Automation risk and local development in micro-regions: evidence from Hungary

doi:10.21203/rs.3.rs-3619607/v1

Download PDF

Research Article

Automation risk and local development in micro-regions: evidence from Hungary

https://doi.org/10.21203/rs.3.rs-3619607/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

This paper investigates automation risk in an Eastern European country, Hungary, at the subnational level. Local economies are increasingly characterised by different development paths and divergence thus efficiently tackling the diverse economic and societal challenges of automation in the local economies requires the understanding of the relationship between automation risk and development perspectives at the local level. Using a national wage survey database, this paper adopts the framework developed by Frey and Osborne in Hungarian NUTS4 regions to understand the spatial distribution of automation risk and its connection to economic development. Results indicate that more productive regions tend to benefit from automation by simultaneously decreasing their exposure over time and creating more jobs. Less developed regions, however, seem to face many challenges. Some of them can reduce exposure, however, at the cost of job destruction. Lagging regions, on the other hand, tend to increase their automation risk over time due to the increasing number of low-quality jobs and/or decreasing number of high-value-added jobs.

JEL code: J31

job polarisation

labour market automation

regional inequality

regional development

HSCO

Income differences in the labour markets of most developed countries increased dramatically since the 1980s, and they continue to grow. Acemoglu and Autor (2011) were among the first authors investigating the wage differences in the United States between 1980 and 2009. Although, since the beginning of the 20th century, wage differences decreased until the 1970s (Goldin and Katz 2008), they realised that this trend had ended as demand for higher-wage employees increased more than the need for middle- or lower-wage ones. The demand for people with higher qualifications increased, making their employment more desirable and increasing wages. Various reasons can be found behind this phenomenon, the most important of which is the professional polarisation of the labour market.

Later, in developed countries, it was also found that professions with wages on the lower or higher end of the wage distribution curve experienced a higher wage premium between 1980 and 2005 than those which were in the middle of the curve (USA (Autor and Dorn 2013), European Union (Goos et al. 2009)). As a result, the proportion of professions with lower or higher qualification levels in the labour market increased, and medium-skilled occupations lost more space. This U-shaped polarised labour market is a well-spread theory in developed countries (Autor et al. 2006; Fonseca et al. 2018; Goos et al. 2014). The shape of the current labour market development is not formed by a single factor but by the combined effect of several variables (Autor 2010).

During the last two decades, many new global trends emerged that significantly affected and reformed the labour market of developed economies. One of these trends is fast technological change. Thanks to the development of information and communication technologies, more and more routine tasks are replaced by mechanical, computer, or robotic solutions in manufacturing. As these phenomena spread, attention turned toward the future of different occupations and wages. The second trend is globalisation, which appears to be an effect of new global economic forces in countries such as China. This affects international trade, which involves import competition from developed countries, which changes the global value chain's production processes (Breemersch et al. 2017). The third trend is closely linked to the second one because the global economic forces' appearance significantly affects the offshoring of occupations. In this paper, the authors will focus on the first trend (technological change) and its impacts on the different regions.

The authors argue that one of the main reasons behind wage differences and job polarisation in technological change is the automation and computerisation of routine tasks. Autor and Dorn (2013) found that most service sector employees are at the lowest end of the wage distribution. Most routine task-based jobs are in the middle, and people with higher education are on the highest end of the distribution. They identify the technological changes behind this phenomenon. Acemoglu and Autor (2011) found evidence that through skill-biased technological change the wages of employees with routine tasks started to decline. That is why the lower-skilled workers turned their labour supply toward the service sector. The occupations within the service sector require direct physical appearance and human communication, and because of that, they are not easily automatised (Autor and Dorn 2013). Acemoglu and Restrepo (2022) document that over the last four decades, 50–70% of changes in the wage structure of the United States are accounted for by wage decreases of workers who specialised in routine tasks in those industries that experience rapid automation.

However, new technology does not necessarily reduce the demand for human work, but demand changes as it requires a different type of work (Autor 2010). In 2018, more than 60% of employment was found in job titles that did not exist more than 50 years ago (Autor et al. 2022). In this context, developed countries' labour markets realised a significant job polarisation which resulted in more highly-skilled, high-salaried, and low-skilled, low-paid jobs (Autor et al. 2006; Goos et al. 2009).

Besides labor market effects, automation and its polarisation effect can also cause various substantial economic impacts. Adapting automated technologies might result in improved productivity, generating additional economic growth and thus inducing additional labour demand. The larger spending caused by growth can also contribute to job growth through the multiplier effect. However, occupations that require lower or higher skills are differently influenced by job creation and job destruction; thus, employment effects depend on the labour market structure and the flexibility of workers to transition from 'old jobs' to newer ones. Finally, the larger share of capital involved in production might increase income inequalities since a small group of people owns capital stock. The interplay of all these factors and effects shapes the actual employment effects.

However, the risks and the potential economic effects of automation are unevenly distributed in space since regions differ significantly in terms of their labour market characteristics, industry mix, productivity, and demographic trends. Metropolitan areas are often found to be less susceptible to automation and better situated to benefit from technological change, while lagging mostly rural regions are more likely to face greater challenges in the near future regarding automation and its economic consequences.

Although studies offer controversial findings on the sign and extent of employment effects, there is a general fear that automation might reduce overall employment, at least in the medium-short term and increase wage- and income inequalities at the national level, and especially at the local level. Assessing and understanding these potential impacts and the mechanisms behind them might contribute significantly to designing and implementing policy interventions to mitigate the negative consequences of automation.

