Automation-induced reshoring and potential implications for developing economies

Nii-Aponsah, Hubert; Verspagen, Bart; Mohnen, Pierre Working Paper Automation-induced reshoring and potential implications for developing economies UNU-MERIT Working Papers, No. 2023-018 Provided in Cooperation with: Maastricht Economic and Social Research Institute on Innovation and Technology (UNU-MERIT), United Nations University (UNU) Suggested Citation: Nii-Aponsah, Hubert; Verspagen, Bart; Mohnen, Pierre (2023) : Automationinduced reshoring and potential implications for developing economies, UNU-MERIT Working Papers, No. 2023-018, United Nations University (UNU), Maastricht Economic and Social Research Institute on Innovation and Technology (UNU-MERIT), Maastricht This Version is available at: https://hdl.handle.net/10419/326868 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-sa/4.0/ #2023-018 Automation-induced reshoring and potential implications for developing economies Hubert Nii-Aponsah, Bart Verspagen and Pierre Mohnen Published 22 May 2023 Maastricht Economic and social Research institute on Innovation and Technology (UNU-MERIT) email: [email protected]u | website: http://www.merit.unu.edu Boschstraat 24, 6211 AX Maastricht, The Netherlands Tel: (31) (43) 388 44 00 UNU-MERIT Working Papers ISSN 1871-9872 Maastricht Economic and social Research Institute on Innovation and Technology UNU-MERIT | Maastricht University UNU-MERIT Working Papers intend to disseminate preliminary results of research carried out at UNU-MERIT to stimulate discussion on the issues raised. Automation-Induced Reshoring and Potential Implications for Developing Economies Hubert Nii-Aponsah, Bart Verspagen and Pierre Mohnen Abstract Technological progress in automation technologies, such as Artificial Intelligence (AI), is expected to impact production activities beyond the home country adopting them as countries interact within the global trade system. Firms tend to offshore production activities to other countries when it is more profitable to produce elsewhere than at home. The adoption of automation technologies reduces the cost of producing in the home country, making previous offshore locations relatively less attractive. From a global perspective, the altered cost structure induces reshoring: a reorganization of production activities back home or to other lower-cost locations. Developing economies, which previously served as low-cost locations, could be adversely impacted by experiencing a drop in the production of the affected sectors and goods. This paper analyses the potential effect of automation on the global portfolio of trade specialization based on the principle of comparative advantage, employed in an extension of Duchin’s World Trade Model to include non-tradable sectors. Through scenario-based analyses within the global economic context and using data, primarily, from the World Input-Output Database (WIOD) and the International Assessment of Adult Competencies (PIAAC), we find that countries in lower-income Asia are likely to be the most adversely affected by reshoring induced by automation in advanced economies. Keywords: Reshoring, Automation, Specialization, Developing Countries, Advanced Countries JEL: D33; E25; F14; F17; F47; J21; O33 2 1. Introduction The past decade has been characterized by growing scholastic interest in the possible effects of new technologies in the wake of advances in automation technologies. Mobile Robotics, 3-D printing, Internet of Things, and Machine Learning are among the areas experiencing notable developments. Novel automation technologies currently possess the capacity to replace a variety of both manual and cognitive tasks. They can manage customer relations through chatbots, identify accounting miscalculations, accurately diagnose sicknesses, and monitor social media content, among others (Howard & Borenstein, 2020; Ivanov et al., 2020; Prettner and Bloom, 2020). Given the early development and adoption of smart automation technologies in the advanced world, most of the research focused on industrialized economies. The seminal paper by Frey and Osborne (2017) estimated that about 47 percent of US jobs face high automation risk (a risk level above 70 percent). Several other studies followed: for instance, Arntz et al. (2017) argued that it is critical to account for differences in tasks performed under the same job as doing so reduced the estimated proportion of US jobs at high risk of automation. Previous work mainly concentrated on the within-country effects of automation technologies on employment and wages in the advanced world (Acemoglu and Restrepo, 2018; Gadberg et al., 2020). Relatively fewer studies have focused on the developing world such as Li et al, 2020, which documented the movement of labour from manufacturing to services in China. In recent years, reports suggest that the impacts of automation-based technologies on output and workers can spill over geographical borders. Some industrial giants that formerly offshored parts of their production to lower-wage developing countries have been reported to relocate or reshore their production activities amidst rising wages in emerging markets and declining automation costs. 1 Adidas established Speedfactories in Ansbach, Germany, and Atlanta, USA, with the view to automating production using additive manufacturing. 2 Firms such as General Electric, Bosch, Philips, and Caterpillar have also been featured in the discussion (The Economist, 2013; Fratocchi et al., 2014; The Economist, 2017). The anecdotal evidence seems to indicate that, even if the diffusion and adoption of automation technologies in developing countries are limited by existing structural bottlenecks, as 1 Offshoring and reshoring are both complex phenomena and can involve multiple motivations beyond labour costs such as flexibility to quickly meet changing demand, quality considerations, among others. Labour cost minimization is, however, a commonly cited reason for both phenomena (Johansson et al., 2019; Dachs et al., 2019). This study analyses the cross-border impact of automation-driven reshoring through the cost-channel. 