Abstract

This paper analyzes cross-country firm-level data from 11 European countries and the USA between 1995-2017 to uncover the effect of technological innovations on increasing labor market polarization within industries in advanced economies. It contributes to the existing literature by making an extension on the theoretical framework of Acemoglu & Restrepo (2018) and by making use of a proxy of Total Factor Productivity (TFP) to represent new technologies. Findings would suggest that technological innovations do contribute to increased polarization in the selected time-period. The share of high-skilled workers increases, presumably due to the replacement effect. This occurs at the expense of middle-skilled workers who are more likely to be replaced by automated technologies, caused by the displacement effect. Low-skilled workers do not experience significant changes. Nevertheless, the findings are different for specific countries and industries and require consideration of other explanatory variables such as offshoring and import competition from China.

2.   Literature overview

The purpose of this chapter is to elaborate upon the existing academic literature on the topic of new technologies and the labor market. The market polarization hypothesis and the supporting empirical evidence is shortly discussed. Subsequently other factors leading to market polarization will be considered.

2.1.         Market polarization hypothesis

Throughout the past decades much attention has been paid to the study of skills and technological change. A common understanding is that Skill-Biased Technical Change (SBTC) is associated with labor share displacement and increasing wage inequality. Tinbergen’s (1974) approach assumes that technological innovations are factor-augmenting. They increase the productivity of skilled workers more compared to the productivity of unskilled workers. Simple tasks performed by unskilled workers are more likely to be replaced by new technologies. Therefore, there might be a displacement of occupations that have routine-based tasks. Goldin & Katz (2008) continued to expand on this framework. Labor-using tasks are replaced with capital due to recent technological innovations. Acemoglu & Restrepo (2018) however argue that besides a displacement effect of new technologies, which entails capital substituting the labor share, there also is a replacement effect prevalent. New tasks and jobs arise due to the introduced technologies, which potentially reinstate the labor share.

Labor economists have predicted different effects of innovations on the composition of the labor market. On the one hand there is a considerable share of supporters of the diffusion hypothesis. They assume that unskilled workers, or in this respect low-skilled workers, are hit hardest by the introduction of new technologies, since their tasks are easily replaceable. The implementation of new technologies has a factor augmenting nature and therefore complements high-skilled workers (Acemoglu & Autor, 2011).

On the other hand, there is the market polarization hypothesis, which expects that rather than low-skilled workers, middle-skilled workers will endure a drop in their labor share due to new technologies (Spiezia, Polder & Presidente, 2016). This paper limits itself to consider only the market polarization hypothesis, since findings of this hypothesis upon this point have been mixed and therefore require further research.

In general, middle-skilled workers are overrepresented in jobs performing routine-based tasks, for example being a bank clerk. High-skilled occupations such as consultants or physicians instead are more likely to perform non-routine tasks (Dao, Das, Koczan & Lian 2017). Traditionally, high-skilled occupations specialize in more complex tasks and are therefore supposedly safeguarded against automation and possibly even perform new tasks associated with technological innovations. Their jobs overall require more analytical or social skills, which are less likely to be displaced (Acemoglu & Restrepo, 2018). Although part of low-skilled occupations is routine-based, they are also frequently employed in occupations performing non-routine tasks such as a cleaning-lady, a hairdresser or a cab driver. Subsequently, due to the nature of the performed tasks, middle-skilled jobs are more susceptible to be replaced by automated and innovative technologies compared to high and low-skilled workers, potentially explaining increased polarization (Dao, Das, Koczan & Lian 2017). Nevertheless, recent technologies such as AI cause risk of replacement of high-skilled tasks as well. For example, consider accounting, financial planning or even surgery. Automation of high-skilled jobs is on the rise, which can be considered to be a counterargument to the market polarization hypothesis (Michaels, Natraj & Van Reenen, 2014).  

2.2.         Empirical evidence on the effect of new technologies on labor market polarization

Evidence of the market polarization hypothesis in the academic literature show different results. The study of Gregory, Salomons & Zierahn (2019) for example identify both labor-replacing and labor-augmenting forces. Results of their study show that between 1999-2010 across 27 European countries there indeed has been strong labor displacing effects. However, the countervailing forces of new jobs that were created through increased product demand outweigh these displacing effects. In addition, Michaels, Natraj & Van Reenen (2018) for example find a positive link between ICT usage and labor market polarization in the European Union between 1980 to 2004. In their study they argue middle-skilled workers tend to have more cognitive routine tasks while performing their jobs. Less-skilled workers are more prevalent in non-routine manual tasks and are therefore less affected by technological innovations. Furthermore, Dao, Das, Koczan & Lian (2017) conclude, based on evidence from 34 advanced economies between 1991-2014 that labor share decline driven by technology and global integration was particularly sharp for middle-skilled workers, resulting in job polarization. Sectors that are initially more specialized in routine-intensive activities experienced a larger decline in the labor share.