In this paper, the authors use detailed labour market survey data to investigate the spatial aspects of susceptibility to automation in the Hungarian low-level NUTS4 regions following the method developed by Frey and Osborne (2013). The paper shows that the most developed parts of Hungary are less exposed to automation while lagging and peripheral areas face larger risks of technological change.

The paper is structured as follows. The next section overviews of automation, its potential economic effects and their spatial characteristics. The third section discusses the methodological details and the data used in the analysis. Results and conclusions are discussed in the fourth and fifth sections.

2.1. The potential effects of automation

Automisation’s effects on employment depend strongly on the specific tasks workers typically perform in each country. In jobs that require tasks that can be easily substituted by automation (performing routine tasks based on codified knowledge), job loss can be expected (displacement effect). While jobs that involve tasks that are complemented by automation are less likely to be automated and can benefit from the positive (productivity) effects of technological development (Acemoglu and Restrepo 2020; Autor 2015; Autor et al. 2003). Through interindustry spillovers, however, other seemingly unrelated, primarily non-tradable activities might experience increased job creation (Autor and Salomons 2018; Mann and Püttmann 2023; Millington 2017; Moretti 2010).

Although there is a wide range of potential consequences of automation, most studies are primarily interested in whether automation will substitute or complement labour (Autor 2015; Autor and Salomons 2018); however, they often offer contradicting findings and predictions about the scope of job replacement. Acemoglu and Restrepo (2017) suggest that robots have a negative effect on local employment and wages. Jäger et al. (2015) and Graetz and Michaels (2018) indicate that robots increase productivity growth with minor job loss in the EU and OECD member states; however, they displace workers and reduce salaries in the USA. Georgieff and Milanez (2021) also found evidence that job loss is not clearly visible in OECD countries, and job creation in high-risk (in terms of automation) jobs is much slower than in low-risk jobs. Similarly, Dauth et al. (2017) found no evidence that robots cause total job losses in Germany; however, Leigh et al. (2020) demonstrated that robots contributed positively to employment in manufacturing in the US, and this finding was also reinforced by Mann and Püttman (2021) in which they found that automation might reduce employment in manufacturing, but job creation in other sectors overcompensates this effect. Thus, the employment effects of automation are not evident and difficult to predict.

Instead of determining the absolute employment effects, many studies focus on the potential extent of the risk of automation (such as Arntz et al. 2016; Frey and Osborne 2017). Estimates are very diverse regarding the extent of automation risk (47% in the US - Frey and Osborne (2013), 57% in the OECD countries - The World Bank (2016), 9% in the OECD countries, according to Arntz et al. (2016) /however they used a task-centered approach which explains the great difference of their results/).

Large automation risk does not mean that there will be technological unemployment in the labour market. First, even if ICT reaches the level that almost every task can be robotised, companies might not be willing to use it to its full extent. This will depend on the machines' prices and how much risk it takes to restructure their production processes. Second, changing the procedures for using the new technologies is a slow and long process from a legal and societal point of view. Lastly, even if the use of ICT intensifies significantly, it might create new workplaces, processes, and tasks (Acemoglu and Restrepo 2017).

Furthermore, adopting automation technologies can increase productivity, reducing production costs and product prices. Lower prices might induce higher demand for products which might be transferred into higher demand for factors of production, including labour (Acemoglu and Restrepo 2020). Thus, automation can contribute to economic and, consequently, employment growth. However, the benefits of technological change are distributed unevenly among economic actors since productivity gains have surpassed wage growth, and inequality has increased (Frey and Osborne 2015).

The potential effect of automation on employment also depends on the flexibility of wages. In countries with flexible wages (decreasing wages due to automation), negative employment effects can be mitigated. Lower labour costs might preserve a larger portion of jobs; however, wage decrease might cause further societal challenges. In the case of wage rigidity, negative employment effects are likely to become more pronounced (Brynjolfsson and McAfee 2011).

In addition to the direct employment effect, changes in the use of factors of production in industries can also affect income distribution (inequality). Due to automation, production technology becomes more capital intensive; thus, a larger share of capital is used in production, which, according to the neoclassical theory, contributes to the total income of households since they own factors of production. However, capital ownership is heavily concentrated in the hands of a small and already wealthier group of households; thus, a relative increase in capital income would intensify income inequalities (Autor 2015; World Bank 2016). Workers predominantly employed in routine jobs and with almost no capital income opportunities might face impoverishment. This shift of income distribution between different groups of households might weaken the multiplier effect of aggregate consumption spending, undermining the overall positive economic impact of automation (Brynjolfsson and McAfee 2011) in terms of economic growth and job creation. Thus, the most significant risk of automation is not a large-scale growth of unemployment but widening income inequality (Millington 2017), where the losers of automation might be the majority of society (Brynjolfsson and McAfee 2011).

