NOVEMBER 2024 Migration, Automation, and the Malaysian Labor Market © 2024 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy, completeness, or currency of the data included in this work and does not assume responsibility for any errors, omissions, or discrepancies in the information, or liability with respect to the use of or failure to use the information, methods, processes, or conclusions set forth. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be construed or considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Attribution: Please cite the work as follows: Abdur Rahman, Amanina, Simon Bilo, Shreya Chatterjee and Natalie Fang Ling Cheng (2024). “Migration, Automation, and the Malaysian Labor Market” (November), World Bank, Washington, DC. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. Cover photo: © Abdul Razak Latif/Bigstock Credit for non-WB images: Bigstock Cover design and layout: Good News Resources Sdn Bhd/www.gnrsb.com Migration, Automation, and the Malaysian Labor Market: A summary of findings 1. Introduction Malaysia is set to achieve high-income status by There is strong sentiment in the country that hiring 2025, with rising educational attainment and a low-skilled migrant workers has adverse effects on rapidly aging population leading to questions on the economy. The 12th Malaysia Plan 2021-2025 has the role that migrant workers will play in the an explicit goal of reducing the number of migrant future of its development. According to official workers. Referring to low-skilled migrant workers statistics, more than 40 percent of the Malaysian as “foreign workers” and high-skilled workers as population aged 25 and older have at least a “expatriates”, the document states that: “Efforts will post-secondary education. This highly educated be undertaken to reduce the dependency on foreign population is arguably less willing to work in low- workers and promote greater local participation and certain mid-skilled jobs – particularly those in the labour market… This undertaking states the deemed to be dirty, dangerous, and difficult (3D plan to reduce the number of foreign workers within jobs) – creating a gap in the labor market that can five years either through employment of locals or potentially be filled by low-skilled migrant workers. automation. Employers hiring expatriates will also At the same time, Malaysia is rapidly aging. In 2020, be subjected to the specific policy” (EPU 2021, Malaysia passed the crucial milestone of becoming p. 10-28). Among the cited effects of hiring low- an aging society.1 It will continue to rapidly age in skilled migrant workers – that are arguably behind the next few decades at a pace as fast as Japan, the goal of reducing the number of migrant workers and significantly faster than the Western developed – includes the weakening of incentives for firms nations such as France, Australia, and the United to substitute labor for technology, or for greater States (World Bank 2020a). The rise in the aging value-adding activities from the employment of population means a decline in the working age high-skilled labor (Ang, Murugasu and Chai 2018). population, creating the need to utilize all possible Relatedly, there is a concern that the employment sources of labor to sustain the economy, including of low-skilled migrant workers is associated with women, youth, and migrant workers. lower productivity, and that it deters the creation of high-skilled and high-paying jobs (BNM 2021). Past studies show that migrant workers have The employment of low-skilled migrant workers played an important role in the Malaysian with low wages relative to Malaysian workers has economy in the last few decades. In a study of also been said to depress wages (Ang et al. 2018). migrant workers in Malaysia, the World Bank (2015) The recent Madani Economy Framework introduced found that they address labor market imbalances by the government echoes the 12th Malaysia Plan by filling labor shortages in low-skilled, labor- as well as these sentiments, stating the need to intensive sectors. Subsequently, findings show that reduce dependence on low-skilled migrant labor, they complement the majority of Malaysian workers which is said to contribute to low wage levels, and and contribute to creating jobs for higher-skilled reduces the morale and incentive for employers Malaysian natives, enabling Malaysians to specialize to innovate and improve competitiveness. At a and increase their wage premiums (Özden and societal level, a study conducted by the World Bank Wagner 2014; World Bank 2015). Migrant domestic and the University of Malaya finds that Malaysians workers have also arguably had a role to play in across regions, ethnicities, and income classes share supporting female labor force participation in increasing resentment toward low-skilled migrant Malaysia (KRI 2018). At the aggregate level, foreign workers (World Bank 2020b). labor in other contexts has been found to support domestic consumption and fuel economic growth However, there is limited to no evidence on the (Ahsan et al. 2014). adverse effects of hiring low-skilled migrant This publication was prepared by Amanina Abdur Rahman, Simon Bilo, Shreya Chatterjee, and Natalie Fang Ling Cheng, and forms part of a broader series on the Future of Work. Harry Moroz provided critical inputs on methodology. This publication benefited from inputs from Matthew Dornan and Alyssa Farha Jasmin. Peer-review comments were provided by Abla Safir and Wendy Cunningham. 1 An aging nation is defined as one with at least 7 percent of the population being age 65 and above (World Bank 2020a). 3 Migration, Automation, and the Malaysian Labor Market: A summary of findings workers. For instance, while it is true that wages A relevant question is whether low-skilled of low-skilled migrant workers are generally lower migration leads to less or slower automation. While than those of Malaysians, there is limited empirical this question has not been studied in the context of evidence on the causal relationship between the Malaysia and is not within the scope of this paper, employment of low-skilled migrant workers and the studies from other countries provide insight. There is wages of Malaysians. Studies that have attempted evidence that an increase in low-skilled labor, driven to analyze the relationship have found that there by migration, decreases the adoption of automation is a statistically significant negative impact on the technologies in the context of Denmark (Mann and employment of low-skilled migrant workers on the Pozzoli 2022) and the United States (Lewis 2011). In wages of lower-skilled Malaysians, but the impact is contrast, a study in the Italian context finds a positive small (Athukorala and Devadason 2012; Özden and causal relationship between low-skilled immigration Wagner 2014). Importantly, there is also evidence and capital intensity (Accetturo, Bugamelli and that the employment of migrant workers increases Lamorgese 2012). In sum, findings from other the employment of Malaysians, and increases the countries do suggest that low-skilled migration can wages of Malaysian workers with at least some indeed lead to less or slower automation, although secondary education (Özden and Wagner 2014; whether this holds for Malaysia remains to be seen. World Bank 2015). A study by Tan and Ng (2018) Independent of migration, the global adoption of found that an increase in foreign worker employment automation technologies has been slower than does not have a statistically significant impact on expected. The Future of Jobs Survey2 by the World the employment of Malaysian workers, or on overall Economic Forum shows that in 2020, 33 percent labor productivity. They also find no relationship of all business tasks were estimated to have been between foreign worker employment and capital automated (World Economic Forum 2023). In 2023, intensity (Tan and Ng 2018). only 34 percent of all business tasks were estimated While automation is often cited as the solution to have been automated, reflecting the slow pace for decreasing dependency on low-skilled migrant of automation (World Economic Forum 2023). In the workers in Malaysia, evidence of this potential same survey, employers have revised predictions for is also limited. Research findings on the impact of future automation from 47 percent automation by 2025 (in 2020), to 42 percent automation by 2027 automation on employment have been somewhat (in 2023) (World Economic Forum 2023). Among mixed (Acemoglu, Koster and Ozgen 2023). There the possible reasons for this include the costs of is indeed evidence that automation displaces low- acquiring and maintaining automation technologies skilled, blue-collar workers performing routine tasks relative to their benefits, the need for developing (Graetz and Michael 2018; Acemoglu and Restrepo laws and regulations to govern the adoption of 2020; Acemoglu et al. 2023). There is also evidence new technologies, and a general resistance to that employment in other types of jobs – particularly technological change (World Bank 2020b). those involving tasks that are complementary to automation technologies – increase as a result of This paper aims to strengthen the evidence automation (Humlum 2019; Acemoglu et al. 2023). base on the employment of migrant workers, Studies on the impact of automation in Spain (Koch, the employment of Malaysian workers, and Manuylov and Smolka 2021) and France (Aghion et al. the automatability of tasks performed within 2021) find that employment for all types of workers – occupations in Malaysia. It provides an updated including low-skilled workers – actually increases with analysis on the relationship between immigration and robot adoption. These studies show that while one labor market outcomes – including the employment can theoretically expect automation to decrease the rate, the unemployment rate, and wages – of demand for low-skilled workers (including migrant Malaysians. A major focus of this paper relates to workers), the impact of automation is ultimately the relative automatability of tasks performed by context dependent and is subject to the type of migrant workers and native workers, which has technology being adopted, among other factors. implications on the employment of both groups. 