On the other hand, Handel (2012) finds no compositional changes in either the Unites States of America (USA) or the European Union (EU) between 1997 and 2009. Using the O*NET database he argues that the frequency of routinized tasks in general have declined during this period. Moreover, Spiezia, Polder & Presidente (2016) estimated the impact of ICT usage on labor demand from 1990-2012 in selected OECD countries. This study found permanent effects on labor demand by industry. In the short run ICT usage leads to increased polarization. However, in the long run these effects will disappear.  

The above-mentioned studies show mixed results of empirical evidence. However, Autor & Salomons (2017) argue several countervailing channels are at play that help determine the eventual effect of automation on the labor market. Namely; the own-industry output effects, cross-industry input output effects, between-industry effects, and final demand effects. They analyze these effects by making use of TFP as a proxy and make use of the EU Klems dataset release of 2007. Subsequently, by taking the sum of five distributed lags, they look into the different direct and indirect effects on labor market dynamics.

2.3.         Factors causing labor market polarization: automation, off-shoring and import low-wage countries

Other studies suggest that there are different factors that should be taken into account while analyzing the effect of technological innovations. Goos, Manning & Salomons (2014) discuss the notion of Routine Biased Technological Change (RBTC) in which is focused on how easily tasks can be routinized in different occupations. Their study analyzes the pervasiveness of job polarization in Europe between 1997 and 2010. They introduce offshoring as another important factor influencing labor market polarization. This argument suggests that routinised tasks, mainly prevalent in middle-skilled jobs, are being off-shored to low-wage countries. Their study however does not vary over time, which is a disadvantage of their identification analysis (Goos, Manning & Salomons, 2014). Another study confirming the importance of offshoring is the study of Oldenski (2014). Observing the USA labor market they find that an increase in off-shoring leads to an increase in the wage gap between middle, high and low-skilled workers.

A different development that is often associated with increased labor market polarization is import competition from China. Bloom, Draca & Van Reenen (2011) find that increased import competition is associated with a fall of the share of unskilled workers. However, they also conclude that import competition leads to more innovation in these sectors, representing the complicated relation between the two. The report of the OECD of 2017 defined the impact of off-shoring, technology and import competition from China to have the largest impact on polarization in the European Union. Their results show that within-industry changes are mainly caused by technological advances, while between-industry change is mainly attributed to the widespread deindustrialization of Europe. Furthermore, they suggest that automation replacing labor is mainly prevalent in the manufacturing industry (Breemersch, Damijan & Konings, 2017).

 

2.4.         Specific channels of technological innovations

The previously discussed papers all make use of proxies for technological innovations similar to the expenditures on R&D of a firm, the largest R&D investors, or ICT usage of a firm. This paper distinguishes itself by making use of TFP. A shortcoming of making use of TFP is that it incorporates productivity growth arising from all production factors. Therefore, you cannot identify the effect of specific technological innovations such as artificial intelligence or robotics. Solely the overall effect of the introduction of new technologies can be identified (Autor & Salomons, 2018).

Nevertheless, there are multiple other papers that look into these specific channels. This is most often done by analyzing the impact of robotics on the labor market. Acemoglu & Restrepo (2018; 2020a) for example make use of the dataset of the International Federation of Robotics to identify the effects of robots on the USA labor market. By making use of an Instrumental Variable (IV) strategy they find large and negative effects of robots on employment and wages in different commuting zones in the USA. In addition, Chiaccho, Petropoulos & Pichler (2018) apply similar data to the European labor market. They identify a significant displacement effect present, particularly for workers in the middle-skilled group. On the other hand, there are studies that find weakly positive effects on the labor market. Dauth Findeisen, Südekum & Wößner (2017) look into the German labor market and do not find evidence that robots destroy jobs between 1990 and 2014. Furthermore Graetz & Michaels (2018) find that investment in industrial robots did not reduce total employment, but potentially did affect the composition of the labor market. They identify negative effects for both the middle-skilled and low-skilled workers caused by these investments between 1993-2007 while observing 17 countries.

This study will not make any inferences on the effects of specific new technologies such as robotics but will refer to other literature to elaborate on this.

2.5.         Summary literature overview

Overall, the results of the previously discussed studies are mixed. Although part of the existing academic literature is able to confirm the influence of new technologies such as robotics on labor market polarization, other research papers have shown that potential countervailing channels take part in the eventual effect. Furthermore, literature has identified other important factors causing labor market polarization, off-shoring being the most important in within-industry analysis. The next section will elaborate upon the economic and theoretical foundation underlying the effect of new technologies on labor market dynamics.