2.2. Automation at the local level

Automation and its effects on labour markets have been discussed widely in the literature, mainly at the national and international levels. At the same time, regions are increasingly characterised by different development paths and divergence (Iammarino et al. 2019), which might influence their exposure to automation. In terms of that, national studies, on the one hand, do not analyse how different regions are susceptible to automation; on the other hand, these studies do not provide information on the potential spatial economic effects of automation. However, since regions vary significantly in their labour market structure, industry mix, and other characteristics regions face different automation challenges, susceptibility to automation and its economic effects are not uniform in space (Wyonch 2018). Nevertheless, the spatial aspects of automation have been relatively scarcely discussed (Bissell and Del Casino 2017; Crowley et al. 2021; Czaller et al. 2021; Frank et al. 2018). This section, briefly accounts for the spatial economic aspects of automation.

Agglomerated regions, on the one hand, are usually considered the hubs of knowledge creation and technological progress (creative change). At the same time, less-developed areas are typically responsible for fewer knowledge-related activities (Andersson et al. 2020). Thus, technology-driven job creation tends to be primarily located in urban areas, where the required knowledge base, innovation capacities, related activities, and different actors of the local innovation system are present (Shutters et al. 2016). As a result, metropolitan regions (cities) are characterised by high demand for non-automatable tasks (Duranton and Puga 2005) since many industries heavily reliant on non-routine jobs are concentrated in these locations. These tasks are performed by high-skilled workers who are usually more specialised at their jobs and less exposed to automation risk (Frank et al. 2018). In addition, due to their size, many low-skilled jobs in the service sector (e.g., health, education, catering-related jobs) are also highly demanded in these locations (Scott 2009). However, routine jobs in manufacturing are generally more exposed to relocation to peripheral areas where both labour costs and land rents tend to be lower (Storper 2013). The intensive use of ICT and automation contributes to this effect as well. They allow firms to relocate routine tasks to peripheral locations and keep creative, knowledge-based, core activities in agglomerations (Eeckhout et al. 2019) that require highly qualified labour or strong social skills (Autor et al. 2003; Frey and Osborne 2017).

Rural economies, on the other hand, can be adversely affected by automation. Since job creation and skilled workers are concentrated mainly in capital regions, demographic trends (e.g., aging, migration, commuting) might also worsen local economic prospects (OECD 2018). The outflow of working-age (especially high-skilled) individuals tends to -educated workers and a lower share of the service sector in which typically jobs are harder to be automated, and they are more likely to rely on a few prominent local employers. These factors might contribute to the higher automation risk in rural areas (OECD 2018). In conclusion, less developed, rural regions with labour market and economic difficulties tend to be more susceptible to automation. Agglomerated areas, cities, on the other hand, are less exposed to automation risk due to their favourable economic conditions and labour market characteristics (Czaller et al. 2021; Ding et al. 2018; Frank et al. 2019; Hilpert 2021; OECD 2018; Wyonch 2018).

Apart from the differences in automation risks, the potential extent and content of its economic effects are expected to differ in regions. As Autor (2019) pointed out, the hollowing-out effect of automation is more pronounced in larger cities. Furthermore, Leigh et al. (2020) found that investments in automation in agglomerated areas can stimulate employment growth in the long run since these technologies complement automatable tasks that might upgrade local skills.

Finally, the potential spatial effects of automation might also be influenced by the automation capabilities of different regions. According to Leigh and Kraft (2017), regions have substantial differences in their robotics knowledge bases, shaping their automation capabilities.

Frey and Osborne (2013) analysed mobile robotics and machine learning to investigate how susceptible jobs are to computerisation. In Mobile Robotics, new algorithms and developments arrived because they can use big data as sources. These processes led to the fact that previously non-routine tasks could be automatised. Occupations that rely heavily on complex perception and manipulation tasks, creative intelligence tasks, or social intelligence tasks are unlikely to be robotised or replaced by computers in the next two decades. Based on these, the authors stated that the probability of an occupation being computerised could be described as a function of these task characteristics. For example, based on their perception, the requirement of a low degree (they created a 0 to 100 automation risk scale, 0 resulting in low probability) of social intelligence in an occupation makes it more susceptible to automation.

Frey and Osborne (2013) used the O*Net database to implement their methodology; the O*NET was initially collected from labour market analysts and regularly updated. It contains detailed information about occupations, most of which correspond closely to the Standard Occupation Classification 2010 (SOC10) of the Labour Department. The O*NET defines those critical features of an occupation which is necessary for their method. The database allowed them to objectively rank occupations according to the mix of KSA-s (knowledge, skills, and abilities) and subjectively categorise them based on their various tasks. (Frey and Osborne 2013)

This approach was applied in this study; therefore, the initial task was to match the SOC10 with the Hungarian Standard Classification of Occupations 2008 (HSCO08). However, this process was not evident, as a direct translator could not be found. Therefore, this translation was done in several steps (Fig. 1.). First, the SOC10 was translated into the International Standard Classification of Occupations 2008 (ISCO 08) (BLS 2012), which was then translated into HSCO08 (KSH 2021). An additional translation was necessary for one year of the database used in this study; for 2007, an HSCO08-HSCO93 (KSH 2008) matching was carried out as there has been realised an ISCO harmonisation in Hungary in 2008. Altogether, 485 different occupations are classified in the HSCO08 system, of which only 28 resulted in no match¹. The complete list of the HSCO08 FO values can be found in Appendix 1.

In the analysis, Hungarian Wage Surveys² data were used (Krtk 2016)for three selected years (2007, 2011, and 2016). The justification of this selection is the following: the aim of the study is to analyse a 10-year-long period. Since 2016 is the latest data available, the time period is 2007–2016, where the initial year is not distorted by the effects of the global financial crisis. As an intermediate year, 2011 is also selected in the middle of the period, which can be evaluated as the peak year of the crisis' impact in Hungary.

