Industrial Development

The previous two sections have summarised evidence on past structural transformation experiences. In this section, we review a set of factors that are shaping current and future paths of industrialisation, and that may make them different from the experiences of advanced economies. These include differences in the global technological environment and other explanations of premature deindustrialisation, growing protectionism, automation, and the interaction between industrialisation and the size structure of firms.

Tristan Reed's Slides, Gustavo de Souza's Slides, Markus Poschke's Slides.

Premature Deindustrialisation

In Section 2, we discussed the differences between the industrialisation paths of late and early industrialisers. Late industrialisers tend to reach lower manufacturing employment shares at the peak of their industrial hump, and these peaks occur at lower levels of income per capita – a phenomenon commonly referred to as premature deindustrialisation. Moreover, for late starters, increases in industrial value added are not accompanied by proportional gains in jobs, which makes industrialisation jobless.

Rodrik (2016) argues that more rapid productivity growth in manufacturing than in the rest of the economy, in combination with trade and globalisation, can explain premature deindustrialisation. As developing countries liberalised trade, those lacking a strong manufacturing advantage faced import competition and falling global manufacturing prices, driven by growth in manufacturing productivity in advanced economies. They thus imported deindustrialisation from advanced economies, limiting their scope for industrial employment, and resulting in premature deindustrialisation. While Rodrik (2016) posits that international trade may be an important contributing factor, he leaves the question open, as addressing it requires quantitative analysis.

Sposi et al. (2025) analyse the role of trade in premature deindustrialisation using a dynamic open-economy model with differences in productivity growth across sectors. They show that these differences lead to premature deindustrialisation, which is further amplified by trade. Fundamentally, premature deindustrialisation reflects faster technological progress in manufacturing than in services. International trade magnifies this process by transmitting global technological change through relative prices and accelerating the decline in the relative price of manufacturing. At the same time, declining trade costs reveal underlying comparative advantages and reinforce specialisation. Over time, this results in fewer countries reaching high manufacturing peaks, as a small set of exporters supply goods to the rest of the world. Importantly, their framework also captures another feature that they document in the data, industry polarisation: the variance in the manufacturing share of value-added across countries has increased over time.

Huneeus and Rogerson (2023) quantitatively explore sector-biased productivity growth in a closed-economy model, highlighting the importance of heterogeneous agricultural productivity paths across countries. Their model incorporates the theoretical mechanisms discussed in Section 2.3. Then, productivity growth in agriculture leads to flow of workers out of agriculture and into manufacturing, and productivity growth in non-agriculture leads to flows of workers out of manufacturing and into services. Jointly, empirically reasonable sectoral productivity growth patterns generate hump-shaped manufacturing employment shares. Their analysis ascribes premature deindustrialisation to relatively sluggish agricultural productivity growth in late starters. As a result, the labour released from agriculture is more likely to move into services rather than manufacturing in these economies, implying that they reach lower peak manufacturing shares, at lower levels of income per capita.

Fujiwara and Matsuyama (2024) attribute premature deindustrialisation to heterogeneous technology gaps across sectors and countries. When technology gaps generate longer adoption lags in services than in agriculture, while agriculture still experiences faster productivity growth than other sectors, poorer countries fall further behind in agriculture than in services. As a result, they reach their manufacturing peaks later in time than richer economies, but at lower productivity levels – hence prematurely. Moreover, if adoption lags in manufacturing are modest, late industrialisers also exhibit lower peak manufacturing shares than early industrialisers. Under these conditions, the model captures three defining features of premature deindustrialisation: later peaks in time, earlier peaks in income per capita, and lower maximum manufacturing shares.

The Protectionist Threat

In Section 3, we discussed the role of international trade for industrial development. In fact, openness to trade – including imports of modern inputs and exports of goods and services – has been a pathway to development for many industries, not just manufacturing. At the macroeconomic level, Spence (2008) identified 13 national success stories, i.e. economies that since 1950 have grown at an average rate of 7% a year or more for 25 years or longer. These successes include manufacturing exporters like the East Asian Tigers, commodity exporters like Botswana and Oman, and services exporters like Malta. Exports in one sector can stimulate growth in others, as beef exports boosted services in Uruguay (Amodio et al. 2025a). Static models of gains-from-trade alone cannot explain these growth stories (Costinot and Rodríguez-Clare 2014). Dynamic models are needed, in which openness enables adoption of productive technologies. One such model is that openness provides firms with a larger market size, allowing economies of scale (Goldberg and Reed 2023a).