2 The Future of Jobs Survey collects data from 803 companies, collectively employing more than 11.3 million workers, across 45 countries from all world regions (World Economic Forum 2023). 4 Migration, Automation, and the Malaysian Labor Market: A summary of findings In short, this paper finds that there is a generally migrant workers) towards being more demand- positive relationship between the employment of driven would be a worthwhile endeavor. migrant workers and the employment outcomes of Malaysian workers. Further, migrant workers are The paper is organized as follows. The next more likely to perform tasks that are at automatable. section presents an overview of Malaysia’s At the same time, there is evidence that Malaysians migration policies. The third section describes the have greater access to jobs composed of tasks employment landscape in Malaysia and the role of that are less susceptible to automation, even migrant workers in the landscape, including findings when compared to migrant workers with similar on the relationship between the employment of qualifications. Importantly, given that Malaysian migrant workers and the employment outcomes workers make up a much larger share of the of Malaysian workers. The fourth section describes workforce compared to migrant workers, the the potential for automation in Malaysia. The fifth adoption of automation technologies will ultimately section presents an overview of the measurement have a substantive impact on Malaysians workers. of the automatability in this paper and calculates This potential impact can be mitigated by a strengthening of active labor market policies. the automatability of tasks performed within jobs Nonetheless, given that migrant workers will likely in Malaysia, as well as the relative automatability of be a persistent feature of the Malaysian economy, tasks performed by migrant workers and Malaysians. strengthening the foreign worker management The final section summarizes the main findings and system (which facilitates the intake of low-skilled presents policy options for Malaysia moving forward. 5 Migration, Automation, and the Malaysian Labor Market: A summary of findings 2. An overview of Malaysia’s migration policies Malaysia has a two-tiered work permit system, employment durations. Category I workers earn at distinguishing between migrant workers employed least RM10,000 per month and have employment in lower-skilled jobs (used interchangeably contracts of up to five years. Category II workers with the term “foreign workers”) and migrant earn between RM5,000 and RM9,999 per month workers employed in high-skilled jobs (used and have employment contracts of up to two interchangeably with the term “expatriates”). years. Category III workers earn between RM3,000 Low-skilled migrant workers are granted a Visitor’s to RM4,999 and have employment contracts of Pass (Temporary Employment) (VP(TE)), while mid- up to 12 months. Other differences between the and high-skilled migrant workers are granted an different visa categories include the ability to bring Employment Pass (EP). Low-skilled migrant workers dependents into the country (allowable for EP include workers who typically engage in manual Category I and II) and the ability to bring a foreign and elementary occupations as well as domestic domestic helper into the country (allowable for helpers. They expect to earn at least the minimum EP). Foreign spouses or dependents of Malaysian wage, which was increased to RM1,500 per month citizens with a Long Term Social Visit Pass – which in May 2022, and the maximum duration of stay is are granted for stays of more than six months and 10 years. There are three categories of the EP, which up to five years and can be extended –are permitted are characterized by different salary levels and to work in Malaysia. Box 1: Types of migrant workers in Malaysia Legal economic migrants include foreigners with valid employment visas and include workers across the skill spectrum. Higher-skilled workers who hold the EP are typically referred to as “expatriates”, while lower-skilled workers who hold the VP(TE) are typically referred to as “foreign workers”. Irregular foreign workers include foreigners without a valid employment visa, and are usually employed in low-skilled jobs. This includes those who have entered the country illegally (i.e. without a valid passport, travel document, or entry permit), persons who have entered the country lawfully but are not authorized to work (e.g. those with a VP(TE) but have changed employers while in Malaysia, those with tourist or student visas but engage in employment activities), and overstayers who do not leave the country after the expiry date or cancellation of their VP(TE) (Loh et al. 2019). Another category of irregular foreign workers include refugees, as described below. Refugees, including asylum seekers are not legally recognized in Malaysia3 and are therefore not authorized to work. However, they do seek employment and most refugees are de facto integrated in the Malaysian society as part of the foreign worker economy (Wurscher 2018). 3 Malaysia is not a party to the 1951 Refugee Convention and its 1967 protocol, which are key legal documents that define a “refugee”, outlines their rights, as well as the legal obligations of states to protect them. 149 states are parties to either or both the convention and the protocol. The United National High Commissioner for Refugees (UNHCR) is the “guardian” of both documents. 6 Migration, Automation, and the Malaysian Labor Market: A summary of findings Migration in Malaysia is primarily composed of and Local Workers Placement Committee. Following temporary migration schemes, with very limited the relevant committee meetings, employers will pathways to permanent residence and citizenship. receive a notice of the outcome of their application. The main migration schemes in Malaysia, namely the Visitor’s Pass (Temporary Employment) scheme The foreign worker landscape in Malaysia has for low-skilled workers and the various Employment been in flux since the COVID-19 pandemic; at Pass schemes for mid- and high-skilled workers present there is a freeze on the admittance of all specify a maximum duration of five years. The foreign workers. During the COVID-19 pandemic, visas are renewable, but they do not automatically the government placed a moratorium on the hiring lead to eligibility to permanent residence or of foreign workers, which was very gradually lifted. To citizenship. While there are official pathways for address the shortage of foreign workers during the permanent residence, anecdotal evidence suggests pandemic, the government has sought to legalize that approvals are few and far between. This irregular foreign workers through a “recalibration restrictiveness is similar to that seen in countries in program” first implemented in November 2020.4 the Middle East such as the United Arab Emirates. The recalibration program (Rekalibrasi Tenaga In contrast, countries such as Australia, the United Kerja, RTK) – which was in place until the end of States, as well as various European nations 2023 – allows employers from the manufacturing, including Germany and Switzerland, offer more construction, agriculture, plantation, and selected straightforward pathways to permanent residence services sector to legalize irregular foreign workers and citizenship. (not including refugees) (see Box 1). Domestic foreign helpers are also eligible. Eligible irregular Employers are required to perform a labor foreign workers include those who have overstayed market test, that is, to advertise vacancies on their Social Visit Pass or VP(TE) from 15 source the national job portal, MYFutureJobs, before countries.5 6 As of November 2021, more than applying to hire foreign workers and expatriates. 280,000 workers were registered under the RTK The vacancies for foreign workers have to be program.7 From January to March 2023, the advertised for a minimum of 30 days, and employers government introduced the Foreign Worker are required to arrange for interview sessions with Recruitment Relaxation Plan, which expedited local candidates within the 30 days. Vacancies for approvals for foreign workers for the manufacturing, expatriates have to be advertised for a minimum of construction, plantation, and agriculture sectors, 14 days. At the end of the period, employers must as well as the food and beverage (restaurants) complete a Hiring Outcome Report and submit it subsector.8 The Relaxation Plan saw approvals of to the Social Security Organisation (SOCSO) (the almost one million foreign workers.9 However, in government agency that oversees MYFutureJobs). March 2023, the government froze the application Subsequently, their application for hiring foreign and approval of foreign workers, and has not lifted workers and expatriates will be considered by the the freeze since.10 This is occurring against the Expatriates Placement Committee and the Foreign backdrop of unemployment among existing migrant 4 Analysis by the World Bank suggests estimates that irregular migrants make up between 40 to 44 of all foreign workers in Malaysia (Ali Ahmad, Simler and Yi 2020). 5 The 15 source countries are Bangladesh, Cambodia, India, Indonesia, Kazakhstan, Lao PDR, Myanmar, Nepal, Pakistan, the Philippines, Sri Lanka, Thailand, Turkmenistan, Uzbekistan, and Vietnam. 6 The workers would need to undergo and pass a medical screening, and employers are responsible for paying a recalibration fee of RM1,500 per worker, as well as a sector-specific levy, a processing fee of RM125 per worker, a VP(TE) fee of RM60, and a country- specific visa fee. In 2023, the sector-specific levy is RM1,850 for the manufacturing, mining and quarrying, construction and services sectors, as well as for security guards, and RM640 for the agriculture and plantation sectors. The program is scheduled to continue until 31 December 2023. 