2 The Robotreport indicated that Adidas eventually shut down the Speedfactories, citing that it was still more profitable to operate in Asia since over 90% of their products are manufactured there. Available at: https://www.therobotreport.com/adidas-closing-german-us-robot-speedfactories/ 3 surmised by World Bank (2016), developing countries could still be negatively impacted by automation efforts in industrialized economies. The academic literature is trailing anecdotal evidence with regards to studying the potential impacts of automation spillover through reshoring, despite the need for a rigorous appreciation. In the International Business (IB) literature, where the study of offshoring is not new, but reshoring is relatively more recent, most of the studies have ignored technological change as one of the key drivers of reshoring (Dachs et al., 2019). Likewise, in the fields of the economics of technological change and international economics, research on the automation-reshoring relationship is thin albeit budding. Carbonero et al. (2018), one of the early studies, found that the adoption of robots in advanced countries has reduced offshoring and further decreased employment in emerging economies by 5 percent. However, robot exposure only accounts for a limited proportion of existing automation technologies. Moreover, most of the existing papers do not fully capture reshoring, which represents a generic relocation of production activities including but not limited to backshoring (production repositioning back home) (Albertoni et al., 2017; Di Mauro et al., 2018). This paper bridges these gaps by analysing the effect of automation on the global reorganization of production activities between advanced and developing economies, thereby addressing the phenomena of offand re-shoring. To conduct the analysis, the study extends the World Trade Model (WTM) proposed by Duchin (2005) to include non-tradable sectors and considers changes in the optimal international allocation of production activities under different (automation) scenarios. The WTM compares production costs globally and (re-)allocates production at the sector level under free trade. We construct the automation scenarios by estimating the labourreducing risk of automation at the sector level. Our primary data sources are the World Input-Output Database (WIOD) and the International Assessment of Adult Competencies (PIAAC). The paper is organized as follows. Section 2 reviews the literature on reshoring and its linkages with automation. Section 3 of the study presents our extended WTM model, as well as its empirical application and data sources. Section 4 reports and discusses the results and section 5 provides the summary and conclusion of the analysis. 4 2. Related Literature: Offshoring, Reshoring, and Automation The literature on reshoring is recent and derives from older literature on offshoring. It is, therefore, worth succinctly introducing the offshoring literature. The theoretical roots that explain the motivations of offshoring date to the seminal contribution by Coase (1937). The paper explains why firms exist, arguing that they do because markets are not without transaction costs. The presence of transaction costs necessitates the existence of firms to offer an alternate means of allocating resources and organizing production. Buckley and Casson (1976) extended the previous work to the multinational firm and introduced the internalization theory. The central idea is that firms “internalize” (i.e. conduct certain value chain activities internally) to exploit and develop firmspecific advantages in knowledge and intermediate products instead of relying on markets as an external coordinating mechanism. From this viewpoint, variations in value chain activities are influenced by underlying factors such as changing costs in the global economic system (Casson, 2013). Thus, the internalization theory adopts a systemic view of global product fragmentation. The eclectic paradigm by Dunning (1977;1980), also known as the OLI framework, focuses more on heterogeneous firm characteristics and builds on the internalization theory by considering three main advantages. First, ownership advantages (O) denote the extent to which a firm holds (or can own) assets that its (potential) competitors do not hold, such as the ownership of capital, patents, or intellectual property rights. Second, location advantages (L) relate to the benefits of complementing the assets owned with the resources or conditions abroad such as lower labour costs, favourable government policies, and nearness to markets. Third, internalization advantage (I) refers to the advantage of making use of the assets rather than selling or leasing them. 3 Put simply, Dunning’s theory stipulates that the OLI advantages are linked and essential for the internationalization of production. Greater ownership advantages tend to increase the internalization advantage, and these advantages combined with more attractive foreign location advantages provide a strong incentive to offshore. Empirical work in the international business (IB) strand of literature has closely followed the theoretical contributions by investigating the main motivations of offshoring, with an emphasis on estimating their effect sizes via regression frameworks. Through probit analyses based on the German Manufacturing Survey, for 3 The study of offshoring is well-established with multiple theories proposed to explain the phenomenon based on the relative costs and benefits of offshoring, including internalization theory, OLI framework, resource-based view, and transaction cost economics (TCE). However, internalization theory and OLI theory are the most widely used theoretical lenses. It is also worth noting that, while TCE and internalization theory are similar, the latter is broader and further assumes awareness of existing costs. Bounded rationality plays a more important role in TCE (Foss, 2003; Delis et al, 2019; Dachs et al, 2019). 5 instance, Kinkel and Maloca (2009) found that, “the reduction of labour costs is the most important single motive for production offshoring activities”. Generally, previous studies on the organization of global production networks tended to focus on offshoring as a uni-directional phenomenon. However, recent reports have indicated a possible counter-trend whereby firms reverse their offshoring decision and relocate production networks or reshore (Economist, 2013; 2017). The IB literature noted the reversal of offshoring early, but it was viewed as stemming from managerial mistakes: miscalculations of the risks of offshoring (Kinkel and Maloca, 2009). 