The Wage Survey contains all of the public sector employees by random sampling, while for the non-public sector, all of the employees of the companies above 50 employees are selected based on random sampling; below this limit, all of the employees of a random sampling of companies are included (5–50 employees)³ (Fig. 2.). The database contains annual data on hundreds of thousands of employees.

Initially, a consistency check was run for all available data; however, it turned out that the public sector had to be removed from the analysis for two reasons. First, in Hungary, after 2016, the number of public workers increased dramatically (Molnár et al. 2019) due to government programs. As all public sector employees are affected by the random sampling, their proportion increased heavily, which increased the FO values since most public jobs are associated with high automation risk. Second, the public sector might have different (e.g., political, solidarity, other societal) purposes when employment-related policy decisions are made (as suggested in the literature, e.g., Kóti 2019). As a result, only the data on the private sector employed occupations remained in the analysis (Table 1). As it can be seen, the sample ratio is larger than 8% in all the years, leading to a high-quality data interpretation and validity; therefore, representativeness is ensured.

Table 1

Representativeness of the analysed data
Year	Total no of employees	No of employees in the non-public sector	Sample data	Ratio
2007	2,760,672	2,012,084	171,895	8.54%
2011	2,691,507	1,956,592	162,673	8.31%
2016	2,977,888	2,105,079	186,568	8.86%
Source: CSO Dissemination Database (2021)

The authors analysed 175 Hungarian NUTS4 micro-regions. Eventually, the data matching and collection process resulted in 2 micro-region-HSCO matrices (175 regions x 482 occupations) regarding the following two areas:

the employees,
the FO values.

Region-specific weighted average FO values (automation risk) were calculated using employment and FO values from these matrices. To interpret the changes in employment and automation risk, regions were categorised into four groups:

General improvement (GenImp): increasing employment and improving FO values,
General deterioration (GenDet): decreasing employment and worsening FO values,
Employment-based deterioration (EmpDet): increasing employment and worsening FO values,
Employment-based improvement (EmpImp): decreasing employment and improving FO values.

This categorisation scheme allows for the identification of the changes, considering whether the FO value improvement or deterioration was due to the general trend of having a lower level of employment or not. Furthermore, to account for the interplay between automation risk, productivity and agglomeration effects, gross value added per employee (Cai and Leung 2020), and urbanisation indices were also estimated⁴.

The spatial distribution of labour productivity is reported in Fig. 1. In 2016, the Hungarian value was 34.6 thousand EUR (315 EUR/HUF considered) (Eurostat 2021). The darker areas can be considered economically more developed areas, as there is a much higher productivity per employee and, as a result, income generating capacity. Regarding the spatial characteristics of productivity, the following tendencies can be identified:

Central areas surrounding the capital have higher productivity since most firms and the headquarters of international companies, with large value-added that mainly require highly qualified labour, are seated there (as stated by Duranton and Puga 2005).
It seems that productivity strictly follows the motorway lines heading toward the capital from the different parts of the country, indicating that accessibility and infrastructure also play a crucial role in economic development, similar to the findings of the literature (e.g., Yang et al. 2020).
Apart from the capital, more developed manufacturing areas are also highly productive in the western part of the country, where a lot of FDI-financed manufacturing takes place (supporting the findings of Eeckhout et al. 2019; Frank et al. 2018; Scott 2009).
More populated regions (such as the county seats) are also more productive than the average, which means that the spatial concentration of economic activities and people might contribute to positive local effects, which is confirmed by the significant correlations between population and GVA values (see later) and previous findings in the literature (confirming the findings of Eeckhout et al. 2019; Shutters et al. 2016).
Knowledge centers, university cities also attract economic activities, making some of the county seats more attractive for firms, promoting productivity and economic growth (in line with the findings of Andersson et al. 2020; Oschinski and Wyonch 2017; Shutters et al. 2016).

Figure 2 reports the regional average FO values. By comparing Figs. 1 and 2, it seems that automation risk tends to be smaller in more developed (productive) areas and vice versa. This is further reinforced by the significant correlation between productivity (GVA in thousands of EUR per employee) and automation risk (FO value) (2016: Pearson correlation=-0.425, p = 0.000; 2011: Pearson correlation=-0.392, p = 0.000; 2007: Pearson correlation=-0.331, p = 0.000). The correlation test was also run for the urbanisation index and automation risk as well and found to be statistically significant as well (2016: Pearson correlation=-0.465, p = 0.000; 2011: Pearson correlation=-0.300, p = 0.000; 2007: Pearson correlation=-0.240, p = 0.000), which indicates that a higher level of urbanisation is accompanied by a lower level of automation risk, which is strengthening the findings of Shutters et al. (2016). Similar results occur in the case of the population-based correlations (2016: Pearson correlation=-0.323, p = 0.000; 2011: Pearson correlation=-0.213, p = 0.000; 2007: Pearson correlation=-0.162, p = 0.032), which also confirm the findings of the literature review (Czaller et al. 2021; Ding et al. 2018; Frank et al. 2019; OECD 2018; Wyonch 2018).

