent with the increased pressure over time to adopt more capital-intensive technologies.

Columns 2–4 of table 6 (panel B) provide some sectoral detail. The study focuses on three sectors: food, beverage, and tobacco products; and, rubber & plastics products; and stone, clay & glass products, which are the largest manufacturing sectors in both countries; and textile, apparel, and leather products (textiles thereafter) as the main “labor-intensive” sectors.²² Ethiopian firms are at about 20 percent of the Czech economy, and Tanzanian firms at around 40 percent.²³

However, textiles are heterogenous encompassing three sectors—textile, apparel, and leather and leather products (including footwear), and these three sectors operate at very different capital-labor ratios. While Czech data do not make it possible to disaggregate, the study can provide finer detail in the countries it studies and can compare them to Vietnam. This is done in fig. 7. In line with international evidence, apparel is generally much more labor intensive than the other sectors, especially compared to textiles, in all three countries. What stands out in the comparison is that Ethiopia and Tanzania have a much greater share of employment in relatively capital-intensive textiles, while Vietnam has been able to shift a significantly greater share of its workforce into apparel. In other words, the high capital-intensity of “textiles” in the African countries is due to the predominance of textiles over apparel (and to some extent leather products) in their employment structure.

It is also instructive to look at trends in manufacturing K/L ratios. Figure 8 shows that the levels of capital intensity in large manufacturing firms in both Tanzania and Ethiopia exceed their economy-wide averages. In Tanzania, the increase in capital-intensity in large manufacturing firms has far outstripped economy-wide capital deepening. By contrast, in the Czech Republic not only is capital intensity lower in manufacturing than in the economy as a whole, the two measures have also moved more or less in parallel in recent years. The study also introduces Vietnam as a comparator, and while its manufacturing K/L is higher than its total economy K/L, they are close and have moved in parallel as in the Czech Republic.

Another striking indicator of low levels of labor-intensity in African manufacturing is the payroll share in total value added. In the Czech economy, the payroll share in aggregate manufacturing is slightly below 40 percent and rises to 50 percent for textiles (UNIDO 2020). In Tanzania and Ethiopia, by contrast, the payroll share is in the range of 11–13 percent and rises to merely 20–24 percent in textiles. It is generally known that labor shares in value added are understated in developing countries because of the predominance of owner-operated firms. Even accounting for that downward bias, the low payroll shares in Tanzania and Ethiopia are striking.²⁴ This evidence on payroll shares also suggests that high labor costs—in relation to per capita incomes or level of productivity—cannot account for the capital intensity of the African firms.²⁵

In sum, the analysis draws four conclusions from this evidence. First, while K/L ratios in African manufacturing are lower than in much richer comparator nations, these ratios are still much higher than would be expected based on their relative labor abundance and low per-capita income levels. Second, when focusing on the largest firms, K/L ratios in Tanzania and Ethiopia are actually comparable to those in much richer OECD countries like Hungary and the Czech Republic. Third, exporting firms or the traditionally labor-intensive textiles firms do not exhibit lower K/L ratios than other manufacturing firms on average. Finally, K/L ratios have increased much more rapidly in Tanzanian and Ethiopian manufacturing than in the economy as a whole.

Implications: The Simple Analytics of Technology Choice

Consider a representative African firm with access initially to two kinds of technologies, a labor-intensive technology, and a capital-intensive technology. The firm operates in an open economy, where the price of output, p₀, is exogenous and determined on world markets. Figure 9 shows the unit costs of the two technologies and how they change with the scale of production. As drawn, production with the labor-intensive technology results in lower unit costs (than production with the capital-intensive technology) over the relevant range of output. This is a natural consequence of lower labor costs in the African country. Firms in the country would choose to employ the labor-intensive technology, and facing price p₀, produce an output q₀.