4 Reshoring was, however, subsequently recognized as the change of a completely rational offshoring decision, motivated by changes in the host or home country conditions (Di Mauro et al., 2018). The motivations for reshoring are discussed within the existing theories of offshoring. Through the perspectives of the internalization and OLI theories, for instance, reshoring is explained in terms of the declining advantages of ownership, location, and internalization. Concerning the location advantages more specifically, the rise of labour costs in developing countries has been identified as a major driver of reshoring due to an emerging disincentive to operate certain value chain activities in these countries (Casson, 2013; Dachs et al., 2019). These studies also generally support the view that different sectors or firms in a given country may be simultaneously offshoring and backshoring at any given time. Thus, determining the magnitude of ultimate impacts on an economy depends on the relative strengths of offshoring and backshoring activities. Some recent studies have confined reshoring to backshoring or back-reshoring, which refers to the relocation of production activities back to the home country (Delis et. al., 2019; Faber, 2020). However, reshoring is broader than backshoring; it can additionally involve a relocation of value chain activities from a host country to another country other than the home country (Albertoni et al., 2017; Di Mauro et al., 2018). Following the earlier work on offshoring, the international business (IB) strand of literature has centred empirical analyses on identifying the most important drivers of reshoring. Like the offshoring literature, labour cost differences stand out as one of the most important considerations in the newer reshoring literature (Johansson et al., 2019; Delis et. al., 2019). Technological change has been largely ignored as a driver of reshoring in the IB strand of literature. Dachs et al., 2019 stress this point and argue, consistent with the reported anecdotal evidence, that Industry 4.0 will likely positively 4 This perspective is rooted in Transaction Cost Economics theory as it attributes a reduction in offshoring or reshoring to bounded rationality in determining foreign location costs. 6 affect reshoring by making labour arbitrage less attractive as factor cost advantages in foreign locations are counteracted. 5 The economics literature is witnessing increasing interest in linking automation technologies to reshoring and employment. This attention is mainly driven by concerns that the adoption of new automation technologies in the advanced world could drive reshoring, as labour cost advantages in developing economies that serve as host countries are eroded, which could in turn adversely impact production and employment in the developing world. De Backer et al. (2018) described this phenomenon as botsourcing. The paper concluded that the use of robots is not yet triggering backshoring based on their full sample covering 2000 to 2014. Conversely, Carbonero et al. (2018) found that offshoring to emerging economies decreased because of automation efforts in advanced economies. Due to the focus on robots, these studies do not substantially account for disembodied technological change present in the advancement of algorithms and software. Besides, most of the recent studies employ regression frameworks involving some measure of reshoring or offshoring such as the share of imported intermediate inputs in the same industry over total non-energy intermediates (Feenstra and Hanson 1996; Faber 2020). Krentz et al. (2021) also proposed a new reshoring measure based on the premise that earlier measures merely capture a reduction in offshoring, which is not necessarily equivalent to reshoring. The paper measures reshoring as the timedifference between the ratio of domestic to foreign inputs over a present and past period. However, the measure equates reshoring to backshoring and thereby does not fully address reshoring. Besides, the use of regression approaches does not provide a system’s view that incorporates critical interdependences within the global economy, such as the coexistence of offshoring and backshoring. The World Trade Model (WTM) proposed by Duchin (2005) possesses properties useful for analysing the global consequences of automation-driven reshoring, taking into account relevant interdependences during trade. It is a linear program that determines worldwide outputs and factor inputs (in the primal program) and world prices (in the dual program) based on each country’s comparative advantage in the global economy. The model is an extension of the 2-country, 2-good, 2-factor World Model of Leontief et al. (1977) to an m-country, n-good, k-factor case, and is an alternative to more complex models (such as Computable General Equilibrium models that requires additional assumptions to be operationalized). Strømman and Duchin (2006) extended the model to determine bilateral trade flows. Several other extensions and applications have been suggested, such as the use of the model to analyse scenarios about potential changes in the future (Dilekli and Cazcarro, 2019; 5 Another technology-backshoring channel that authors point out is flexibility. In addition to the labour cost channel, they argue that Industry 4.0 technologies promise flexibility in production which could induce firms to backshore or reshore close to advanced-country customers. 13 4. Empirical Findings This section presents the findings of our analysis based on the approach explained in the previous section. As explained in detail in Appendix C, our data are taken from an actual description of the global economy (the WIOD database). This describes a situation in which trade takes place, and the appendix explains how we calculate autarky values of the variables based on these data. However, trade in this real-world situation is not as far-developed as the WTM (or WTMNT) linear programs predict, for example, because the models do not include any transportation or other trading costs. Therefore, the autarky (no-trade) linear program as well as the WTM and WTMNT linear programs are abstractions that are necessarily different from the actual data observed in the WIOD. The main focus of our analysis is the impact of automation on the distribution of production across the world. For this, we specify two automation scenarios, which entail changing the labour coefficients according to the automation risk estimates that were discussed above. In the first automation scenario, only