7 Source: https://www.nst.com.my/news/nation/2021/11/750153/employers-must-immediately-hire-foreign-workers-under-rtk- programme 8 Source: https://www.nst.com.my/news/nation/2023/02/880073/govt-resolve-foreign-labour-shortage-3-months-foreign-worker- recruitment 9 Source: https://www.nst.com.my/news/nation/2023/03/890353/application-and-approval-foreign-worker-quota-be-temporarily- discontinued 10 Source: https://www.nst.com.my/news/nation/2024/10/1122976/saifuddin-freeze-foreign-worker-employment-quota-continue 7 Migration, Automation, and the Malaysian Labor Market: A summary of findings workers in Malaysia, including those who entered be a recognition of the importance of improving the country as part of the Relaxation Plan. the process of admitting migrant workers and in attracting global talent to improve business The recently introduced Madani Economy competitiveness. Recently, the government Framework, like the 12th Malaysia Plan, explicitly announced its plans of reducing the duration of mentions the need to reduce dependence on low- foreign worker recruitment from more than 29 months skilled foreign labor, including through the use of to less than 16 months.12 The government has also a multi-tiered levy system. The Prime Minister’s announced an initiative to streamline the approval speech announcing the framework states that process for the hiring of expatriates. Under the new the dependence on low-skilled foreign labor, “… Xpats Gateway system, a one-stop center will be contributes to the overall low wage levels. This also used to process applications (currently handled by reduces the morale and incentive for employers about 11 separate government agencies), which will to innovate and improve competitiveness. The see a reduction the time taken for the application Government aims to implement tiered foreign and approval process from about 80 days to about worker levies, where part of the levy increase will be 22 days.13 These initiatives suggest an interest to allocated to automation and training programs for better facilitate migration into Malaysia. Moreover, local workers (Anwar Ibrahim 2023).” Using a multi- in 2022, the government introduced a digital nomad tiered levy system for the hiring of foreign workers visa, named the DE Rantau program, which is aimed as a tool to manage the demand for foreign labor at establishing Malaysia as the preferred digital has been discussed since at least 2018 (see ILMIA nomad hub in the Association of Southeast Asian 2018), but its implementation has been delayed. In Nations (ASEAN), while boosting digital adoption 2021, the Minister of Human Resources announced and promoting digital professional mobility and that its implementation has been postponed to tourism across the country. In a recent editorial, the 2022,11 after which there has not been any official announcements of its implementation. Minister of Economy stated that Ministry of Economy will work to facilitate the existence of a startup hub Despite the goal of reducing the number of both in Malaysia, for which a talent pipeline – including foreign workers and expatriates, there appears to international talent – is to be one of the enablers.14 11 Source: https://www.nst.com.my/news/nation/2021/06/696117/multi-tier-levy-system-foreign-workers-postponed-2022 12 Source: https://bernama.com/en/news.php?id=2263063 13 Source: https://www.thestar.com.my/news/nation/2023/06/01/fastlane-approval-for-skilled-expats-to-come-into-force 14 Source: https://www.malaysiakini.com/columns/667119 8 Migration, Automation, and the Malaysian Labor Market: A summary of findings 3. The role of migrant workers in the Malaysian labor market The structural transformation of the Malaysian The share of skilled employment has been economy from an agriculture-centric economy increasing gradually over time, but low-skilled to one driven by services has been accompanied jobs remain important in the Malaysian economy. by an educational upgrading of the Malaysian High-skilled jobs make up the largest share of jobs workforce. In 1957, agriculture accounted for 58 created (based on the number of workers employed percent of employment (World Bank 2019a). Since in high-skilled jobs at the beginning and at the end then, its share of employment has declined to 10.3 of the period) in Malaysia between 2013 and 2022, percent in 2021 (see Figure 1), releasing labor first at 70 percent of jobs (see Figure 3).16 This is followed to the manufacturing sector, and then increasingly by mid-skilled jobs, which make up 37 percent of to the services sector (Abdur Rahman and Schmillen jobs created. Low-skilled decreased the overall 2023). In the same period, the Malaysian population number of jobs created, with a share of negative has become increasingly more educated. Figure 7 percent. The share of employment by skill level 2 shows that among persons born in 1956, only has only changed very gradually between 2013 16.4 percent have a post-secondary education. In and 2022 (see Figure 4). In the period, the share of comparison, 38.7 percent of the population born in high-skilled jobs increased by 5.4 percentage points 1996 have a post-secondary education.15 over 9 years. The shares of mid- and low-skilled jobs decreased slightly, by 2.9 percentage points and 2.5 percentage points respectively. FIGURE 1: Share of employment by sector, 1990- FIGURE 2: Educational attainment by year of birth, 2021 (%) 2021 100 100 Share of population (%) Share of employment (%) 80 80 60 60 40 40 20 20 0 1966 1970 1974 1992 1996 1956 1958 1960 1962 1964 1968 1972 1976 1978 1980 1982 1984 1986 1988 1990 1994 0 1990 1993 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 Year of Year of birth birth No formal education education Primary Primary Agriculture Mining Manufacturing Secondary Post-secondary Post-secondary Construction Services Tertiary Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) 15 The data used for Figure 2 is that for 2021. Thus, persons who are born in 1996 – the youngest cohort in the figure – are or will be 25 years old within the year at the time the data is collected. 16 The occupational skill level is based on the mapping of ISCO-08 to the International Standard Classification of Education 1997 (ISCED-97) as detailed in ILO (2012). High-skilled occupations include managers, professionals, and technicians and associate professionals, which require tertiary education. Mid-skilled occupations include clerical support workers, services and sales workers, skilled agricultural, forestry, and fishery workers, craft and related trades workers, and plant and machine operators, and assemblers, which require secondary education. Low-skilled occupations include elementary occupations, which require primary education. 9 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 3: Share of jobs created between 2013 and FIGURE 4: Share of employment by skill level, 2013- 2022 (%) 2022 (%) 110 110 100 100 employment (%) 90 90 Mid-skilled Mid-skilled 80 80 Share of jobs created (%) Share of jobs created (%) of employment 70 70 60 60 50 50 40 40 Share of High-skilled High-skilled Share 30 30 20 20 10 10 0 0 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Low-skilled Low-skilled -10 -10 2013-2022 2013-2022 High-skilled Mid-skilled Low-skilled High-skilled Mid-skilled Low-skilled Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) The number of migrant workers in the labor force jobs increased by 4.4 percentage points, while the has declined over the last decade, and a higher share of migrant workers in mid-skilled jobs actually share of migrant workers are employed in low- declined by 9.3 percentage points. skilled jobs.17 At 11.6 percent, the share of migrant The shares of migrant workers by occupation workers (as defined in Box 1) in the labor force in and sectors reinforce the finding that they are 2022 is lower than in 2013, when it stood at 15.4 more likely employed in low-skilled jobs. More percent (see Figure 5). The relatively steep decline specifically, migrant workers are very likely to be in both the number and the share of migrant workers employed in elementary occupations (which is the since 2019 has arguably been due to a moratorium only low-skilled MASCO 1-digit occupation) (see in hiring foreign workers imposed during the Figure 7). Migrant workers make up 44.7 percent COVID-19 pandemic that was only temporarily of total employment in elementary occupations. lifted (see Section 2). The share of migrant workers Further, between 2013 and 2022, the share of is highest in low-skilled jobs. As a share of total migrant workers employed in some mid-skilled jobs employment in low-skilled jobs, the share of migrant – including skilled agricultural workers, plant and workers is 44.7 percent in 2022 (see Figure 6). In machine operators and assemblers, and craft and comparison, the share of migrant workers in mid- related trades workers – had declined substantially. and high-skilled jobs are 8.9 percent and 3.3 percent Similarly, migrant workers are likely to work in respectively. Separately, in 2013, 41.5 percent of all sectors that require low-skilled, manual labor such migrant workers in Malaysia were employed in low- as the agriculture, manufacturing, and construction skilled jobs. By 2022, this share had increased by sectors (see Figure 8). The shares of citizens and 4.9 percentage points to 46.4 percent. In the same non-citizens employed by sector are largely similar period, the share of migrant workers in high-skilled in 2013 and 2022, with a notable difference being 17 This paper primarily utilizes the Labour Force Survey (LFS) administered by the Department of Statistics Malaysia (DOSM) in its analysis of migrant workers. The LFS is a household-based sample survey which excludes communal housing, where many foreign workers working in the agriculture, manufacturing, and construction sectors are likely to live. This leads to an underestimation of the number of foreign workers. At the same time, the LFS captures both regular and irregular foreign workers that are not captured through other official data sources (see Footnote 5 and Box 1) (World Bank 2015). While the resulting representativeness of the LFS in capturing foreign workers is unknown, underestimation remains to be likely given the prevalence of communal housing among foreign workers, both regular and irregular alike. 