Though global trade still grows in real terms, recent policy initiatives raise questions about its future (Goldberg and Reed 2023b). On the demand side, some now question whether high-income countries should buy exports from developing countries. This view links foreign trade surpluses to domestic trade deficits and decline (Klein and Pettis 2020, Obstfeld 2024). In the US and UK, growing inequality has coincided with calls for protection, especially in import-competing areas (Bonomi et al. 2021, Choi et al. 2024). This popular backlash, however, was not an inevitable outcome of globalisation. In Germany and France, for instance, inequality has not increased (Milanovic 2016)

On the supply side, industrial policies in low- and middle-income countries restrict exports of industrial raw materials and some food crops, motivated by security or in order to lower domestic commodity costs (OECD 2025). These policies threaten access to intermediate inputs crucial for exports (Amiti and Konings 2007, Fernandes 2007), though in the medium-term there may be adaptation and substitution away from these inputs (Alfaro et al. 2025). Export restrictions on advanced technology, despite the attention they receive, are much more limited. A January 2025 US rule limiting computer chip exports to 150+ countries was rescinded, now focusing mainly on one country.

Three facts clarify the threat of protectionism for industrial development, with different impacts across agriculture, industry, and services.

Protectionism is rising, but only in a few markets

Though the US has now raised tariffs on many countries, its global import share is just 15% today, down from 20% in the 2000s. Neighbours like Canada, Mexico and Haiti are highly dependent, but it counts for less than a fifth of most other countries' exports. This suggests existing relationships can be used to develop other markets rather than reducing total exports, especially as other economies grow.

Another trade barrier has been in the EU, which now protects against trade in goods linked to deforestation or carbon intensity. Yet these policies are limited to a small number of commodity products: the Carbon Border Adjustment Mechanism targets just six (i.e. iron and steel, aluminium, cement, fertiliser, hydrogen, and electricity) and the Deforestation Regulation targets just seven (i.e. cattle, cocoa, coffee, oil palm, rubber, soya, and wood). These policies do not affect most export opportunities in manufacturing or services, suggesting a limited impact outside commodity sectors.

With declining demand from the US and EU, the biggest export growth opportunity is in middle-income countries that are growing faster than high-income countries, or so-called South-South trade. This story played out in the 2001–2014 period, as commodity demand in East Asia drove rapid growth in commodity exporters, especially in Africa, though this did not drive industrial development in Africa, potentially due to competition from East Asia itself in value-added manufacturing. Yet as East Asian countries' costs rise and move into still higher-value exports, opportunities could open for other regions, as has happened in textiles and apparel. New regional trade agreements like the Africa Continental Free Trade Agreement and the Regional Comprehensive Economic Partnership, which lower trade barriers but do not materially discipline industrial policies, create more opportunities to experiment with new exports in local markets.

A challenge for South-South trade is that many Southern countries export similar goods, limiting trade gains. Heterogeneous growth paths, coupled with the development of new industries across Southern economies, are needed to generate distinct comparative advantages and reduce reliance on the North.

Protectionism against China creates opportunities for other countries

China's rise as a competitive exporter has been a major theme of recent decades. This rise has been driven by a combination of savings, openness, and industrial policy, with subsidies that are phased out after industries mature (Barwick et al. 2025, Fang et al. 2025). Mature industries remain competitive given domestic scale, supply chains, and research and development.

Protectionism focused on China specifically opens space for bystander countries with fewer advantages. Evidence shows that other countries' exports often substitute Chinese imports rather than complement them (Fajgelbaum et al. 2024). Consistently, the late 2010s US-China tariff war coincided with trade growth and increased exports from Mexico and Vietnam to the US, both in terms of locally manufactured substitute goods and transshipments (Alfaro and Chor 2023, Iyoha et al. 2024). McKinsey Global Institute (2025) finds that European exporters are best positioned to fill gaps, though these opportunities could be exploited by other regions.