Considering the change in automation risk over time, in general, Hungarian regions experienced a decrease in the risk of automation⁵. The average FO value in 2007 was 62.24, lowered in 2011 to 57.47; in 2016, it was further decreased to 54.90. Furthermore, the majority (61%) of the regions experienced a steady and significant improvement in automation risk⁶. The shrinking automation risk seems to be accompanied by productivity improvements in the more productive regions (and vice versa) which is confirmed by the One-Way ANOVA analysis: in 2016, the GVA for continuous improvement was 7.20 thousand of EUR, while for non-continuously improving, only 3.97, F = 12.978 Levene p = 0.001 Welch p = 0.000 ANOVA p = 0.000; and the urbanisation value for constant improvement was 60.27%, while for non continuously improving only 49.18%, F = 13.306 Levene p = 0.373 Welch p = 0.000 ANOVA p = 0.000). It appears that especially the eastern parts of Hungary are simultaneously characterised by large automation risk, low productivity and these values are not improving over time. Thus, the spatial distribution of productivity and automation risk seem to be persistent over time, which might indicate that the chances of decreasing spatial disparities are limited, partially due to automation risks.

In the last analysis, the previously introduced micro-regions categorisation is used to map the simultaneous change in automation risk (change in FO values due to structural change in the employment in occupations) and in employment. Figure 3 shows the territorial distribution. The majority of the micro-regions (GenImp: 98 pcs, 56% of the micro-regions which contains the 79% of the employees) experienced a simultaneous improvement in automation risk and employment level, which might mean net job creation did happen in occupations less exposed to automation risk. The second largest group of regions is in the category of employment-driven improvement (EmpImp: 61 pcs, 35%), where the improvement of the FO values is accompanied by a decrease in the number of employed people, which might mean that automation risk was reduced due to larger job loss in occupations that are more susceptible to automation and this effect was not compensated by job creation in higher qualified jobs. Regions in ‘employment-driven deterioration’ (EmpDet: 7 pcs, 4%) also faced an increase in automation risk and job growth. This is an unhealthy economic and social situation because these regions could increase employment in occupations that are more exposed to automation risk which is not sustainable in the longer run. The last region group consist of the most lagging regions (general deterioration - GenDet: 9 pcs, 5%) in which economies suffered from losing jobs in occupations that are less susceptible to automation, increasing overall automation risk.

Considering the economic development of regions, there are strong spatial dependencies in how regions evolved over time in terms of employment and automation risk. Table 2 reports the values of the automation risk (FO), productivity (GVA), and urbanisation of the four categories of regions and the relationship between these variables. According to the results, the GenImp territories are typically categorised as more productive, urbanised⁷ regions that seem to cope efficiently with automation risks (e.g. the risk of automation is the lowest in these regions). This category consists of the country’s most developed areas (the capital, industrialised regions, county seats, university towns). EmpImp regions seem to cope with automation as well (second lowest risk); however, due to their lower economic development, they struggle with job growth. This group seems to be the ‘hinterland’ of more developed areas (e.g., commuting zones). EmpDet and GenDet regions account for only 9% of the total regions, and these territories are categorised as the most lagging (low productivity, less urbanised) economies of the country.

Furthermore, these areas simultaneously suffer from losing quality jobs and worsening automation risk (and the highest levels among the regions), which indicates that long-lasting economic and social challenges are in place and policy actions are necessary, especially for these regions.

Table 2

One-Way ANOVA of FO, GVA, and urbanisation by FO and employment improvement categories from 2007 to 2016
	N	FO	GVA	Urbanisation
GenImp	98	52.07	7.08	58.00%
EmpImp	61	53.12	4.80	54.00%
EmpDet	7	63.73	3.04	44.00%
GenDet	9	65.47	1.99	48.00%
Total	175	53.59	5.86	56.00%
F		13.977	3.864	1.766
Levene		0.003	0.011	0.025
Welch		0.000	0.000	0.000
ANOVA		0.000	0.010	0.156
Source: own edition based on Krtk (2016)

This paper studied automation risk, productivity, and urbanisation in 175 Hungarian micro-regions in 2007, 2011, and 2016. This spatial analysis is based on the classification of occupations which is relatively new in Europe. Only a few studies (e.g., Goos et al. 2009) can be identified that predominantly use different analysis frameworks, and no similar studies have been found for Hungary. Furthermore, the paper provides the description of the process of the translation process of the FO values into the HSCO system by completing it with an estimation for the 28 occupations without a FO matching.

The study’s most important findings are the following. First, in general, automation risk has decreased in Hungary over this 10-year period, but significant spatial differences can be observed. Second, regions typically less susceptible to automation are primarily more productive, urbanised areas or strongly connected to these areas by transport infrastructure (e.g., motorway) and vice versa. Third, considering the temporal dimension of automation risk, the gap between developed and less developed regions seems to have increased. Fourth, regions tend to cope with these challenges differently in each group: developed, more productive regions tend to improve their capacity to cope with automation risk by either creating more less-susceptible, high-value-added jobs or destroying more highly-susceptible, low-quality jobs while lagging regions tend to face increasing risks of automation by either losing quality jobs or creating low value-added and highly susceptible jobs. Fourth, based on that, it seems that automation tends to substitute labour in less developed regions (Autor 2015; Autor and Salomons 2018), creating more unemployment, while in developed regions, there could be a complementary relationship between employment and technology, creating more high-value-added jobs. As a result, this study argues that automation and economic development must be examined at the local level to formulate policies that can effectively tackle local challenges⁸. The results are in line with the findings of the following researchers: Czaller et al. (2021); Ding et al. (2018); Frank et al. (2019); OECD (2018); Wyonch (2018).