10 Migration, Automation, and the Malaysian Labor Market: A summary of findings the increase in the share of migrant workers being been a decrease in the share of citizens employed employed in the agriculture sector, from 21.1 in the agriculture sector, reflecting the different percent in 2013 to 28.1 percent in 2022. There has profiles of the different workers. FIGURE 5: Number and share of migrant workers in FIGURE 6: Share of employment by citizenship and the labor force, 2013-2022 (million and %) skill level, 2013-2022 (%) Number of workers in labor force workers in labor force Share of immigrant workers in labor workers in labor 20 20 20 20 100100 Share of employment (%) Share of employment (%) 80 80 15 15 15 15 60 60 40 40 Number of (millions) (millions) force (%) force (%) 10 10 10 10 Share of immigrant 20 20 5 5 5 5 0 0 High-skilled High-skilled Mid-skilled Mid-skilled Low-skilled Low-skilled High-skilled High-skilled Mid-skilled Mid-skilled Low-skilled Low-skilled 0 0 0 0 2013 2014 2013 2015 2014 2016 2015 2017 2016 2018 2017 2019 2018 2020 2019 2021 2020 2022 2021 2022 2013 2013 2022 2022 Non-citizen Non-citizen Citizen Citizen Citizen Non-citizen Citizen Non-citizen Non-citizen Non-citizen (%, (%, right right axis) axis) Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) FIGURE 7: Share of employment of migrant workers FIGURE 8: Share of employment by citizenship and by occupation, 2013 and 2022 (%) sector, 2013 and 2022 (%) Elementary Elementary Occupations Occupations Non-citizen PlantWorkers Plant Workers 2022 Service Service and andSales SalesWorkers Workers Craft Craft Workers Workers Citizen Skilled Skilled Workers AgriculturalWorkers Agricultural Associate Associate Professionals Professionals Non-citizen 2013 Professionals Professionals Managers Managers Citizen Clerical Clerical Support Support Workers Workers 0 20 40 40 60 60 80 80 100 100 00 10 20 10 20 30 30 4040 50 50 Share of employment (%) of employment (%) ofmigrant Shareof Share workers (%) migrantworkers (%) Agriculture Agriculture Mining Mining Manufacturing Manufacturing 2022 2013 2022 2013 Construction Construction Services Services Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) 11 Migration, Automation, and the Malaysian Labor Market: A summary of findings Indonesian workers – almost a third of whom are Migrant workers are less educated and younger working in the plantation sector – make up the than Malaysian workers on average, suggesting largest share of low-skilled migrant workers in that migrant workers complement higher-skilled Malaysia, but the share of Bangladeshi workers Malaysians in a country that is aging. The majority has been rising over time. About 35.2 percent of migrant workers – 67.0 percent in 2022 – have a of low-skilled migrant workers in Malaysia are primary education or less (see Figure 11). While the from Indonesia, followed by 28.4 percent from share of Malaysians with at least a post-secondary Bangladesh, and 15.8 percent from Nepal (see education has increased between 2013 and 2022, Figure 9). While not shown here, data presented the higher levels of educational attainment among in Ali Ahmad et al. (2020) show that the share of migrant workers have been concentrated at the low-skilled migrant workers from Bangladesh has secondary level, rising from 14.8 percent in 2013 to increased dramatically between 2015 and 2018, 24.0 percent in 2022. In the same period, the share making it the second largest source country in 2019. of migrant workers with at least a post-secondary There appears to be some degree of differentiation education has increased by 4.0 percentage points, in the sector of employment by nationality. but remains a small share of the labor market. Indonesian workers are more likely to work in the Migrant workers are also significantly younger than plantation sector, with 28.6 percent of low-skilled Malaysians, with 71.5 percent of migrant workers migrant workers from Indonesia being employed in being between the ages of 15 to 34 in 2022, the sector in 2019 (see Figure 10). In comparison. compared to 47.3 percent of Malaysians (see Figure Bangladeshi workers are more likely to work in the 12). Notably, this is a contrast from 2001, when construction and manufacturing sectors, with 38.1 the age distributions of Malaysians and migrant percent and 36.5 percent of workers in each sector workers were more similar, and is reflective of the respectively. The majority of Nepalese workers (72.4 aging Malaysian population (World Bank 2015). This percent) work in the manufacturing sector. Similarly. pattern is likely to persist, given that Malaysia is 71.6 percent of Vietnamese workers and 62.4 further along in the aging process than its primary percent of Sri Lankan workers are employed in the source countries. manufacturing sector, and 61.6 percent of Filipino workers are employed as domestic workers. FIGURE 9: Share of low-skilled migrant workers by FIGURE 10: Share of low-skilled migrant workers by nationality, 2019 (%) sector and nationality, 2019 (%) 40 40 100.0 100.0 Share of foreign workers (%) Share of foreign workers (%) Share of foreign workers (%) Share of foreign workers (%) 30 30 80.0 80.0 60.0 60.0 20 20 40.0 40.0 10 10 20.0 20.0 0 0.0 0.0 0 Indonesia Bangladesh Nepal Indonesia Bangladesh Nepal ya N a l h ia Th ietames a bo ka Cin d yamapal Tahila am n na s Pa diaar ne n ast ia es ngdeesi N ladsh oia aso La d C l an p sta e s Baglaon ia m Lakna aind La hia Innmr n Vtn in ki Ind Lo bd Manufacturing Construction Plantation n e ip kia n n d es M ep Vi ilipip Manufacturing Construction Plantation Sr h am Ba Inon C Sr i Ph P M Services Agriculture Domestic workers Ph i d C l a i Services Agriculture Domestic workers In Source: Ali Ahmad et al. (2020) Source: Ali Ahmad et al. (2020) 12 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 11: Share of labor force by citizenship and FIGURE 12: Share of labor force by citizenship and education level, 2013 and 2022 (%) age group, 2022 (%) Non-citizen 2022 Non-citizen Non-citizen 2022 Citizen Non-citizen Citizen Non-citizen 2013 Non-citizen 2013 Citizen Citizen Citizen Citizen 0 20 40 60 80 100 0 20 40 Share of 60 (%)80 labor force 100 0 20 40 60 80 100 Share of labor force (%) 0 20 No certificate Primary 40 60 Share of labor force (%) 80 100 No certificate Secondary Primary Post-secondary Share of labor force (%) 15-24 25-34 35-44 45-54 55-64 Secondary Bachelor's degree or higher Post-secondary 15-24 25-34 35-44 45-54 55-64 Bachelor's degree or higher Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Excluding Sabah and the Federal Territory of to Malaysia’s GDP in 2021, contributing 24.8 Labuan, migrant workers make up higher shares of percent, 15.7 percent, 9.5 percent, and 7.1 percent employment in or near the major economic centers to GDP respectively.18 Negeri Sembilan is located of the country. Out of all workers in Kuala Lumpur, at the border of Selangor and anecdotally houses 13.4 percent are migrant workers. Similarly, the people who work in Selangor and Kuala Lumpur share of migrant workers make up 9 to 11 percent of (which is geographically located within Selangor). total employment in the Federal Territory of Kuala Sabah and Labuan (which is geographically located Lumpur, Johor, Negeri Sembilan, Pulau Pinang, within Sabah) house an exceptionally high share and Selangor (see Figure 13). Kuala Lumpur, Johor, of migrants as Sabah houses refugees from the Pulau Pinang, and Selangor are major economic Mindanao region of the Philippines,19 as well as centers in the country. Data from DOSM show that irregular migrant workers from Indonesia and the the states of Selangor, Kuala Lumpur, Johor, and Philippines given its proximity to the two countries Pulau Pinang were among the highest contributors (Azizah Kassim 2009). 18 In 2021, the state of Sarawak also contributed 9.5 percent of GDP, making it one of the five states or federal territories with the greatest contribution to GDP. Sarawak’s economy is driven by its natural resources and particularly petroleum, with mining and quarrying making up 21.1 percent of its contribution to GDP, and the manufacturing of petroleum and petroleum-related products making up 23.3 percent of its contribution to GDP. 19 These refugees are allowed to stay in Sabah using the IMM13 visa, which is renewed every year upon payment of an annual fee of RM90 (see Azizah Kassim 2014). IMM13 holders are given limited access to employment, social services, and public amenities. 13 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 13: Share of migrant workers by state, 2022 (%) 40 Share of migrant workers (%) 30 20 10 0 Sabah Negeri Sembilan Pulau Pinang WP Labuan Sarawak Pahang Melaka Kedah Perak WP Putrajaya Kelantan Terengganu Perlis WP Kuala Lumpur Johor Selangor Source: DOSM Migrant workers earn less than Malaysian Malaysian men who are managers or professionals workers on average, although there is evidence are likely to earn about 5.3 percent and 7.5 percent that high-skilled migrant workers earn more less than non-Malaysian men who are managers than high-skilled Malaysian workers. In 2021, or professionals on average. This is unsurprising, the median wage for Malaysian men and women given that migrant workers who are managers or were RM12.58 and RM12.18 per hour respectively professionals are likely to be working in the most (see Figure 14). In contrast, the median wage for senior positions, and are typically granted globally migrant men and women were RM7.07 and RM6.57 competitive compensation packages. After taking per hour respectively. On average, Malaysian into account socioeconomic factors, Malaysian workers earned about 79.3 percent more than women who are plant and machine operators and migrant workers in 2021. That said, this gap has assemblers are likely to earn about 5.1 percent declined from 116.9 percent in 2010. After taking less than non-Malaysian women, although the into account socioeconomic factors, analysis finds possible reasons for this is unclear. Other than these that on average, Malaysian workers are still likely categories of workers, Malaysians are more likely to to earn more than migrant workers (see Figure 15). earn more than migrant workers across occupations. 