Export restrictions on commodities in lower-income countries

Protectionism in low- and middle-income countries mainly uses export restrictions, a second-best industrial policy intended to promote learning-by-doing in commodity-using industries by lowering their input costs. Countries with limited fiscal space may prefer this over direct subsidies, the first best tool to promote learning-by-doing. Since industrial development depends on many factors other than input costs, including access to markets, human capital, and regulation, there is good reason for scepticism that commodity export bans alone can promote development. Regardless, countries are increasingly experimenting.

For example, Indonesia's 2014 nickel ore export moratorium triggered foreign investment in processing, raising processed nickel exports that are inputs to steel and batteries. Early evidence shows increased domestic value added and downstream entry as a result, though many entrants are small and potentially unproductive, and there remains dependence on imported steel (Kee and Xie 2025). Further study is needed, and perhaps more time, to identify whether there has been an aggregate impact on productivity, wages, and employment.

Policy Implications

Trade shocks stress emerging economies but are not insurmountable. Opportunities to grow agribusiness, industry, and services exports continue to exist. Further, as middle classes grow, they can serve as an alternative source of demand (Goldberg and Reed 2023a).

The export promotion toolkit, which can be implemented cheaply and does not violate international trade rules, remains highly relevant as countries develop new markets (Reed 2024). Experimentation with market interventions like subsidies, export restrictions, and local content requirements, could theoretically yield gains if they address market failures, but more research is needed to test this hypothesis.

Automation and Jobless Industrialisation

Industry in advanced economies has been transformed over the last decades by mechanisation and the adoption of robots. Industrial robot adoption remains much lower among firms in low- and middle-income countries than in rich countries, but adoption rates are rising. The diffusion of industrial robots raises the possibility that productivity gains from industrialisation may no longer translate into broad-based job creation, reinforcing concerns about ‘jobless industrialisation’. Understanding whether automation primarily destroys jobs, reshapes the task composition of work, or alters countries’ integration into global value chains is therefore crucial for assessing its implications for structural transformation and inequality.

This section reviews the emerging evidence on automation in low- and middle-income countries, focusing on three interrelated questions. First, how does robot adoption affect employment, wages, and occupational structure in developing economies, both directly and through trade and production networks? Second, what economic forces are driving the recent acceleration in automation, and how do these forces differ from those emphasised in advanced economies? Third, to what extent are industrial robots an appropriate technology for labour-abundant countries, and what frictions limit the realisation of productivity gains? The discussion concludes by drawing out the policy implications of these findings.

Automation in the Developing World

The effects of robot adoption in emerging markets are similar to those identified in rich countries. Rather than creating massive unemployment, robots have induced a reallocation of workers across occupations and sectors. Evidence also shows that the impacts of industrial robots can reach countries with little or no adoption through global value chains and trade links.

In 2020, high-income countries accounted for roughly 60% of the global stock of industrial robots, China for about 30%, and other low- and middle-income countries for the remaining 10%. In advanced economies, the stock of industrial robots more than doubled between 2000 and 2020, increasing by approximately 125% (IFR 2020). Evidence from these economies indicates that robot adoption tends to displace low-skill, routine workers, while its effects on aggregate employment are more mixed. In the US, Acemoglu and Restrepo (2020) estimate the negative impacts of robot exposure on employment and wages in commuting zones, concentrated among routine jobs with lower skills. The cross-country evidence in Graetz and Michaels (2018) similarly points to displacement of lower-skill workers, even as robots contribute meaningfully to productivity growth – illustrating the trade-off between efficiency gains and distributional costs.

In emerging countries – excluding China – the diffusion of industrial robots has accelerated notably since the late 2000s, with their share of the global robot stock rising from 2% in 2000 to 10% in 2020. Despite this growth, the predominant labour market effect appears to be reallocation of workers across occupations rather than rising unemployment, and in certain cases even employment gains. In Thailand, Jongwanich et al. (2022) find shifts in employment between skilled and unskilled jobs. For Brazil, Rodrigo (2022) shows that automation induced a reallocation of workers across occupations, moving from production to support activities even within the same firms. de Souza and Li (2025) highlight that these effects are concentrated among low-skill workers. For Mexico, Faber (2020) documents no statistically significant effect of domestic robots, but a robust negative effect of automation in an advanced-economy trade partner (notably the US) that transmits through the global value chain, depressing employment and relative wages in more integrated Mexican sectors and regions.