This section briefly accounts for the most important limitations of this study and aims to designate potential future research areas.

First, this study applied the automation risks Frey and Osborne (2013) developed for the US economy. Consequently, the same underlying assumptions, processes, trends, and determining factors are applied to Hungarian occupations, too.

Second, the dataset used in this study is only available for Hungary, which should be extended to more countries of the EU to allow for cross-country comparisons.

Third, in this study, automation risk is calculated based on occupations and not tasks as suggested by Arntz et al. (2016); therefore, as a next step, this alternative method could be considered in terms of the additional insights it could provide besides the findings of the present paper.

Fourth, this study only accounts for automation risk and not the offshorability of occupations investigated by Blinder (2007). However, these two processes simultaneously happen in the global competition, putting pressure on local economies. Thus, this approach also should be analysed in future works.

Fifth, a further research area could be to apply the approach of Goos et al. (2009) or Autor (2010) to rerun the analyses for the three years and investigate the eventual differences along with them.

Finally, other approaches, such as Brynjolfsson and McAfee (2014), could be applied where a specific industry or occupational classification could be placed in the focus to identify the related peculiarities.

FO value

Frey and Osborn automation values

HSCO

Hungarian Standard Classification of Occupations

ISCO

International Standard Classification of Occupations

NUTS

Nomenclature of Territorial Units for Statistics

SOC

Standard Classification of Occupations

Availability of data and materials

The data that support the findings of this study are available from Centre for Economic and Regional Studies but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Centre for Economic and Regional Studies.

Competing interests

The authors declare that they have no competing interests

Funding

Project no. TKP2021-NKTA-19 has been implemented with the support provided from the National Research, Development and Innovation Fund of Hungary, financed under the TKP2021-NKTA funding scheme.

Authors' contributions

JN and NS analysed and interpreted the Wage Tariff Database of Hungarian microregions, and created the conceptual model. NSz applied the spatial approach in interpreting the data and was a major contributor in writing the manuscript. All authors read and approved the final manuscript.

Conflicts of interest: The authors declare no conflicts of interest.

Code availability: The calculations were applied on an online database which requires access, and the original data cannot be exported. The calculations afterward and the STATA codes can be provided on request.

Ethics approval: Not applicable.

Consent to participate: Not applicable.

Consent for publication: Not applicable.