the developed world implements automation technology, and hence the labour coefficients only change in the developed countries. In the second automation scenario, automation is global, i.e., all countries have lower labour coefficients. The simplest way to implement the three scenarios (baseline and automation) in the three models (No Trade model, WTMNT, and WTM) is to impose the labour and capital endowments as they appear in the data. In this case, either capital or labour may provide a constraint in each particular country, but it is generally the case that (global) final demand is satisfied while some countries do not become constrained in any production factor. The reason is that each of the three models assumes that the final demand is exogenous and equal to what we observe in the real-world inputoutput table, and those values correspond to a situation in which no country produces at full capacity. This is the first way in which we solve our models, but we only use this to broadly check the consistency of the solutions obtained under autarky, WTM, and WTMNT. Table 2 presents the optimal values for these model solutions. The left column uses factor coefficients as they are observed in the data, while the middle column assumes that automation takes place in developed countries, and the rightmost column assumes that automation takes place in all countries. In each of the three columns, the results confirm Duchin’s basic result that trade permits countries with comparative advantage to produce goods to satisfy worldwide consumption requirements at a lower global cost than autarky. This is seen from the fact that the bottom line (WTM) in the table has lower values than the top line (no trade, or autarky). The WTMNT is an intermediate case where some but not all of the benefits of trade are reaped. Furthermore, automation reduces global factor costs by improving productivity and thus reducing the cost of labour in output. 14 Table 2: Global Factor Use Cost (US$ millions), Baseline and Automation Scenarios Model Baseline Developed world Automation Global Automation No Trade Model 74,366,875 60,592,661 51,120,128 World Trade Model with Non-Tradables 49,897,034 38,361,280 35,259,063 World Trade Model 34,884,787 23,956,648 21,521,377 Note: In both baseline and automation scenarios, global factor use cost of the World Trade Model with Non-Tradables lies between the Duchin’s No Trade Model and World Trade Model. A second possible way to solve the models assumes that each country produces at maximum capacity of at least one production factor. This is achieved by progressively increasing global final demand, by multiplying the final demand vector by ever-larger scalar numbers, until a solution to the linear program becomes infeasible. Note that this means that final demand increases everywhere in the world by the same multiplicative factor. At the maximum feasible multiplication factor, at least one of the endowment constraints for labour and capital is binding in all countries (i.e., factor use equals factor endowments). We solved the model in this way, but do not document the results to save space (these results are available on request). Furthermore, we consider a third way to solve the models, which is focused on the implications of labour rather than capital being the scarce factor, under a situation of full employment (i.e., a binding labour constraint) in every country. This is implemented by first increasing capital endowments in every country to (very) large amounts, to ensure that capital never becomes the binding factor, and then progressively increasing final demand until labour becomes binding in every country, in the same way as before. Obviously, this yields a hypothetical solution to the linear programs, not only in the sense that full employment is reached globally, but also in the sense that final demand is at the maximum possible value given the state of technology in every country. Solving the models in this way yields a very specialized international distribution of labour. In the baseline scenario, in 19 of the 44 tradeable sectors, production takes place in just one country, 14 other sectors have production concentrated in 2 countries, and 5 sectors concentrated in 3 countries, leaving 6 sectors with more than 3 countries. The largest number of countries in which production takes place for a single sector is 11, which happens in wholesale trade. 15 Table 3: Shares of Global GDP and share of rents in GDP, Baseline Scenario Shares of global total Shares of GDP in country group Actual GDP (data) Factor price income Scarcity rents Benefit of trade rents Total GDP Scarcity rents Benefit of trade rents NA&A 0.276 0.241 0.257 0.000 0.246 0.968 0.000 EU 0.240 0.225 0.221 0.517 0.233 0.878 0.090 H’Asia 0.084 0.096 0.073 0.000 0.070 0.955 0.000 L’Asia 0.178 0.175 0.173 0.158 0.173 0.930 0.037 LAM 0.045 0.052 0.087 0.230 0.092 0.880 0.101 ROW 0.176 0.211 0.189 0.094 0.186 0.942 0.021 Table 3 documents the shares of global GDP of 6 country groups in these optimizations, as well as the share of rents related to the shadow prices in each of the country groups. In this case, i.e., the baseline scenario without labour saving due to automation, the maximum feasible multiplication factor for final demand was 1.606. The table also documents the share of global GDP of the country groups in the actual data. The country groups are as follows: • North America (comprising USA and Canada) and Australia (NA&A); • Europe (EU): Austria, Belgium, Bulgaria, Switzerland, Cyprus, Czech Republic, Germany, Denmark, Spain, Estonia, Finland, France, United Kingdom of Great Britain and Northern Ireland, Greece, Croatia, Hungary, Ireland, Italy, Lithuania, Luxembourg, Latvia, Malta, Netherlands, Norway, Poland, Portugal, Romania, Slovakia, Slovenia, and Sweden; • High-income Asia (H’Asia), which includes Taiwan, Japan, and Korea; • Lower-income Asia (L’Asia), which includes China, India, and Indonesia; • Latin America (LAM), covering Brazil and Mexico; • Rest of the World (ROW): Russian Federation, Turkey, and WIOD-ROW. We note that in Table 3, the optimization assigns a much higher share of GDP to LAM as compared to the actual data, and a slightly higher share to ROW. The 4 other country groups get a smaller share of global GDP than what they have in the actual data. In this particular setup, i.e., production at global full employment, scarcity rents (for labour, as capital is never scarce) are