China stands out among emerging economies for its dramatic rise in robot adoption. Its share of the global industrial robot stock increased from less than 1% in 2000 to 30% in 2020, catching up with advanced economies in terms of robot density – i.e. the number of robots per 10,000 employees (IFR 2020). Giuntella et al. (2024) document that this expansion was accompanied by reductions in employment and hourly earnings, concentrated among the less educated. The effect was larger for male, prime-age, and older workers.

Still, by 2020, nearly 140 developing countries had a robot density below one robot per 10,000 employees, 84% of them with no record of installed robots (IFR 2020), partly because cheap labour discourages investments in labour-saving technology (see Section 4.3.3). While these nations might be exempt from the direct effects of robots, they might still be indirectly exposed to the effects of automation technology through changes in trade patterns with more technologically advanced trade partners (Krenz et al. 2021, Cilekoglu et al. 2024). 

Kugler et al. (2020) and Calì and Presidente (2025) analyse the effects of automation in two developing countries – Colombia and Indonesia – where robot adoption has been minimal. Kugler et al. (2020) show that Colombian workers in industries more exposed to US robot adoption experience lower cumulative earnings and longer job tenures. They further find that workers who switch firms within the same industry largely avoid earnings losses, whereas those who change industries or locations suffer the largest income declines. In contrast, Calì and Presidente (2025) document positive labour market and plant-level employment effects from robot adoption in Indonesia. They attribute these gains to industrial automation having diminishing productivity returns, suggesting that robots can be employment-enhancing at early stages of industrialisation when untapped automation possibilities abound and the productivity gains from adoption are the largest.

Drivers of Automation

In this section, we examine three forces driving the surge in robot adoption: falling robot prices, rising wages of low-skilled workers, and population aging. Together, these forces push firms to automate tasks once done by workers by making capital cheaper relative to labour.

Falling prices have been the main driver of industrial robot adoption. Over the past 20 years, robots of all types have become more affordable. Adachi et al. (2024), tracking unit prices in Japan, show that, since the 1980s, welding robots have become 60% cheaper, while robots in other applications cost from 20% to 40% less. Because Japan is one of the world's largest exporters, these price declines have fuelled adoption worldwide. Adachi (2025) shows that cheaper robots in Japan lead to higher adoption of them in the US, causing lower employment. Ayyagari et al. (2025) shows that the sudden yen depreciation in the first half of 2010s also led to an increase in robot adoption in the US.

These studies find that robot adoption is highly elastic to robot prices. Adachi et al. (2024) finds an elasticity of 1.5% for Japan, while de Souza and Li (2025) estimate an elasticity of 7% for Brazil. In the US, Ayyagari et al. (2025) find that a 10% yen depreciation raises the probability of robot adoption by 2.3 percentage points.

Across several studies, industrial robots have been shown to replace low-skilled workers. Firms face a trade-off: they can produce a given task using either robots or low-skilled production workers. Therefore, if the wage of low-skilled workers increases, firms may increase their use of robots. Dechezleprêtre et al. (2025) show that an increase in the wage of low-skilled workers increases automation by up to 5%. They also show that this effect has implications for policies. The German Hartz labour market reforms, which tightened unemployment benefits and expanded low-wage employment opportunities, increased labour force participation among low-skilled workers and led to a decrease in automation. In India, using granular firm-level data on machinery and computer investment, Gauthier (2025) shows that firms intensive in routine tasks invest more in machinery and computers following minimum wage hikes.

Population aging is another key force behind automation. Japan, one of the main producers of industrial robots, illustrates this phenomenon well. Between 1980 and 2000, the share of Japanese aged over 65 rose from 9% to 17%. In the same period, robot adoption jumped from near zero to five per thousand workers – the highest rate worldwide (Adachi et al. 2024). The surge in retirees created a shortage of production workers, which in turn led to robot adoption (Deng et al. 2023).

Acemoglu and Restrepo (2021) shows that the link between aging and robot adoption extends well beyond Japan. In a cross-country panel, they show that countries that aged faster also adopted more robots, produced more automation-related patents, and exported more automation technologies. Their model explains the rationale behind this pattern: as aging makes middle-aged workers relatively scarce, their wages rise, prompting firms to substitute costly labour with industrial robots.