Acemoglu D, Autor DH (2011) Skills, tasks and technologies: Implications for employment and earnings. In: Card D, Ashenfelter O (eds) Handbook of labor economics, vol 4. Elsevier, North Holland, pp 1043-1171
Acemoglu D, Restrepo P (2017) Robots and Jobs: Evidence from the U. S. Labor Markets NBER Working Paper doi:10.3386/w23285
Acemoglu D, Restrepo P (2020) Robots and jobs: Evidence from US labor markets Journal of Political Economy 128:2188-2244
Acemoglu D, Restrepo P (2022) Tasks, Automation, and the Rise in U.S. Wage Inequality Econometrica 90:1973-2016 doi:10.3982/ecta19815
Andersson L-F, Danley T, Eriksson R, Henning M (2020) Workers’ participation in regional economic change following establishment closure Small Business Economics 54:589-604 doi:10.1007/s11187-018-0036-2
Anelli M, Colantone I, Stanig P (2019) We were the robots: Automation and voting behavior in western europe BAFFI CAREFIN Centre Research Paper
Antonietti R, Cattani L, Pedrini G (2023) Robots and the productivity of local manufacturing systems in Emilia-Romagna: the mediating role of occupational similarity and complexity European Planning Studies:1-25 doi:10.1080/09654313.2022.2161816
Arntz M, Gregory T, Zierahn U (2016) The risk of automation for jobs in OECD countries: A comparative analysis OECD Social, Employment and Migration Working Papers
Autor D, Chin C, Salomons A, Seegmiller B (2022) New Frontiers: The Origins and Content of New Work, 1940–2018 MIT Working Paper doi:10.3386/w30389
Autor DH (2010) The polarization of job opportunities in the US labor market: Implications for employment and earnings Center for American Progress and The Hamilton Project 6:11-19
Autor DH (2015) Why Are There Still So Many Jobs? The History and Future of Workplace Automation Journal of Economic Perspectives 29:3-30 doi:10.1257/jep.29.3.3
Autor DH (2019) Work of the Past, Work of the Future AEA Papers and Proceedings 109:1-32 doi:10.1257/pandp.20191110
Autor DH, Dorn D (2013) The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market American Economic Review 103:1553-1597 doi:10.1257/aer.103.5.1553
Autor DH, Katz LF, Kearney MS (2006) The Polarization of the U.S. Labor Market American Economic Review 96:189-194 doi:10.1257/000282806777212620
Autor DH, Levy F, Murnane RJ (2003) The Skill Content of Recent Technological Change: An Empirical Exploration The Quarterly Journal of Economics 118:1279-1333 doi:10.1162/003355303322552801
Autor DH, Salomons A (2018) Is automation labor share–displacing? Productivity growth, employment, and the labor share Brookings Papers on Economic Activity 49:1-87 doi:10.3386/w24871
Barca F, McCann P, Rodríguez-Pose A (2012) The Case for Regional Development Intervention: Place-Based Versus Place-Neutral Approaches Journal of Regional Science 52:134-152 doi:10.1111/j.1467-9787.2011.00756.x
Bissell D, Del Casino VJ (2017) Whither labor geography and the rise of the robots? Social & Cultural Geography 18:435-442 doi:10.1080/14649365.2016.1273380
Blinder AS (2007) How Many US Jobs Might Be Offshorable CEPS Working Paper
Crosswalks between the 2010 SOC and systems used by other Federal and international statistical agencies (2012) https://www.bls.gov/soc/soccrosswalks.htm. Accessed 24 May 2021
Breemersch K, Damijan JP, Konings J (2017) Labour market polarization in advanced countries: impact of global value chains, technology, import competition from China and labour market institutions OECD Social, Employment and Migration Working Papers doi:10.1787/06804863-en
Brynjolfsson E, McAfee A (2011) Race against the machine: How the digital revolution is accelerating innovation, driving productivity, and irreversibly transforming employment and the economy. Digital Frontier Press, Lexington, Massachusetts
Brynjolfsson E, McAfee A (2014) The second machine age: Work, progress, and prosperity in a time of brilliant technologies. WW Norton & Company,
Cai J, Leung P (2020) A note on linkage between gross value added and final use at the industry level Economic Systems Research 32:428-437 doi:10.1080/09535314.2020.1718617
Crowley F, Doran J, McCann P (2021) The vulnerability of European regional labour markets to job automation: the role of agglomeration externalities Regional Studies 55:1711-1723 doi:10.1080/00343404.2021.1928041
Czaller L, Eriksson R, Lengyel B (2021) Automation risk along individual careers. static and dynamic upgrades in cities Papers in Regional Science 100:1545-1569 doi:10.1111/pirs.12635
Munkaerő, Keresetek, Nemzetgazdaság havi kereseti és létszám adatai (2021) http://statinfo.ksh.hu/. Accessed 22 May 2021
Dauth W (2014) Job polarization on local labor markets IAB-Discussion Paper
Dauth W, Findeisen S, Südekum J, Woessner N (2017) German robots-the impact of industrial robots on workers. CEPR Discussion Paper
Ding L, Leigh EW, Harker P (2018) Automation and Regional Employment in the Third Federal Reserve District. Federal Reserve Bank of Philadelphia,
Duranton G, Puga D (2005) From sectoral to functional urban specialisation Journal of Urban Economics 57:343-370 doi:10.1016/j.jue.2004.12.002
Eeckhout J, Hedtrich C, Pinheiro R Automation, Spatial Sorting, and Job Polarization. In: 2019 Meeting Papers, 2019. vol 581. Society for Economic Dynamics,
Apparent labour productivity by NACE Rev. 2 1 000 EUR (2021) http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=tin00152&lang=en. Accessed 5 May 2021
Fine D, Havas A, Hieronimus S, Jánoskuti L, Kadocsa A, Puskás P (2018) Transforming our Jobs: Automation in Hungary. McKinsey & Company,
Fonseca T, Lima F, Pereira SC (2018) Job polarization, technological change and routinization: Evidence for Portugal Labour Economics 51:317-339 doi:10.1016/j.labeco.2018.02.003
Frank MR et al. (2019) Toward understanding the impact of artificial intelligence on labor Proceedings of the National Academy of Sciences 116:6531-6539 doi:10.1073/pnas.1900949116
Frank MR, Sun L, Cebrian M, Youn H, Rahwan I (2018) Small cities face greater impact from automation Journal of The Royal Society Interface 15 doi:10.1098/rsif.2017.0946
Frey CB, Osborne M (eds) (2015) Technology at work. The future of innovation and employment. Citi GPS: Global Perspectives & Solutions,
Frey CB et al. (2016) Technology at work v2. 0: The future is not what it used to be. CityGroup and University of Oxford, Oxford
Frey CB, Osborne MA (2013) The future of employment: How susceptible are jobs to computerisation? Mimeograph, Oxfod Martin School
Frey CB, Osborne MA (2017) The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change 114:254-280 doi:10.1016/j.techfore.2016.08.019
Georgieff A, Milanez A (2021) What happened to jobs at high risk of automation? OECD Social, Employment and Migration Working Papers doi:10.1787/10bc97f4-en
Goldin C, Katz LF (2008) The Race between Education and Technology. The Belknap Press of Harvard University Press, Cambridge, Massachusetts. doi:10.2307/j.ctvjf9x5x
Goos M, Manning A, Salomons A (2009) Job Polarization in Europe American Economic Review 99:58-63 doi:10.1257/aer.99.2.58
Goos M, Manning A, Salomons A (2014) Explaining Job Polarization: Routine-Biased Technological Change and Offshoring American Economic Review 104:2509-2526 doi:10.1257/aer.104.8.2509
Graetz G, Michaels G (2018) Robots at Work The Review of Economics and Statistics 100:753-768 doi:10.1162/rest_a_00754
Hilpert YM (2021) Glowing cities and the future of manufacturing in the US and Europe: How digitalization will impact metropolitan areas depending on sectoral dominances and regional skill distribution European Planning Studies 29:1672-1689 doi:10.1080/09654313.2021.1963052
Iammarino S, Rodriguez-Pose A, Storper M (2019) Regional inequality in Europe: evidence, theory and policy implications Journal of Economic Geography 19:273-298 doi:10.1093/jeg/lby021
Jäger A, Moll C, Som O, Zanker C (2015) Analysis of the Impact of Robotic Systems on Employment in the European Union. Fraunhofer Institute for Systems and Innovation Research,
Kóti T (2019) Efficiency of active labour market policy in Hungary: detransitive settlement structure of supported public employment Folia Geographica 61:45-70
Bértarifa (2016) http://econ.core.hu/adattar/bertarifa.html. Accessed 22 May 2021
Conversion key between the items of the current version of classification and those of the previous version. (2008) https://www.ksh.hu/docs/osztalyozasok/feor/fordkulcs_feor93_feor08_eng.pdf. Accessed 22 May 2021
Correspondence table between the elements of the recent valid Hungarian and international versions of the classification (2021) https://www.ksh.hu/docs/osztalyozasok/feor/fordkulcs_isco_feor_eng.pdf. Accessed 20 May 2021
Leigh NG, Kraft B, Lee H (2020) Robots, skill demand and manufacturing in US regional labour markets Cambridge Journal of Regions, Economy and Society 13:77-97 doi:10.1093/cjres/rsz019
Leigh NG, Kraft BR (2017) Emerging robotic regions in the United States: insights for regional economic evolution Regional Studies 52:804-815 doi:10.1080/00343404.2016.1269158
Lindley J, Machin S (2013) Labour market polarization, urbanization and skillbiased consumption. Paper presented at the CESifo Area Conference on Economics of Education, Munich, 6-7 September
Mann K, Püttmann L (2023) Benign Effects of Automation: New Evidence from Patent Texts Review of Economics and Statistics 105:562-579 doi:10.1162/rest_a_01083
Manyika J, Chui M, Miremadi M (2017) A future that works: AI, automation, employment, and productivity McKinsey Global Institute Research, Tech Rep 60
Millington AK (2017) How changes in technology and automation will affect the labour market in Africa. K4D Helpdesk Report. Institute of Development Studies, Brighton, UK
Molnár G, Bazsalya B, Bódis L, Kálmán J (2019) Public works in Hungary. actors, allocation mechanisms and labour market mobility effects SOCIOHU:117-142 doi:10.18030/socio.hu.2019en.116
Moretti E (2010) Local Multipliers American Economic Review 100:373-377 doi:10.1257/aer.100.2.373
OECD (2018) Job Creation and Local Economic Development 2018: Preparing for the Future of Work. OECD Publishing, Paris. doi:10.1787/9789264305342-en
Oschinski M, Wyonch R (2017) Future shock? The impact of automation on Canada's labour market Commentary
Rodríguez-Pose A (2018) The revenge of the places that don’t matter (and what to do about it) Cambridge Journal of Regions, Economy and Society 11:189-209 doi:10.1093/cjres/rsx024
Santos I (2016) Labor market polarization in developing countries: challenges ahead. The World Bank,
Schwanen D (2017) Innovation Policy in Canada: A Holistic Approach CD Howe Institute Commentary 497
Scott AJ (2009) Human capital resources and requirements across the metropolitan hierarchy of the USA Journal of Economic Geography 9:207-226 doi:10.1093/jeg/lbn051
Shutters ST, Muneepeerakul R, Lobo J (2016) Constrained pathways to a creative urban economy Urban Studies 53:3439-3454 doi:10.1177/0042098015616892
Storper M (2013) Keys to the city: How economics, institutions, social interaction, and politics shape development. Princeton University Press, Princeton
World Bank (2016) World development report 2016: Digital dividends. The World Bank, Washington DC
Wyonch R (2018) Risk and Readiness: The Impact of Automation on Provincial Labour Markets. SSRN Electronic Journal. doi:10.2139/ssrn.3112877
Yang Z, Li C, Jiao J, Liu W, Zhang F (2020) On the joint impact of high-speed rail and megalopolis policy on regional economic growth in China Transport Policy 99:20-30 doi:10.1016/j.tranpol.2020.08.007