the largest share of GDP in each region, and benefit-of-trade rents are much smaller shares, which are highest in EU and LAM. Factor payments at exogenous factor prices represent a very small share of GDP. Next, we present the results of the optimizations where labour saving is introduced. We still focus on the results corresponding to the full employment setup, 16 as in Table 3. If automation/labour saving is introduced in the developed countries, which are the NA&A, EU, and H’ASIA groups, plus the Russian Federation (part of ROW), we are able to multiply final demand by 2.440. If labour saving is global, then we are able to multiply by 2.492. Table 4 shows the shares of production value of each country group in each broad sector and the changes of this in the automation scenarios relative to the baseline. The table only includes sectors that were treated as tradeable in the analysis. Table 4: Shares of sectoral GDP and changes relative to Baseline Scenario A: Agriculture, Forestry, and Fisheries B: Mining C: Manufacturing JK: Trade, Hotels, Restaurants MN: Services Baseline NA&A 95.6 12.9 1.3 0 0 EU 4.4 36.4 18.9 27.1 0.6 H’ASIA 0 0 20.7 17.4 3.6 L’ASIA 0 28.4 51.8 20.3 28.9 LAM 0 22.3 0 26.8 43.1 OTH 0 0 7.3 8.3 23.7 Developed countries automation NA&A 77.1 (-18.5) 13.6 (0.8) 24 (22.7) 40.3 (40.3) 0 (0) EU 22.9 (18.5) 8.8 (-27.7) 46.8 (27.9) 43.2 (16.1) 0.6 (0) H’ASIA 0 (0) 0 (0) 19.1 (-1.6) 0 (-17.4) 4.8 (1.1) L’ASIA 0 (0) 37.7 (9.4) 10.1 (-41.6) 9.7 (-10.6) 43.8 (14.9) LAM 0 (0) 24.5 (2.2) 0 (0) 0 (-26.8) 3.7 (-39.4) OTH 0 (0) 15.3 (15.3) 0 (-7.3) 6.8 (-1.6) 47.1 (23.3) Global automation NA&A 80.4 (-15.2) 0 (-12.9) 22.7 (21.4) 10.7 (10.7) 4.6 (4.6) EU 19.6 (15.2) 2.5 (-34) 42.8 (23.9) 12.6 (-14.5) 12.4 (11.8) H’ASIA 0 (0) 30 (30) 18 (-2.7) 0 (-17.4) 0 (-3.6) L’ASIA 0 (0) 67.5 (39.1) 9 (-42.8) 48.6 (28.3) 40.4 (11.5) LAM 0 (0) 0 (-22.3) 7.5 (7.5) 14 (-12.7) 38.2 (-4.9) OTH 0 (0) 0 (0) 0 (-7.3) 14 (5.7) 4.4 (-19.3) Note: Numbers between brackets are the changes relative to the baseline scenario In the agriculture, forestry, and fisheries sector, production is concentrated in the NA&A group in the baseline, with a small part in EU, but nothing in the rest of the world. Automation shifts a part of the production from NA&A to EU, with little difference between the two automation scenarios. 17 In Mining, results are biased because we do not consider mineral resources as a separate production factor. Production is fairly spread out in the baseline scenario, and with the automation in the developed scenario, EU loses out an important part of its share, which is large in the baseline. In the global automation scenario, LAM and NA&A also lose, and L’ASIA becomes the largest producer. In manufacturing, L’ASIA is by far the largest producer in the baseline, but it loses out most of its share in the developed world automation scenario, mostly to EU and NA&A. In the global automation scenario, L’ASIA loses an even larger part of its baseline share. In Trade, Hotels, and Restaurants, LAM, L’ASIA, and H’ASIA are the large losers in the developed world automation scenario, while NA&A and EU gain. In the global automation scenario, L’ASIA becomes the winner instead of EU and NA&A. Finally, in the business services sector, LAM loses in the developed world automation scenario, with L’ASIA and ROW as the gainers. In the global automation scenario, ROW is the only big loser. Table 5 presents the GDP data for the developed world automation scenario, in the form of differences to the baseline of Table 3. Hence in each of the columns for shares of the global total, the numbers in Table 5 will add to zero. The table shows that the developed countries will gain significantly in GDP, with EU gaining the most (6.5 points), and NA&A and H’ASIA gaining marginally less. L’ASIA loses the most (7 points), closely followed by ROW. In terms of the components of GDP, EU gains most in terms of the benefit-of-trade rents, which become almost one-third of the total GDP in the EU group. Table 5: Shares of Global GDP and share of rents in GDP, Differences of Developed world Automation scenario to Baseline Scenario Shares of global total Shares of GDP in country group Factor price income Scarcity rents Benefit of trade rents Total GDP Scarcity rents Benefit of trade rents NA&A 0.045 0.074 0.000 0.049 -0.040 0.000 EU 0.018 0.008 0.404 0.065 -0.243 0.214 H’Asia -0.009 0.058 0.044 0.048 -0.046 0.037 L’Asia -0.038 -0.062 -0.158 -0.070 -0.029 -0.037 LAM -0.010 -0.012 -0.230 -0.026 0.073 -0.101 ROW -0.005 -0.067 -0.060 -0.066 -0.098 0.008 In Table 6, we document the results of the global automation scenario. Again, the table reports differences from the baseline scenario of Table 3. In terms of total GDP, the ROW country group is the big winner (+6 points). H’ASIA also gains, but marginally, and EU stays virtually constant, but at a very small positive difference. The other country groups lose in terms of the global GDP share, with L’ASIA as the largest 18 loser (-4.4 points). However, the benefits of trade rents shift towards L’ASIA in this case. Scarcity rents now fall very significantly as a share of GDP in all country groups. Table 6: Shares of Global GDP and share of rents in GDP, Differences of global Automation scenario to Baseline Scenario Shares of global total Shares of GDP in country group Factor price income Scarcity rents Benefit of trade rents Total GDP Scarcity rents Benefit of trade rents NA&A 0.057 -0.209 0.000 -0.019 -0.912 0.000 EU 0.040 -0.081 -0.068 0.003 -0.725 -0.052 H’Asia -0.008 0.025 0.000 0.018 -0.671 0.000 L’Asia -0.025 -0.135 0.387 -0.044 -0.853 0.048 LAM 0.002 0.049 -0.224 -0.017 -0.407 -0.100 ROW -0.065 0.352 -0.094 0.060 -0.369 -0.021 5. Summary and Conclusions The objective of this study is to use an input-output-based model of global trade to investigate the possible consequences of the introduction of automation technologies on global development. The main results are obtained under the assumption that labour is the scarce production factor (and that capital is abundant everywhere in the world). Broadly, the analysis finds that the adoption of new automation technologies in advanced economies is likely to lead to significant relocation of production, including a fair amount of reshoring of production activities away from developing countries, back to developed countries. As the main part of the analysis assumes full employment, the consequences of this relocation are experienced in the form of a changing global distribution of income (GDP). The results revealed that lower-income Asia is likely to be the most adversely impacted developing region in a scenario where automation takes place in the advanced world only. In this case, the advanced regions of the world gain a significant share of global GDP. The loss in lower-income Asia was especially manifest in manufacturing. When automation takes place globally, i.e., also in developing countries, the “rest of the world” category appears as a main winner. However, again, lower-income Asia loses out the most in terms of the share of global GDP, with manufacturing likely to experience the hardest hit. We also find that advanced economies adopting automation technologies would rather benefit via growth in incomes including positive factor earnings. 