VoxEU Column Labour Markets Productivity and Innovation

Technological progress and hollowing-out of the middle-skilled labour share of income

The declining labour share of income is a global phenomenon that has affected primarily low-skilled and middle-skilled workers. This column examines the effects of trade and technology on the labour shares of different skill groups using a new dataset covering both advanced and developing economies. Both trade and technology have contributed to the declining labour share of middle-skilled workers but have had little effect on low-skilled and high-skilled labour. Policies should be designed with the goal of spreading the benefits of globalisation to the entire labour force.

One of the most remarkable facts about the decline in the global labour share of income is that it is widespread across economies in different regions of the world, of different income levels, and with varying exposure to international trade and technological diffusion. This global development has been noted and discussed in a number of recent papers (e.g. Elsby et al. 2013, Karabarbounis and Neiman 2014, Autor et al. 2017). 

Less known, but equally remarkable, is that the decline in the global labour share has been borne disproportionately by low-skilled and middle-skilled labour. Between 1995 and 2009, their combined labour income share fell by more than seven percentage points, while the high-skilled labour share rose by more than five percentage points (Figure 1). This pattern emerges in both advanced and emerging economies. Particularly striking is that the decline in the middle-skilled share of labour income was driven primarily by a drop in their relative wage rate, while their share of employment in the total workforce remained stable or even rose. The decline in the labour share of low-skilled labour and the increase for high-skilled labour were largely driven by their diverging trends in employment composition, reflecting rising levels of education. 

Figure 1 Labour share evolutions and labour force composition by skill level (in percent)

Sources: World Input-Output Database; and IMF staff calculations.
Note: AEs = advanced economies; EMDEs = emerging market and developing economies

The diverse evolution of the labour share of income by skill group has important implications for income inequality and social cohesion. However, the forces behind these within-skill trends have received much less attention. Existing research, focused predominantly on the US and Europe, has highlighted that technological progress is biased in favour of high-skilled labour (Karabarbounis and Neiman 2014) and pointed to persistent losses of jobs in middle-skilled occupations, resulting in wage and employment polarisation (Autor and Dorn 2013, Goos et al. 2014, Pierce and Schott 2016).

In a recent paper, we examine the effects of trade and technology on the labour shares of different skill groups (Dao et al. 2019). We assemble a new data set based on primary sources from national authorities, expanding sources beyond the OECD and the data set of Karabarbounis and Neiman (2014), and include a much larger set of emerging and developing economies. The analysis is based on countries with at least ten years of data on labour shares over the 1991–2014 period, resulting in a sample of 27 advanced economies and ten emerging economies on labour shares by skill groups. Low skilled refers to workers with primary and lower secondary education, middle skilled to those with upper secondary or postsecondary, non-tertiary education, and high skilled to those with first-stage tertiary education or higher.

We examine the distributive effects of technology and trade, including whether these have contributed to polarisation and the so-called ‘hollowing out’ of the middle class in advanced economies by looking at the drivers of the labour shares of high-skilled, middle-skilled, and low-skilled workers separately. In order to get a more nuanced picture of the effects of technological change, we introduce measures of the exposure to routinisation to assess whether the declining price of investment goods has led to a greater decline in labour shares in more exposed countries and industries. To measure routine exposures, we use employment-weighted scores of occupations in the economy and across industries (for details, see Das and Hilgenstock 2018). 

Figure 2 Contribution to aggregate labour share change by skill, 1995-2008

Sources: World Input-Output database and authors' calculations. 
Notes: Decomposition derived from aggregate labor share regressions by skill groups. Data for labor share by skill from WIOD. Contribution of "Skill supply and other shifts in composition" is combined effect of educational composition and regression constant term.

Our results suggest that both technological advancement and participation in global value chains have lowered the income share of middle-skilled workers but have had little discernible effect on those of low-skilled or high-skilled workers. New to the literature, we find strong evidence that countries with higher exposure to routinisation and greater increase in participation in global value chains have experienced stronger declines in the middle-skilled labour income share, which has been especially pronounced in Austria, Germany, and the US (Figure 2). 

This finding corroborates existing evidence that automation and import competition and offshoring have led to long-term losses in middle-skill occupations and displacement of middle-skilled workers to lower-wage occupations in advanced economies.

What does this imply for policies? Our findings, together with existing evidence that technology and globalisation have contributed to job polarisation and long-term unemployment for some groups of workers present challenges to policymakers to find new ways to keep on harnessing the benefits of globalisation and the rapid advancements in technology, while spreading those benefits more widely to all groups of workers.

In general, policies in advanced economies should be designed to help workers better cope with disruptions caused by technological progress and global integration, including through skill upgrading. Some of the required policies will have transitory effects in nature, such as ‘trampoline policies’ needed to facilitate the reallocation of displaced workers to new jobs. These include social safety nets such as unemployment benefits, as well as strong job search support, retraining programs, and well-designed temporary subsidies. But to the extent that some workers are permanently affected by the growth-enhancing forces of technology and trade, as suggested by this skill-level analysis, longer-term redistributive measures might be required as well, including redesigning tax and benefit policies by introducing or scaling up negative income taxes or universal basic incomes. With the prospect of an AI-led economy at hand, redesigning education and training for a future AI-led economy will also be needed.


Autor, D H and D Dorn (2013), “The Growth of Low-Skill Service Jobs and the Polarization of the US Labour Market”, American Economic Review 103(5): 1553–97.

Autor, D H, D Dorn, L F Katz, C Patterson and J Van Reenen (2017), “Concentrating on the Fall of the Labour Share”, NBER Working Paper 23108.

Dao, M C, M Das and Z Koczan (2019), “Why Is Labour Receiving a Smaller Share of Global Income?”, Economic Policy 34(100): 723–759.

Das, M and B Hilgenstock (2018), “The Exposure to Routinization: Labour Market Implications in Developed and Developing Economies”, IMF Working Paper.

Elsby, M W, B Hobijn and A Şahin (2013), “The Decline of the US Labour Share”, Brookings Papers on Economic Activity (2): 1–63.

Goos, M, A Manning and A Salomons (2014), “Explaining Job Polarization: Routine-Biased Technological Change and Offshoring”, American Economic Review 104(8): 2509–26.

Karabarbounis, L and B Neiman (2014), “The Global Decline of the Labour Share”, Quarterly Journal of Economics 129(1): 61–103.

Pierce, J and P Schott (2016), “The Surprisingly Swift Decline of U.S. Manufacturing Employment”, American Economic Review 106: 1632–62.

2,310 Reads