This substitution of middle-aged workers with robots has important implications for growth. As countries age, labour force participation and worker productivity decrease. Yet, as Acemoglu and Restrepo (2017) shows, robots offset these losses by replacing scarce middle-aged workers, allowing economies to keep growing despite shrinking workforces.

These forces also carry important implications for developing countries: robot adoption there is likely to rise. Like advanced economies, developing countries face aging populations, and as they grow, the wages of low-skilled workers tend to increase. Combined with the steady decline in robot prices from technological progress, these trends point to a future where developing countries adopt industrial robots at accelerating rates.

Industrial Robots and Appropriate Technology

In developing countries, low-skilled labour is abundant while capital is scarce. The adoption of robots in these countries creates a mismatch: industrial robots rely on the scarce factor (capital) and replace the abundant one (labour). This mismatch has long been recognised in growth theory. When technologies are designed in advanced economies to reflect their factor endowments, they may be ill suited for developing economies, where labour-intensive technologies would be more productive (Acemoglu and Zilibotti 2001, Basu and Weil 1998). Because R&D is concentrated in advanced economies, developing countries often end up adopting technologies that are not well aligned with their relative factor abundance, which depresses the productivity benefits they can reap.

This tension plays out at the firm level as well. When firms adopt a new technology such as robots, they face a skills and organisational mismatch that must be resolved before productivity gains can materialise. The adoption of robots depreciates the existing organisational capital, forcing companies to invest in training and reorganisation until finding the new skill mix to effectively operate the new technology. Rodrigo (2022) provides evidence from Brazil illustrating this process: robot adoption triggered major labour reallocation within firms, shifting workers from production to support roles. Such reorganisation was costly and gradual enough to explain the slow emergence of productivity gains. The study also shows that this process is slower in labour markets with high firing costs.

If capital-intensive technologies such as robots are often inappropriate for developing countries, why do firms adopt them? de Souza (2022) develops a model that answers this question. In his framework, firms face a trade-off between two characteristics of a technology: its factor bias and its productivity level. Consider industrial robots. Their capital-intensive nature makes them poorly aligned with the input mix typical of developing economies, which discourages adoption. Yet, if robots have sufficiently large productivity, firms may still choose to adopt them despite the mismatch. The result, at the aggregate level, is that robots raise productivity but also shift factor demand, reducing the need for low-skilled labour while increasing the return to capital. Firms in developing countries thus become more productive than they would be without these technologies, but remain less productive than firms in advanced economies where the technologies are truly appropriate.

Policy Implications

If the realisation of the productivity gains associated to automation technology depends on firms' capacity to adjust their organisation, then subsidies for retraining, providing managerial capacities, as well as unemployment benefits and adjusting the labour regulation towards more flexible systems that facilitate labour mobility, might be more effective policies than tax cuts or other type of incentives for technology adoption.

As for the problem of appropriate technology, there are two possible policy solutions. One is to subsidise the development of local technologies. However, recent evidence suggests that such programmes often fail to address the problem because firms in developing countries tend to use these subsidies to imitate foreign technologies (de Souza 2023, König et al. 2022). The other possible policy solution, emphasised by de Souza and Li (2025), is to subsidise technologies that complement labour rather than replace it. In a model calibrated to Brazil, they distinguish between labour-replacing technologies, such as robots, and labour-complementing technologies, such as tools. Their results show that subsidising tools raises welfare by promoting redistribution.

Firm Size Distribution

Industrialisation also goes along with changes in the size of firms in an economy. The distribution of firm sizes and the organisation of production differ markedly between rich and poor countries. The average size of firms is much higher in richer countries (Poschke 2018, Bento and Restuccia 2017, 2021) and, by reflection, the share of self-employed workers is much lower (Gollin 2008, Donovan et al. 2023). In the US, more than 80% of employment is in firms with 10 or more employees, while it is only 15% in low-income countries (Gottlieb et al. 2025). Because larger firms pay higher wages, this also affects inequality. Recent evidence shows that workers’ individual income and business size are more tightly correlated in less developed countries compared with advanced economies, even conditional on individual characteristics and sector of employment (Eslava et al. 2023).