19 REFERENCES Acemoglu, D., & Restrepo, P. (2018). Demographics and Automation. Boston University - Department of Economics - The Institute for Economic Development Working Papers Series dp-299, Boston University - Department of Economics. Albertoni, F., Elia, S., Massini, S., & Piscitello, L. (2017). The reshoring of business services: Reaction to failure or persistent strategy? Journal of World Business, 52, 417– 430. Arntz M., Gregory T., & Zierahn U. (2017). Revisiting the risk of automation. Journal of Economic Letters 159: 157‐160. Autor, D., and Dorn, D. (2013). The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market. American Economic Review 2013, 103(5): 1553–1597 http://dx.doi.org/10.1257/aer.103.5.1553. Buckley, P. J., & Casson, M. (1976). The future of the multinational enterprise. Basingstoke: Macmillan. Cadarso, M.A., Gómez, N., López, L.A., and Tobarra, M.A. (2012). Offshoring components and their effect on employment: firms deciding about how and where. Journal of Applied Economics 44, 1009–1020. Carbonero, F., Ernst. E., & Weber. E. (2018). Robots worldwide: the impact of automation on employment and trade. International Labour Office (ILO), Research Department Working Paper No. 36. Casson, M. (2013). Economic Analysis of International Supply Chains: An Internalization Perspective. Journal of Supply Chain Management Volume 49, Issue 2: 8-13 Cavenaile, L. (2021). Offshoring, computerization, labor market polarization and top income inequality. Journal of Macroeconomics 69 (2021) 103317. Coase, R. (1937). The nature of the firm. Economica, 4(16), 386–405. Dhanaraj, C., & Parkhe, A. (2006). Orchestrating innovation networks. The Academy of Management Review, 31(3), 659–669. Dachs, B., Kinkel, S. and Jägerc, A. (2019). Bringing it all back home? Backshoring of manufacturing activities and the adoption of Industry 4.0 technologies. Journal of World Business 54 (2019) 101017. De Backer, K., DeStefano, T., Menon, C., Suh, J. R. (2018). Industrial robotics and the global organisation of production, OECD Science, Technology and Industry Working Papers, 2018/03, OECD Publishing Paris. Delis, A., Driffield, N. & Temouri, Y. (2019). The global recession and the shift to reshoring: Myth or reality? Journal of Business Research 103 (2019) 632–643. 20 Di Mauro, C., Fratocchi, L., Orzes, G., & Sartor, M. (2018). Offshoring and backshoring: A multiple case study analysis. Journal of Purchasing and Supply Management 24 (2018) 108–134. Dilekli, N. and Cazcarro, I. (2019). Testing the SDG targets on water and sanitation using the world trade model with a waste, wastewater, and recycling framework. Journal of Ecological Economics Volume 165, 106376. Duchin, F. (2005) A world trade model based on comparative advantage with m regions, n goods, and k factors, Economic Systems Research, 17, pp. 141–162. Dunning, J. H. (1977). Trade, Location of Economic Activity and the Multinational Enterprise: The Search for an Eclectic Approach. In The International Allocation of Economic Activity, edited by B. Ohlin, P. O. Hesselborn, and P. M. Wijkman, 395–418. London: Macmillan. Dunning, J. H. (1980). Towards an eclectic theory of international production: some empirical tests. Journal of International Business Studies, 11 (1), 9–31. Dunning, J. H. (2000). The eclectic paradigm as an envelope for economic and business theories of MNE activity. International Business Review 9 (2000) 163–190. Faber, M. (2020). Robots and reshoring: Evidence from Mexican labor markets. Journal of International Economics 127 (2020) 103384. Feenstra, R. C., and Hanson, G. H. (1996). Globalization, outsourcing, and wage inequality. American Economic Review 86(2), 240-245. Feenstra, R. C., Hanson, G. H. (1999). The Impact of Outsourcing and High-Technology Capital on Wages: Estimates for the United States, 1979-1990, Quarterly Journal of Economics, 114, 907-941. Feenstra, R.C. (2017). Statistics to Measure Offshoring and its Impact. NBER Working Papers 23067, National Bureau of Economic Research, Inc. Foss, N. (2003). Bounded rationality in the economics of organization: Much cited and little used. Journal of Economic Psychology, 24(2), 245–64. Foster-McGregor, N., Nomaler, O. and Verspagen, B. (2019). Job Automation Risk, Economic Structure and Trade: A European Perspective. Foster-McGregor, N., Poeschl, J and Stehrer, R. (2015). Offshoring and the Elasticity of Labour Demand. Open Econ Rev 27, 515–540 (2016). https://doi.org/10.1007/s11079-0159384-6 Fratocchi, L., Di Mauro, C., Barbieri, P., Nassimbeni, G., Zanoni, A. (2014). When manufacturing moves back: concepts and questions. Journal of Purchasing & Supply Management 20 (1), 54–59. Frey, C. and Osborne, M. (2017). The Future of Employment: How Susceptible are Jobs to Computerization? Technological Forecasting and Social Change. 21 Fuster, B., Lillo-Banuls, A. & Martínez-Mora, C. (2019). The effects of service offshoring on employment. Structural Change and Economic Dynamics 51 (2019) 529–538. Gardberg, M., Heyman, F., Norbäck, P., and Persson, Lars. (2020), "Digitization-Based Automation and Occupational Dynamics". Economics Letters 189. Günlük-Senesen, G. and Bates, J.M. (1988) Some Experiments with Methods of Adjusting Unbalanced Data Matrices. Journal of the Royal Statistical Society, Series A, 151, 473–490. Hijzen, A. and Swaim, P. (2010). Offshoring, labour market institutions and the elasticity of labour demand. European Economic Review 54 (2010) 1016–1034 Howard, A., & Borenstein, J. (2020). AI, robots, and ethics in the age of COVID-19. Ivanov, S., Kuyumdzhiev, M., and Webster, C. (2020). Automation fears: Drivers and solutions. Journal of Technology in Society, 63 (2020) 101431. Johansson, M. and Olhager, J. (2018). Comparing offshoring and backshoring: The role of manufacturing site location factors and their impact on post-relocation performance. International Journal of Production Economics 205 (2018) 37–46. Johansson, M., Olhager, J., Heikkiläb, J. and Stentoftc, J. (2019). Offshoring versus backshoring: Empirically derived bundles of relocation drivers, and their relationship with benefits. Journal of Purchasing and Supply Management 25 (2019) 100509. Kinkel, S., Maloca, S., 2009. Drivers and antecedents of manufacturing offshoring and backshoring – a German perspective. J. Purch. Supply Manag. 