Note that the shapes of the two cost curves imply that costs rise more rapidly under the capital-intensive technology. This is also a likely consequence of economic conditions in African countries. Labor can be drawn from the countryside or from informal activities without a steep rise in wages. Capital, on the other hand is scarce. Equally important, it is possible to interpret capital more broadly here, as including other production inputs that are likely to be strong complements to capital in manufacturing—in particular, skilled labor and infrastructure. Those are also likely to be comparatively scarce in a low-income country, which would contribute to the steepness of the cost curve for the capital-intensive technology.

Now consider the implications of a significant improvement in technology that affects only the capital-intensive mode of production. This is an extreme form of biased technological change that will simplify the exposition and help make the argument. This example assumes that this innovation takes place in the advanced countries, but the resulting technological innovation is also available to the African country.

The innovation affects the diagram in two ways. First, the price on world markets falls as rich countries—using the capital-intensive technology—experience a reduction in costs, which is in turn passed on to world prices. This is shown in fig. 9 by a reduction from p₀ to p₁.

Second, the relevant cost curve for the capital-intensive technology shifts down. As drawn, the vertical shift of the cost curve is less than the fall in prices. This is because it is assumed that all the cost savings in rich countries are passed on (at least on impact) to prices and (realistically) that capital costs are higher in the African country so that the cost benefits of the new technology are lower than in the rich countries. These imply that the reduction in African costs (with the new technology) is less than the drop in world prices.

At the new level of prices, the labor-intensive technology is no longer cost-effective. The African firm will now shift to the capital-intensive technology. The result is a fall in its level of production from q₀ to q₁. Even with full and costless access to the new technology, the African firm is disadvantaged. Moreover, the adverse employment impact is even larger than the output effect, since production now takes place using a more capital-intensive technology.

Once firms have adopted the capital-intensive technology, the ability of the economy to increase manufacturing output (and generate employment) in response to new opportunities is reduced. That is because the cost curve of the new technology is steeper: due to scarcity of capital and complementary inputs (skills, infrastructure) any potential expansion of employment and output is choked off by rising costs. Not only is the comparative advantage of the economy in manufactures undermined, but its supply curve is also less responsive to higher prices (or lower costs).

This story is likely to apply especially to formal parts of manufacturing insofar as larger firms competing with imports or engaged in exports are the ones that face the pressure to upgrade to capital-intensive technologies. Smaller, informal firms serving sheltered home markets or operating in niche areas can survive while hanging on to older technologies. They will correspondingly be in a position to absorb employment to a much greater extent, albeit at stagnant or declining levels of productivity.

From a normative standpoint, the benefits of global technological advance are reduced—and possibly negated—in this economy by two considerations. First, to the extent that the country was previously an exporter of the good in question, a more capital-intensive technology produces a reduction in the gains from trade (worse terms of trade). Secondly, to the extent that there are wage premia associated with employment in the formal sector (a common form of misallocation in low-income countries, implying formal employment is too low) the reduction of employment in firms adopting the new technology is an additional source of efficiency loss. This is the sense in which one can consider the new technology “inappropriate” and its capital intensity “excessive.”

The study shows that firm-level employment growth is indeed negatively associated with capital intensity at the sector level, particularly for large firms. First a sector-level capital-labor ratio is defined as the total capital in a 4-digit ISIC subsector divided by the total employment. The analysis then regresses firm-level employment (measured in logs) on this industry-level measure of capital intensity (also measured in logs), with firm fixed effects. Controls for industry-level sales, industry, and year are also included. To adjust for any bias introduced by a firm accounting for a large share of capital and/or labor in a subsector, the study adjusts the measures of industry-level capital-labor ratios and industry-level sales by removing the own-firm contribution. The relationship is modeled by the following equation:

The results from equation 6 are reported in table 7. In Ethiopia, for large firms there is a negative relationship between the industry-level capital-labor ratio and firm-level growth in employment—a 1 percent increase in industry-level capital labor ratios correlates to a 7 percent decline in employment. In Tanzania, a 1 percent increase in industry-level capital-labor correlates to a 6 percent decline in firm-level employment. In Vietnam this relationship is even stronger—a 1 percent increase in the industry-level capital-labor ratio correlates to a 30 percent decline in employment for large firms and 11 percent for small firms.