14 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 14: Median hourly real wages, 2010-2021 FIGURE 15: Difference in hourly wages of Malaysians (2021 RM) and migrant workers by gender (%) Median hourly real wages (2021 RM) Difference in hourly wages (%) 14 30 Median hourly real wages (2021 RM) Difference in hourly wages (%) 14 30 12 20 12 20 10 10 10 10 8 8 0 0 6 6 -10 4 -10 s s ls s s s er er er sn er ls s s 4 lsa s er er s rs s io lsa an k k r rk k r otn r rs rs an er e o oe r oe k r eg o o tia osn o k oe k r rk k i ga w sn ap 2 r w w w r i rk r n so eis wlo eo rto o w pu eis na s ft o ew 2 inw w cc a osf s M rtaw cu rto e osf ulr a asl he o fre pfr ta op M e in o yc als ulr af p c P C 0 o aro r ltu p d h e ca atp pu r Pr s ic m nrty C dn 0 ls cru tie da su a ta rig 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 m asn la iac ne a co ic dn se m dg 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 cra ss lee a ic lea so lie atn ve Em A ild Cr eirc tn As le e e la rv illk El Pn C S S Se a Sk Pl Men Citizens Women Citizens Men Citizens Women Citizens Men Women Men Non-citizens Women Non-citizens Men Women Men Non-citizens Women Non-citizens Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Notes: Hourly wages refer to wages including overtime Notes: A positive difference indicates that Malaysian workers payment, divided by hours of work. The median wages earn more than migrant workers on average. Hourly wages are adjusted by the consumer price index (2021=100). The refer to wages including overtime payment, divided by hours difference in median wages between citizens and non-citizens of work. The analysis is conducted using data for the years (for each gender) are statistically significant for all years plotted. 2017-2021. The analysis controls for age, age-squared, gender, marital status, the interaction between gender and marital status, education level, urban-rural location, occupation, state, and year. The difference in hourly wages between citizens and non-citizens are statistically significant across all occupations for men. For women, the differences are statistically significant for all but managers, agricultural (and related) workers, and craft and related trades workers. FIGURE 16: Change in the number of Malaysians FIGURE 17: Change in the number of Malaysians employed, unemployed, and out of the labor force employed due to 1 additional migrant due to 1 additional migrant 1.5 1.5 0.3 0.3 Malaysians of Malaysians Malaysians of Malaysians 1 1 0.2 0.2 0.1 0.1 0.5 0.5 Number of 0 Number Number of 0 0 0 Number -0.1 -0.1 above and above Certificate/diploma secondary secondary Certificate/diploma Upper secondary Lower secondary formal -0.5 15-19 20-29 30-49 50-64 No formal Primary -0.5 15-19 20-29 30-49 50-64 Primary Degree and -1 -1 No Female Female Female Overall Overall Overall Female Female Female Male Male Male Overall Overall Overall Male Male Male Degree Upper Lower Malaysian Malaysian Malaysian Malaysian Malaysian Malaysian Malaysians Malaysians employment employment Malaysian employment Malaysian employment employment unemployment out by age by age by by education education employment unemployment out of of labor force labor force Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Note: The analysis is conducted using data for the years 2010-2020. Note: The analysis is conducted using data for the years 2010-2020. 15 Migration, Automation, and the Malaysian Labor Market: A summary of findings Econometric analysis conducted for this paper findings show that there is no evidence of an generally finds a favorable, or no significant adverse relationship between immigration and labor relationship between immigration and labor market outcomes of Malaysians. Instead, there is market outcomes of Malaysians. 20 More some evidence of a fall in unemployment resulting specifically, an ordinary least squares analysis using from immigration, as well as a positive relationship data for the period 2010 to 2020 finds that one between immigration and the employment rate of additional migrant is associated with 0.2 additional Malaysians, particularly among those who are aged employed Malaysians in a given state and sector, 30 and older and those with a secondary education, with the relationship being statistically significant which may suggest complementarity between (see Figure 16). Similar analysis finds that there is migrant workers and Malaysian workers. no significant relationship between immigration with unemployment and labor force participation of In contrast to the findings in this study, which Malaysians in a given state. Further, an additional finds no significant relationship between the migrant in a given state and sector is associated employment of migrant workers and the wages with higher levels of employment among Malaysians of Malaysians, previous studies have found a aged 30 and older, and Malaysians with secondary small negative relationship. Using data for the education, as reflected by statistically significant period 1990 to 2014, World Bank (2015) finds that positive relationships (see Figure 17). Reinforcing immigrants do not affect the unemployment rate these findings, an instrumental variable analysis and labor force participation rate of Malaysians, but (which accounts for reverse causality and allows do increase the employment for Malaysians. The for estimation of the impact of migration on labor study finds that for every 10 new migrant workers market outcomes21) focusing on the relationship in a state, there are 5.2 additional Malaysians between the immigration rate and labor market employed (World Bank 2015). The positive impact outcomes of Malaysians reveals similar findings. on employment is largest for Malaysians aged 20 The immigration rate, as measured by the ratio to 29, and those who are more educated, with of non-citizens to citizens in a given state, was upper secondary or a university degree, who can found to have no significant relationship with the better complement unskilled foreign workers (World employment rate, the labor force participation rate, Bank 2015). However, the wages of less-educated and wages of Malaysians in the state (see Table A1 in Malaysians were found to be negatively affected by Annex 1). There is, however, a statistically significant immigration, perhaps due to the substitutability of negative relationship between the immigration rate the workers with foreign workers (World Bank 2015). and the unemployment rate, suggesting that a one Specifically, a 10 percent increase in the employment percentage point increase in the immigration rate of low-skilled migrant workers decreases the wages leads to a 0.29 percentage point decrease in the of Malaysians working in less-skilled occupations unemployment rate of Malaysians. Similar results by 0.7 percent (World Bank 2015). Similar findings are found even when one analyzes the relationship were made by Athukorala and Devadason (2012) between the immigration rate of migrants with and Tan and Ng (2018), who found that an increase different levels of education and the labor market in migrant workers is associated with a small decline outcomes of Malaysians using instrumental variable in wages. To elaborate, Athukorala and Devadason analysis (see Table A2 in Annex 1). Overall, these (2012) found that a 10 percent increase in the 20 The findings follow Özden and Wagner’s (2014) methodology. The dependent variable is the number of employed citizens in the respective sub-groups for a given region/industry/year. The independent variables are the number of employed non-citizens in the region/industry/year and region by sector, year by sector, and year by region fixed effects. The reported results are based on ordinary least squares (OLS) regressions with standard errors clustered by industry and region. The results for the unemployment and out of labor force outcomes follow and adjusted strategy, as people who are not employed cannot be categorized by sector. The dependent variable is the number of unemployed or out-of-labor-force citizens in the region/year. The independent variables are the number of employed non-citizens in the region/year and region and year fixed effects. The reported results are based on OLS regressions with standard errors clustered by region. The overall analysis is based on the national sample as collected in the Labour Force Survey. 21 The goal of the analysis is to estimate the causal effect of a higher concentration of non-citizens on labor market outcomes. The difficulty is that non-citizens do not necessarily immigrate to a given state randomly, but they are driven by economic conditions across the states. These are likely correlated with the labor market outcomes, leading to an omitted variable bias in the OLS estimates. The instrumental variable (IV) approach is designed to overcome the omitted variable bias by using the past levels of immigrant concentrations as an instrument. Please see Annex 1 for the full technical details of the analysis. 16 Migration, Automation, and the Malaysian Labor Market: A summary of findings employment of low-skilled migrant workers results Overall, this section presents evidence on the role in a 1.3 percent decline in the wages of all low- of migrant workers in the Malaysian economy, skilled workers. Tan and Ng (2018) found that an and importantly points to the generally positive increase in the employment of foreign workers in relationship between the employment of migrant a given sector and year by one percentage point workers and the employment outcomes of is associated with a 3.8 percent decrease in overall Malaysian workers. It contradicts the narrative average wages. They also suggest that the decrease that the employment of foreign workers depresses in overall average wages is due to the relatively low the wages of Malaysians and deters the creation of wages earned by foreign workers, rather than the high-skilled and high-paying jobs (Ang et al. 2018, suppression of overall wages, and that the wages of BNM 2021). Data also show that migrant workers are native workers have actually increased in the relevant largely and to an increasing extent being employed period (Tan and Ng 2018). Moreover, once sectoral in low-skilled jobs that are arguably unattractive to effects are taken into account, the relationship increasingly educated Malaysians. Migrant workers between the share of foreign workers and wages is can also complement Malaysian workers in an aging statistically insignificant (Tan and Ng 2018). economy given their younger age profile. 