15 (3), 154–165. Krenz, A., Prettner, K., and Strulik, H. (2021). Robots, reshoring, and the lot of lowskilled workers. European Economic Review 136 (2021) 103744. Leontief, W., and Duchin, F. (1984). The Impacts of Automation on Employment, 19632000. Retrievable at: https://files.eric.ed.gov/fulltext/ED241743.pdf. Leontief, W., Carter, A. P. and Petri, P. (1977). The Future of the World Economy (New York: Oxford University Press). Li, X., Hui, E. C., Lang, W., & Zheng, S., Qin, X. (2020). Transition from factor-driven to innovation-driven urbanization in China: A study of manufacturing industry automation in Dongguan City. China Economic Review 59 (2020) 101382. Luenberger, D.G. (1989). Linear and Nonlinear Programming. (Reading, Massachusetts, Addison-Wesley Publishing Co.). Nedelkoska, L. and Quintini G. (2018). Automation, skills use and training. In OECD Social, Employment and Migration Working Papers, No. 202, OECD Publishing. Nii-Aponsah, H. (2022). Automation exposure and implications in advanced and developing countries across gender, age, and skills. UNU-MERIT Working Papers ISSN 1871-9872. Prettner, K. and Bloom, D. (2020). Automation and Its Macroeconomic Consequences: Theory, Evidence, and Social Impacts, 1st Edition. 22 Rocco, M, V., Golinucci, N., Ronco, S.M., and Colombo, E. (2020). Fighting carbon leakage through consumption-based carbon emissions policies: Empirical analysis based on the World Trade Model with Bilateral Trades. Rodrik, D. (1997). Has globalization gone too far? Institute for International Economics, Washington, DC. Senses, M. Z. (2010). The effects of offshoring on the elasticity of labor demand. Journal of International Economics, 81:89–98 Strømman, A. H. and Duchin, F. (2006). A world trade model with bilateral trade based on comparative advantage, Economic Systems Research, 18:3, 281-297, DOI: 10.1080/09535310600844300 Temurshoev, U., Miller, R.E. & Bouwmeester, M. C. (2013). A Note on The Gras Method. Journal of Economic Systems Research, 25:3, 361-367. The Economist. (14th Jan 2017). Adidas’s high-tech factory brings production back to Germany. Business Report. The Economist. (17th Jan 2013). Reshoring Manufacturing: Coming Home. Special report. Ten Raa, T. and Mohnen, P. (2001) The Location of Comparative Advantages on the Basis of Fundamentals Only, Economic Systems Research, 13:1, 93-108, DOI: 10.1080/09535310120026265 Timmer, M. P., Dietzenbacher, E., Los, B., Stehrer, R. and de Vries, G. J. (2015). "An Illustrated User Guide to the World Input–Output Database: the Case of Global Automotive Production", Review of International Economics., 23: 575–605 World Bank. (2016). World Development Report 2016: Digital Dividends. 29 Hungary, Ireland, Italy, Japan, Republic of Korea, Lithuania, Luxembourg, Latvia, Malta, Netherlands, Norway, Poland, Portugal, Romania, Russian Federation, Slovakia, Slovenia, Sweden, Taiwan, and the United States. Furthermore, the study uses PIAAC data to estimate the risk of automation needed to construct the automation scenarios, which are compared to the baseline scenarios. The PIAAC dataset, created by the OECD, is based on a survey that provides information on tasks that workers perform and the frequency with which they undertake them. Although the survey includes about 37 countries across three (3) rounds between 2011 and 2019, we use only countries with publicly available 4-digitlevel ISCO-08 job codes and estimate the automation risks for 2-digit sectors, consistent with WIOD. The sample entails 20 advanced countries based on the 20202021 World Bank Income Classification system. The advanced countries include the following: Belgium, Chile, Czech Republic, Denmark, France, Greece, Hungary, Israel, Italy, Japan, the Republic of Korea, Lithuania, Netherlands, New Zealand, Poland, Russia, Slovakia, Slovenia, Spain, and the UK. We, however, compute the automation risks for countries in WIOD, which exclude Chile, Israel, and New Zealand. Variables From the raw input-output data, we first compute intermediate input coefficients in the usual way, yielding a square matrix with 44×56 rows and columns. Next, we calculate the autarky intermediate input coefficients (𝐴𝑖), which, for each country including ROW, forms a square matrix with 56 rows and columns. To calculate these, we sum, in each column of the original (large) intermediate input coefficients matrix, the cells belonging to the same sector. This yields 56 values, which form the column of the autarky intermediate input coefficients matrix of the column-country in the original intermediate inputs matrix. For instance, if sector 1 in Germany needed to use $10 million worth of sector 2 intermediate input domestically while importing (from the other 43 countries) $2 million worth of sector 2 intermediate input to produce $24 million worth of sector 1 output, then the autarky intermediate input coefficient in row 2, column 1 becomes (10+2)24 ⁄. Final demand under autarky (𝑦𝑖) is constructed using the same approach as autarky intermediate input coefficients. We first aggregate the 5 WIOD components of final demand (final consumption expenditure by households, final consumption expenditure by non-profit organisations serving households, final consumption expenditure by government, gross fixed capital formation, and changes in inventories and valuables) into a single vector. Then for every country, we aggregate all values in this vector for each of the 56 sectors. The analysis further needs the factor inputs per unit output (𝐹𝑖), which we calculate by dividing labour (number of persons engaged) and capital input quantities from the Socio-economic accounts (SEA) database by the corresponding country- 30 sector outputs in the WIOT. Since the SEA database does not contain data for the Rest