As the Armed Force Occupations of ISCO08 and HSCO08 are not present in SOC10, the authors had to identify automation risks for 25 occupations. This was realised by analysing similar professions and using the combined value of at least two Frey-Osborne (hereinafter FO) scores.
https://adatbank.krtk.mta.hu/en/adatbazisok/elerheto-adatbazisok/
However information is not available regarding the demographic information about individuals, such as marital status or the number of children, because the surveys only collect data in connection with the analysed companies (including the company’s location, industry, or geographical, and territorial data).
Where GVA is needed to be estimated since official data is only available at the NUTS3 level and above. Urbanisation index is calculated as the percentage of population living in urban areas.
The FO values for the micro-regions in 2007, 2011, and 2016 can be seen in Appendix 2.
A continuous improvement was noted with Yes (102 pcs) or No (73 pcs), visualised in Appendix 3.
Even if urbanisation is not significant, the post-hoc test confirms the difference between improvement and deterioration categories.
Potential policy response to automation is discussed in detail in Appendix 4.

Appendices.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Automation risk and local development in micro-regions: evidence from Hungary

Status:

Version 1

Abstract

Figures

1. Introduction

2. Automation, economic effects and their spatial distribution

2.1. The potential effects of automation

2.2. Automation at the local level

3. Methodology

4. Results

5. Conclusions and discussion

6. Limitations and areas of future research

Abbreviations

Declarations

References

Footnotes

Supplementary Files

Status:

Version 1