The most recent findings from Ethiopia and Tanzania highlight a sharp dichotomy between large industrial firms that exhibit strong productivity performance but generate few jobs, and small manufacturing firms that absorb labour without corresponding productivity gains (Diao et al. 2025). The empirical likelihood that informal microenterprises become larger and pay higher wages is also low (Eslava et al. 2024). A possible explanation for these findings is that larger firms nowadays use less labour-intensive production technology than in the past, while smaller firms produce local non-tradable goods – e.g. food stands, customised garment making – whose demand increases with local income. This implies, however, a tighter link between the number of jobs at modern industrial firms and employment at smaller informal firms. Indeed, recent evidence from Ethiopia shows that when the number of industrial jobs increases, employment at small informal firms also rises, challenging the dichotomous view presented above (Amodio et al. 2025b).

The literature seeking to explain the relative scarcity of large firms in poor countries points to a range of frictions and distortions, including entry costs (Moscoso Boedo and Mukoyama 2012, Poschke 2010), labour market regulation (Hopenhayn and Rogerson 1993, Poschke 2009, Ulyssea 2010), financial frictions (Buera et al. 2011, Midrigan and Xu 2014), delegation frictions (Akcigit et al. 2021, Grobovšek 2020, Guner et al. 2018), and generic wedges (Bartelsman et al. 2013, Hsieh and Klenow 2009, Restuccia and Rogerson 2008). Others have argued that small firm sizes may be an optimal reaction to a different environment, for example in terms of the level of capital (Gollin 2008) or of technology (Poschke 2018). High levels of churn in labour markets in poor countries also affect employers’ incentives to create formal jobs (Donovan et al. 2023, Poschke 2025).

More recent literature highlights the role of labour market structure and limited competition in shaping employment outcomes (e.g. Amodio and de Roux 2024, Felix 2022, Sharma 2024). When industrial jobs are concentrated among a few dominant firms, these firms can internalise the aggregate impact of their own labour demand, reducing job opportunities, suppressing wages, and offering worse working conditions relative to a perfectly competitive benchmark. This helps explain why self-employment remains relatively attractive to unskilled workers in poor countries (Blattman and Dercon 2018). At the same time, the availability of self-employment as a readily accessible outside option increases workers’ responsiveness to wage changes and can mitigate the wage-setting power of large firms. By the same token, however, policies that seek to expand wage employment by reducing reliance on self-employment may inadvertently strengthen employer market power, making such policies less effective than intended (Amodio et al. 2025c, Amodio et al. 2025d).

A crucial aspect of employment at large firms is that it also entails higher skills and white-collar occupations such as managers, accountants, purchasing agents, and clerks. For this reason, recent papers have focused on skill supply or lack thereof in explaining differences in the firm size distribution and development across countries. Engbom et al. (2025) show that differences in aggregate skills entirely account for the lower share of white-collar workers in poorer vs richer countries. In addition, Gottlieb et al. (2025) show that the skill intensity of larger firms is higher in poorer countries, where the skill premium and cost of middle management are relatively higher (Hjort et al. 2022). Cox (2025) shows that the 1996 higher education reform in Brazil that deregulated private colleges increased the prevalence and employment share of large firms. These studies suggest that education policies that expand the supply of skills can be powerful tools for promoting firm expansion in low-income countries.

Finally, some recent studies prompt us to reconsider the very same way we define the boundaries of firms and the separation between capital and labour within firms in developing countries. Surveying manufacturing firms in Uganda, Bassi et al. (2022) find that small firms in informal clusters are connected through an active rental market for machines that allows them to mechanise production. Crucially, when moving from a management-based (workers under the same management), to a machine-based (workers using the same machine) definition, the share of firms with more than 10 employees increases from 5% to 33%. At the same time, however, Bassi et al. (2023) show that somewhat larger firms in this context lack vertical specialisation between employees and entrepreneurs, and are little more than a collection of self-employed individuals sharing the same space. To conclude, McCasland et al. (2024) provide experimental evidence that workers supply both labour and capital to their employers, with important implications for the measurement of labour and capital compensation and returns, and their dispersions across firms.

For full reference list see the end of the final chapter.