of the World (ROW), we calculate the ROW factor coefficients as the average over the corresponding developing countries-sectors (China, Brazil, India, Indonesia, Mexico, and Turkey). Thus, ROW is treated as primarily reflecting the developing world. Moreover, because capital is measured in national currencies, the study first adjusts it to US dollars by dividing capital by the PPP exchange rates from the International Comparison Program (2017). Capital is adjusted further by multiplying the PPPadjusted capital by economy-wide capacity utilization rates in 2014 from Eurostat and the OECD statistics (see Appendix B for these data). This is to ensure that capital coefficients reflect capital use per unit output rather than capital stock per unit output. Factor endowments (𝑓𝑖) are calculated as follows. The aggregate capital endowments are the total PPP-adjusted capital from the WIOD SEA (without the utilization rate adjustment), whereas the implied aggregate labour endowments are determined by dividing the total number of persons engaged in each country from the WIOD SEA by 1 minus the unemployment rate from WDI (2021). 11 To determine ROW endowments, we first calculate the total ROW labour and capital use by multiplying the ROW factor input coefficients by output and summing them up. The employment and utilization rates for ROW are the averages over the rates for the developingregion sample: Brazil, China, Indonesia, India, Mexico, and Turkey. Finally, the analysis needs the factor input prices (𝜋𝑖). The wage rate is calculated by dividing total labour compensation (converted to $ using the WIOD exchange rates) by the total number of persons engaged per country (labour input). The price of capital is determined likewise; that is, the capital cost is first converted by the WIOD exchange rate to US dollars and subsequently divided by the total PPP-adjusted capital stock (in this case unadjusted for utilization) in the country. Input prices for the ROW are the averages over the developing regions in the sample. Implementation We observed which fraction of the gross output of each sector is exported in the WIOT, and based on this, designated twelve sectors as non-tradable: electricity supply (D35), water supply (E36), construction (F), postal activities (H53), accommodation and food (I), real estate activities (L68), public administration and defence (O84), education (P85), health (Q), other services (R_S), activities of households as employers (T), and activities of extraterritorial organizations (U). Note that none of these are exactly non-tradable, i.e., we observe some trade even in these sectors, but trade is minor as compared to the tradeable sectors. 11 Our analysis compares the computed labour and capital endowments per country with their corresponding No Trade endowments and uses the maximum endowments (in each scenario) to satisfy the restriction: fi ≥ fnt,i. 31 Some other critical considerations before running the models are as follows. The WIOT contains some sectors that produce no gross output, for example, because these sectors are merged with other sectors for specific countries. We set the intermediate coefficients and final demands of these sectors to zero to assure that they also produce no output under both autarky (the No Trade) and trade (the WTMNT), and set the factor input coefficients to be prohibitively high to ensure that these sectors do not attract positive output in any of the optimization scenarios (we checked ex-post to make sure this indeed did not happen). We run the linear programming optimizations in Matlab using the linprog command for the baseline and automation scenarios. The UNU-MERIT WORKING Paper Series 2023-01 Can international mobility shape students' attitudes toward inequality? The Brazilian case by Cintia Denise Granja, Fabiana Visentin and Ana Maria Carneiro 2023-02 Demand-led industrialisation policy in a dual-sector small open economy by Önder Nomaler, Danilo Spinola and Bart Verspagen 2023-03 Reshoring, nearshoring and developing countries: Readiness and implications for Latin America by Carlo Pietrobelli and Cecilia Seri 2023-04 The role of product digitization for productivity: Evidence from web-scraping European high-tech company websites by Torben Schubert, Sajad Ashouri, Matthias Deschryvere, Angela Jäger, Fabiana Visentin, Scott Cunningham, Arash Hajikhani, Lukas Pukelis and Arho Suominen 2023-05 More than a feeling: A global economic valuation of subjective wellbeing damages resulting from rising temperatures by Stephan Dietrich and Stafford Nichols 2023-06 Was Robert Gibrat right? A test based on the graphical model methodology by Marco Guerzoni, Luigi Riso and Marco Vivarelli 2023-07 Predicting social assistance beneficiaries: On the social welfare damage of data biases by Stephan Dietrich, Daniele Malerba and Franziska Gassmann 2023-08 Related or unrelated diversification: What is smart specialization? By Önder Nomaler and Bart Verspagen 2023-09 Breach of academic values and digital deviant behaviour: The case of Sci-Hub by Giulia Rossello and Arianna Martinelli 2023-10 The effect of lobbies’ narratives on academics' perceptions of scientific publishing: An information provision experiment by Giulia Rossello and Arianna Martinelli 2023-11 Making impact with agricultural development projects: The use of innovative machine learning methodology to understand the development aid field by Lindsey Moore, Mindel van de Laar, Pui Hang Wong and Cathal O’Donoghue 2023-12 Green windows of opportunity in the Global South by Rasmus Lema and Roberta Rabellotti 2023-13 The green and digital transition in manufacturing global value chains in latecomer countries by Rasmus Lema and Roberta Rabellotti 2023-14 Impact of prestigious-STEM Education of corporate board members on innovation effort: Evidence from India by Rituparna Kaushik, Sourabh Bikas Paul and Danilo Spinola 2023-15 Reducing environmental impact through shared ownership: A model of consumer behaviour by Francesco Pasimeni and Tommaso Ciarli 2023-16 Welfare losses, preferences for redistribution, and political participation: Evidence from the United Kingdom’s age of austerity by Patricia Justino, Bruno Martorano and Laura Metzger 2023-17 Inequality, social mobility and redistributive preferences by Isabel Günther and Bruno Martorano 2023-18 Automation-induced reshoring and potential implications for developing economies by Hubert Nii-Aponsah, Bart Verspagen and Pierre Mohnen