17 Migration, Automation, and the Malaysian Labor Market: A summary of findings 4. The potential for automation in Malaysia The Malaysian government is committed 43.69 for upper middle income countries (Oxford to increasing the adoption of automation Insights 2023). The report states that Malaysia’s technologies within the next decade. The National strength is largely in dimensions related to AI skills Robotics Roadmap 2021-2030 situates Malaysia as and education, suggesting that “…the country is set an emerging economy in terms of its automation up to be a source of much-needed AI talent in the readiness. It states that there is room for improvement years to come” (Oxford Insights 2023, p. 9). in the legislation, research and talent development, The adoption of automation technologies has been investment, and commercialization in order to found to have an impact on the labor market, with increase the adoption of automation (MOSTI 2021). global research suggesting that jobs composed of In the same document, the government sets the goal routine tasks having a greater likelihood of being of increase the robot density per 10,000 employees affected. The task-based framework for automation in manufacturing from 55 units per 10,000 employees posits that given that jobs are composed of bundles in 2019 to the expected global average of 195 units of tasks that are either performed by capital or per 10,000 employees by 2030 (MOSTI 2021). The labor, new technologies – which is a form of capital Roadmap also states Malaysia’s goal of becoming – affect labor demand by changing the task content a regional robotics hub in services, agriculture, and of production being performed by labor (Acemoglu manufacturing by 2030. In line with this objective, and Restrepo 2019). On one hand, automation Malaysia installed almost 2,000 industrial robots in technologies may take over tasks previously 2021, which represented a 37 percent increase in performed by labor through the displacement industrial robots (IFR 2022). Other than facilitating effect. On the other, automation technology also an increasing robot density, the National Robotics increases productivity, and via the productivity Roadmap also details strategies for strengthening effect, can contribute to the demand for labor in local robotics research, ensuring that workers in non-automated tasks (Acemoglu and Restrepo Malaysia have the relevant skills and qualifications 2019). Several studies have found that automation to complement automation technologies, and to displaces low-skilled, blue-collar workers performing ensure that local standards and regulations are fit routine tasks (Graetz and Michael 2018; Acemoglu for purpose. and Restrepo 2020; Acemoglu et al. 2023). At the same time, there is also evidence that employment Malaysia also aspires to nurture expertise in other types of jobs, particularly those involving and foster innovation in artificial intelligence tasks that are complementary to automation (AI), toward AI-driven value creation. The New technologies, increase as a result of automation, Industrial Master Plan (NIMP) 2030 details an reflecting the productivity effect (Humlum 2019; action plan to this effect. The goal is to empower Acemoglu et al. 2023). Studies on the impact of local system integrators to assume pivotal roles automation in Spain (Koch et al. 2021) and France in driving technological advancements and (Aghion et al. 2021) find that employment for all enhancing efficiency within the manufacturing types of workers – including low-skilled workers and manufacturing-related services sectors (MIDA – actually increases with robot adoption. These n.d.). In December 2023, the Prime Minister of studies show that while one can theoretically expect Malaysia introduced a new Digital Ministry, which automation to decrease the demand for low-skilled is positioned to lead the government’s digital workers (including migrant workers), the impact of transformation, including a focus on AI. Relatedly, automation is ultimately context dependent and is Malaysia has made important strides in advancing subject to the type of technology being adopted, the implementation of AI in the delivery of public among other factors. services. The Government AI Readiness Index 2023 by Oxford Insights (2023) ranks Malaysia 23rd globally Around half of all jobs in Malaysia have the in its readiness to implement AI in the delivery of possibility of being replaced or reshaped by public service. Malaysia’s overall score of 68.71 is automation. This translates into about seven million also substantially higher than the average score of jobs. This finding is based on the Frey and Osborne 18 Migration, Automation, and the Malaysian Labor Market: A summary of findings (2017) approach in a study undertaken by World and maintaining such technologies. A survey of Bank in partnership with TalentCorp (see World Bank medium- and large-sized furniture manufacturers 2020b). Jobs that are highly likely to be affected in Malaysia on the readiness for mechanization and are mainly composed of routine tasks that can be automation finds that among the reasons employers relatively quickly and easily automated through the are keen to utilize automation technologies include use of computers and robots. In comparison, jobs the consistent quality and high productivity that can that have low likelihood of being affected involve be achieved (Ratnasingam et al. 2019). However, non-routine tasks requiring socioemotional skills such the majority of the 312 manufacturers surveyed as creativity and persuasion, for which automation were wary of the high initial investment required, is more challenging. The same study suggests the lack of information technology infrastructure, that the depth of technological penetration, cost, and data management within their manufacturing and regulations will affect the actual impact of facility (Ratnasingam et al. 2019). Importantly, the automation in Malaysia’s labor market (World Bank study finds that the Malaysian furniture industry has 2020b). a relatively low diffusion of even more rudimentary technologies, and thus lacks the foundation to move Studies on the barriers to mechanization and into production processes utilizing more advanced automation in Malaysia find that among the main technologies (Ratnasingam et al. 2019). Studies barriers include the high cost of investing in the on the barriers to technology adoption in the technologies. The 2015 Malaysian Enterprise Survey construction industry in Malaysia finds that among shows that the likelihood of adoption an automation the main challenges to the adoption of construction technology generally increases with firm size. This robotics is the high cost of acquiring, maintaining, shows that due to cost, the economies of scale, and and updating new technologies (Yap et al. 2022). other factors, large and globally connected firms are This also includes the cost of training workers to more likely to be able to afford to automate a wider use such technologies, given their relatively low range of processes compared to small businesses, technology literacy (Mahbub 2012; Kamaruddin, which are more likely to persist with technologies Mohammad and Mahbub 2016). There is, however, that require manual labor (World Bank 2020b). an acknowledgement that technology adoption can Similarly, the National Robotics Roadmap 2021-2030 improve the safety of working in the sector, and acknowledges that the low level of robot penetration decrease the need for foreign workers (Mahbub in Malaysia is due to the prohibitive costs of acquiring 2012; Kamaruddin et al. 2016; Yap et al. 2022). 19 Migration, Automation, and the Malaysian Labor Market: A summary of findings 5. How automatable are tasks performed by migrant workers? The automatability of a job is generally estimated into five categories of tasks: non-routine cognitive using either an occupation-based approach or analytical tasks, non-routine interpersonal tasks, a task-based approach. An occupation-based non-routine manual tasks, routine cognitive tasks, approach considers the automatability of entire and routine manual tasks. They provide evidence jobs, while a task-based approach takes into that job polarization in the United States has account the task structure of different jobs, and resulted from “routinization”, or the displacement of the automatability of the different tasks. Focusing jobs involving routine tasks by recent technological on machine learning and mobile robotics, Frey advancements (Acemoglu and Autor 2011). While and Osborne (2017) utilized an occupation-based more recent advances in technology have allowed approach and estimated the automatability of for the automation of some non-routine tasks, the 702 occupations in the United States. Following share of routine tasks in a country arguably defines expert opinions on whether 70 selected jobs can the lower bound of the potential automatability be automated based on their task structure, they of tasks given existing technology (whether or not identified engineering bottlenecks that hinder this potential translates in automation depends automation, and estimated the automatability of on various factors, which will be discussed later). the remaining 632 occupations in the United States As mentioned earlier, the automation of some using a probabilistic model. Subsequently, they tasks does not necessarily imply the automation of produced a list of “computerization scores” by entire jobs, although it would result in the change occupation, which reflect the automatability of jobs in the task content of jobs. Advancing the task- in the United States. This approach has since been based approach, Arntz, Gregory and Zierahn (2016) criticized, with favor being given to the task-based consider the task structure of jobs in estimating approach (described below). The Frey and Osborne the likelihood of automation of jobs in the United (2017) approach has also been found to have States and other OECD countries, building on the limitations related to its subjectivity and inability to