6. Conclusion: Implications for Growth and Growth Policy in Low-Income Countries

The analysis of the manufacturing sectors in Ethiopia and Tanzania reveals a dichotomy between larger firms that exhibit superior productivity performance but do not expand employment much, and small firms that absorb employment but do not experience much productivity growth. Typically, economic development happens when the productively dynamic parts of the economy absorb resources from the rest. By contrast, the choice that African manufacturers seem to face is either to increase productivity or to increase employment.

It is unlikely that this pattern can be explained (only) by factor-price distortions or other institutional shortcomings specific to the African setting. This study's interpretation is that the technologies available on world markets restrict the range of production techniques that can be used by firms. As the capital- (and skill-) intensity of global technology has increased, the gap with low-income countries’ factor endowments has opened wide. Becoming more productive requires adopting technologies with factor input combinations that are increasingly at variance with African countries’ factor abundance.

From the standpoint of trade theory, this study's interpretation amounts to an argument that Ethiopia and Tanzania have been losing comparative advantage in traditionally labor-intensive manufactures due to a trend reduction in their labor intensity. This implies a loss in the gains from trade. It also lowers the ceiling on industrialization and constrains the capacity of manufacturing to absorb labor productively.

This is not to say that manufacturing cannot play an important role in the development of these countries. After all, productivity growth in the large manufacturing firms in Tanzania and Ethiopia has been impressive and could create jobs indirectly. For example, while the manufacturing of food products is capital intensive, smallholder farming is labor intensive. Worker training programs associated with industrialization strategies like Ethiopia's Technical and Vocational Education and Training School (TVET) could also enhance the capabilities of smaller firms. And the managerial and logistical capabilities of large manufacturing firms could be transferred to other activities through worker turnover or informal networks (Abebe et al. 2022). But tempering expectations is important especially in politically fragile countries like Ethiopia.

Data Availability

These data can be made publicly available and publishing them with the article is possible. As a pre-condition for receiving funding from the Private Enterprise Development in Low-income countries, Center for Economic Policy Research (CEPR), the study entered into two separate agreements for data sharing upon completion of the construction of the two firm-level panel datasets. The study as agreement letters with the Central Statistical Agency in Ethiopia and the National Bureau of Statistics in Tanzania (NBS). The datasets along with details about their construction are currently available from the authors upon request and will soon be published by CEPR and uploaded to the website of NBS.

Author Biography

Xinshen Diao is a senior research fellow at the International Food Policy Research Institute (IFPRI), Washington D.C., USA; her email address is [email protected]. Mia Ellis was a research analyst at the International Food Policy Research Institute (IFPRI), Washington D.C., USA. Margaret McMillan (corresponding author) is a professor at Tufts University, Medford, MA, USA; her email address is [email protected]. Dani Rodrik is a professor at John F. Kennedy School of Government, Harvard University, Cambridge, MA, USA; his email address is [email protected].

The research for this article was financed by Private Enterprise Development in Low-Income Countries (PEDL), the foreign, Commonwealth and Development Office (FCDO), and the Consultative Group on International Agricultural Research's (CGIAR) research program Policies, Institutions, and Markets (PIM) led by IFPRI. The authors thank the editor and two anonymous referees of this article. They also thank the National Bureau of Statistics in Tanzania and the Central Statistical Agency in Ethiopia for providing access to the data as well as invaluable support in creating and cleaning the panels, Inigo Verduzco for his work creating the Ethiopia panel identifiers for 1996–2013; Andualem Telaye Mengistu and his team at the Ethiopian Development Research Institute and Oxford University for providing access to their matching process for the Ethiopia panel for 2013–2017; Brian McCaig for information about Vietnam; Gaaitzen de Vries and Emmanuel Mensah for their feedback on total manufacturing employment estimates; and Marco Sanfilippo for helpful discussions. Alison Campion and Feiruz Mohammed provided excellent research assistance. A supplementary online appendix is available with this article at the World Bank Economic Review website. The authors have no financial or proprietary interests in any material discussed in this article.

Footnotes

References