computerization scores generated by Frey and explain changes in employment in the United States Osborne (2017). Importantly, using data from the between 2013 and 2018 (Coelli and Borland 2019). Program for the International Assessment of Adult Competencies (PIAAC), they look at the specific tasks The task-based approach suggests that involved in jobs, such as “exchanging information”, because jobs are composed of various tasks, “solving complex problems”, and “using fingers or the automatability of a job depends on the hands”, instead of collapsing them into aggregate automatability of the different tasks. Hence, task categories as in Acemoglu and Autor (2011). while the task content of jobs may change to complement technologies, it is less likely for entire Different approaches yield very different results, jobs to become obsolete. The task-based approach with the occupation-based approach estimating originated from Autor, Levy and Murnane (2003) a much higher risk of automation compared to who suggest that jobs that require more routine the task-based approach. The findings from Frey tasks are more likely to be automatable, thus coining and Osborne (2017) using the occupation-based the term “routine-biased technological change” approach suggest that 47 percent of employment (RBTC). They also suggest that since mid-skilled in the United States are at high risk of automation, jobs are more likely to compose of routine tasks, and will potentially be automated over the next job polarization – that is, the decline of mid-skilled two decades. In comparison, using the task-based jobs, and the increase in low- and high-skilled jobs approach, Arntz et al. (2016) find that only 9 percent – can be attributed to declines in the real price of of workers in the United States are at high risk of information and communication technologies. This being displaced by automation. This pattern is also is further elaborated by Acemoglu and Autor (2011), seen for other countries. Using the occupation- who proceed to categorize tasks in the United based approach and Frey and Osborne’s (2017) States’s Occupational Information Network (O*NET) computerization scores, the share of jobs that 20 Migration, Automation, and the Malaysian Labor Market: A summary of findings are at risk of automation is 35 percent in Finland are used to illustrate the routineness of (Pajarinen and Rouvinen 2014) and 59 percent in tasks performed within jobs, augmenting Germany (Brzeski and Burk 2015). The task-based the Acemoglu and Autor (2011) approach. approach taken by Arntz et al. (2016) estimates Like the Acemoglu and Autor (2011) approach, that the relevant shares are 7 percent in Finland this approach does not allow the estimation of and 12 percent in Germany. Thus, it is clear that automatability of jobs but provides an indication the occupation-based approach estimates a much of the susceptibility of the task content in jobs to higher risk of automation compared to the task- automation technologies. The RTI is a composite based approach. measure constructed by Lewandowski et al. (2022) which increases with the importance This paper estimates the automatability of of routine content of work, and decreases jobs in Malaysia using three different methods with the importance of non-routine content of described below. These approaches provide a work (see Annex 3). The main difference – and range of considerations and estimations for the advantage – is that the RTI by occupation (as automatability of jobs in Malaysia and allow for the defined by 1- and 2-digit International Standard comparisons between the potential automatability of Classification of Occupations (ISCO) codes) for jobs held by Malaysian citizens and migrant workers. Malaysia is estimated using a regression-based Each approach provides different information and methodology utilizing data from 47 countries has different limitations, which are described below. as opposed to applying data for the United 1. The shares of jobs with high importance in non- States. More specifically, the RTI for Malaysia, as routine analytical, non-routine interpersonal, well as for other countries without occupational non-routine physical, routine cognitive, and task-related data was estimated based on a routine manual tasks are estimated and regression model that relate the RTI of each analyzed following the approach taken by occupation for 47 countries with task-related Acemoglu and Autor (2011). Given that the data to four key factors (Lewandowski et al. relative amount of time spent in these different 2022). The four factors are: (1) development level, tasks are not estimated (due to the unavailability measured by GDP per capita, (2) technology use, of data), this approach does not allow for the approximated by the number of internet users estimation of the automatability of jobs. It per 100 inhabitants, (3) globalization, quantified does, however, allow for an indication of the by foreign value-added share of domestic relative importance of routine versus non- output, and (4) the supply of skills, measured by routine tasks in jobs held by migrant and native the average years of schooling. The distribution workers. Subsequently, the approach allows for of the RTI for the United States has been found to an estimation of the potential for the job scope be consistent with the routineness of tasks within of the different type of workers to change as a occupations as measured by Acemoglu and result of automation. Since such technologies Autor (2011) using O*NET data. One limitation are arguably more rudimentary than other of this approach is that the RTI is measured at technologies that are presently available (such the 2-digit ISCO level, which is quite highly as machine learning and mobile robotics), this aggregated. Subsequently, this can affect the reflects a relatively conservative measure of the precision of the measure. For example, ISCO automatability of tasks. This approach uses data 26 (“legal, social, and cultural professionals”) on the task content of jobs from the 2019 version includes legal professionals, social and religious of O*NET, collected in the United States. It should professionals, authors, journalists, and linguists, be noted that this approach is not equivalent to and creative and performing artists, all of whom the task-based approach, as it does not estimate arguably perform very different tasks. Despite the automatability of jobs based on their task this shortcoming, this approach does provide an structures – as was done in Arntz et al. (2016) – empirical basis for identifying the routineness – due to the unavailability of data for Malaysia. and therefore automatability – of tasks performed in jobs held by Malaysian citizens and migrant 2. Routine task intensity (RTI) scores estimated workers (instead of relying on data for the United by Lewandowski, Park and Schotte (2023) States). 21 Migration, Automation, and the Malaysian Labor Market: A summary of findings 3. The computerization scores by Frey and 2010 version of O*NET – developed by and for Osborne (2017) are applied to data from the the United States – reflects task structures that Malaysia Labour Force Survey 2016-2021 to are relevant for Malaysia in the period between provide an estimate of the automatability of 2010 to 2021. Nonetheless, it remains to be an jobs. This method allows for a direct estimation important source of information in the absence of the automatability of jobs in Malaysia, using of other sources of data. an occupation-based approach. Aside from the limitations of the approach that have already Findings from the Acemoglu and Autor (2011) been discussed, there are two major caveats to and Frey and Osborne (2017) approaches are this approach. First, the computerization scores largely consistent. As shown in Figure 18, jobs are estimated based on the task structures of that place high importance in routine cognitive occupations in the United States, which are and routine manual tasks (based on the Acemoglu likely to greatly differ from the task structures of and Autor (2011) approach) being more likely to occupations in Malaysia.22 Despite this caveat, have high likelihood of automation (based on the as described above, this is a commonly used Frey and Osborne (2017) approach) compared approach in the literature, given the lack of data to jobs in which non-routine analytical and non- on country-specific task structures. Second, the routine interpersonal tasks are highly important. computerization scores are based on the task That said, jobs that in which non-routine physical structures as described in the 2010 version of tasks are highly important are also more likely to O*NET data. Given that the task content of jobs have high likelihood of automation based on Frey changes over time (Autor et al. 2003, Spitz-Oener and Osborne’s (2017) classification. This differs to 2006), the scores arguably do not remain relevant Acemoglu and Autor’s (2011) position, and can be for the entirety of the analysis period (i.e. 2010 to partly attributed to the technological advancements 2021). Subsequently, it is uncertain whether the made between the time of the publication of the two FIGURE 18: Share of employment in jobs with high task importance by probability of automation (%) 100 Share of employment (%) 80 60 40 20 0 Non-routine Non-routine Non-routine Routine Routine analytical interpersonal physical cognitive manual High risk Medium risk Low risk Source: World Bank staff calculations using data from LFS (DOSM) 22 For example, when a survey of occupational tasks in Indonesia (referred to as Indotask) was conducted, it was found that there were major differences in the relative importance of skills in Indonesia and the United States, with three out of 35 skills being relatively more important for Indonesia than the United States. The other skills are important for the United States than for Indonesia (World Bank 2020c). 22 Migration, Automation, and the Malaysian Labor Market: A summary of findings papers, as described in Frey and Osborne (2017). low RTI, and a relatively low computerization score. Acemoglu and Autor (2011) argue that non-routine In comparison, metal, machinery, and related trades physical tasks are those that require situational workers have a relatively high task importance adaptability, visual and language recognition, and in routine manual tasks, a relatively high RTI, and in-person interactions, and therefore would need to a relatively high computerization score. The task be performed by human workers. This paper takes importance of routine cognitive tasks for science the position of Acemoglu and Autor (2011) and and engineering professionals is (slightly) higher focuses on the automatability of routine cognitive than that for metal, machinery and related trades and routine manual tasks as indicators of overall workers, which is as indication of more routineness routineness of the tasks performed in a job. for the former group of workers. However, the task importance of non-routine analytical tasks is Direct comparisons of the different approaches substantially higher, hence explaining the overall are shown below. Taking two 2-digit ISCO less routine nature of the tasks performed by science occupations as examples, Table 1 shows that science and engineering professionals, as indicated by the and engineering professionals have a relatively low RTI and computerization scores. task importance in routine manual tasks, a relatively TABLE 1: Examples of automatability of tasks by measure and selected occupations Occupation Science and Measure of automatability Metal, machinery, and engineering related trades workers professionals (MASCO 72) (MASCO 21) Acemoglu and Autor (2011) task importance scores Non-routine analytical tasks 4.3 2.7 Non-routine interpersonal tasks 3.2 2.8 Non-routine physical tasks 1.4 3.3 Routine cognitive tasks 3.7 3.5 Routine manual tasks 1.8 4 Lewandowski et al.’s (2023) RTI -0.6 0.5 Frey and Osborne’s (2017) computerization score 0.1 0.7 Source: World Bank staff calculations using data from LFS (DOSM), Lewandowski et al. (2023), and Frey and Osborne (2017) The following sections apply the methodologies described to analyze the automatability of jobs in Malaysia, with a focus on the relative automatability of jobs held by migrant workers. 23 Migration, Automation, and the Malaysian Labor Market: A summary of findings 5.1 Analyzing the task content tasks have average importance scores of 2.6 and 2.5 respectively. At first glance, it appears that there are of jobs as an indication of sizeable differences in the average task importance automatability based on the for jobs held by migrant and non-migrant workers, Acemoglu and Autor (2011) with migrant workers having higher average task approach scores for non-routine physical tasks (3.3 versus 2.4) and routine manual tasks (3.4 versus 2.6), with all On average, routine cognitive tasks are the most of the differences being statistically significant (see important type of task for jobs in Malaysia, and Figure 20). However, once the occupational skill level there is a minimal difference in the type of tasks of jobs is taken into account, the average difference performed by migrant and non-migrant workers in the task scores for migrant and non-migrant after taking into account the occupational skill workers are narrower, with a maximum difference level. On a scale of 1 (least important) to 5 (most of 0.5 points (although all of these differences are important), the average importance of routine statistically significant) (see Figure 21 and Figure cognitive tasks is 3.9, while the average importance 22). This suggests that the initial difference seen of non-routine interpersonal tasks and non-routine reflects the relatively high share of migrant workers analytical tasks are 3.2 and 3 respectively (see Figure employed in low-skilled jobs.24 19).23 Routine manual tasks and non-routine physical FIGURE 19: Average task importance for jobs in FIGURE 20: Average task importance for jobs in Malaysia, 2021 (1-5) Malaysia by citizenship, 2021 (1-5) 55 55 importance task importance 44 importance task importance 44 Average task 33 Average task 33 Average Average 22 22 11 Non-routine Non-routine Non-routine Non-routine Routine Non-routine Non-routine Routine Routine Routine 11 analytical interpersonal analytical physical cognitive interpersonal physical cognitive manual manual Non-routine Non-routine Non-routine Non-routine Routine Non-routine Non-routine Routine Routine Routine Citizens Citizens Non-citizens Non-citizens analytical interpersonal analytical physical cognitive interpersonal physical cognitive manual manual Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Note: The difference in average task importance between citizens and non-citizens are statistically significant across all task categories. 23 The average task importance scores are weighted by the share of employment. Full details of the methodology are presented in Annex 2. 24 About 49.2 percent of migrant workers are employed in low-skilled jobs in 2021 compared to 7.4 percent of Malaysian workers, thus increasing the task importance of non-routine physical and routine manual tasks, which are more important in low-skilled jobs. 24 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 21: Average task importance for high- FIGURE 22: Average task importance for low- skilled jobs in Malaysia by citizenship, 2021 (1-5) skilled jobs in Malaysia by citizenship, 2021 (1-5) 55 55 Average task importance Average task importance Average task importance Average task importance 44 44 33 33 22 22 11 11 Non-routine Non-routine Non-routine Non-routine Routine Non-routineNon-routine Routine Routine Routine Non-routine Non-routine Non-routine Non-routine Routine Non-routine Non-routine Routine Routine Routine analytical interpersonal analytical physical cognitive interpersonal physical cognitive manual manual analytical interpersonal physical analytical interpersonal physical cognitive cognitive manual manual High-skilled High-skilledCitizens Citizens High-skilled High-skilledNon-citizens Non-citizens Low-skilled Low-skilledCitizens Citizens Low-skilled Low-skilledNon-citizens Non-citizens Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Note: The difference in average task importance between Note: The difference in average task importance between citizens and non-citizens are statistically significant across all citizens and non-citizens are statistically significant across all task categories. task categories. Higher shares of migrant workers are employed There are some differences in the composition in jobs with high importance in routine manual of Malaysians and migrant workers employed tasks, while higher shares of Malaysian workers in jobs with high importance in routine tasks by are employed in jobs with high importance in occupational skill level. A few patterns stand out routine cognitive tasks (see Figure 23). A job with when analyzing the shares of Malaysian and non- high importance in a task category is one with a task Malaysian workers employed in jobs with high task importance score within the top three deciles of importance by occupational skill level (see Figure 24). the overall distribution of task importance scores in First, there are high shares of both Malaysians and the country within a given year. About 53.1 percent non-Malaysians employed in low-skilled jobs that of migrant workers are employed in jobs with high have high importance in routine manual tasks. That importance in routine manual tasks, compared to said, a higher share of migrant workers is employed 23.6 percent of Malaysian workers. At the same time, in jobs with high importance in routine manual 17.9 percent of Malaysians workers are employed in tasks. Second, a higher share of Malaysian workers jobs with high importance in routine cognitive tasks, (19 percent vs 4.5 percent of migrant workers) is compared to 4.5 percent of migrant workers. A employed in mid-skilled jobs with high importance higher share of migrant workers is employed in jobs in routine cognitive tasks. Further, a higher share of with high importance in non-routine physical tasks, non-Malaysian workers (34 percent vs 29.2 percent while higher shares of Malaysians are employed of Malaysian workers) is employed in mid-skilled in jobs with high importance in non-routine jobs with high importance in routine manual tasks. interpersonal and analytical tasks. Since routine Together, these data suggest that high shares of tasks are more automatable than non-routine tasks both Malaysian and non-Malaysian workers perform from a technical perspective, these shares provide tasks in jobs at different occupational skill levels that some indication of the susceptibility of the tasks are susceptible to automation. performed by Malaysian and migrant workers to automation. 25 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 23: Share of workers employed in jobs with FIGURE 24: Share of workers employed in jobs with high task importance by citizenship, 2021 (%) high task importance by occupational skill level and citizenship, 2021 (%) 100 100 (%) 100 100 employment (%) 80 of employment 80 Share of employment (%) Share of employment (%) 80 80 60 60 60 60 40 40 Share of 20 20 Share 40 40 00 High-skilled Mid-skilled Low-skilled High-skilled Mid-skilled Low-skilled High-skilled Mid-skilled Low-skilled High-skilled Mid-skilled Low-skilled 20 20 0 0 Non-routine Non-routine Non-routine Routine Routine Citizens Non-citizens Non-routine Non-routine analytical Non-routine interpersonal physical Routine cognitive Routine manual Citizens Non-citizens analytical interpersonal physical cognitive manual Non-routine analytical Non-routine interpersonal Non-routine Non-routine analytical physical cognitive interpersonal Non-routine Routine Citizens Non-citizens Citizens Non-citizens manualphysical Non-routine Routine Routine cognitive Routine manual Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Note: A job that requires high importance in a task category is Note: A job that requires high importance in a task category is that with a task importance score within the top three deciles of that with a task importance score within the top three deciles of the overall distribution of task importance scores in the country the overall distribution of task importance scores in the country within a given year. The shares do not sum to 100 percent by within a given year. The shares do not sum to 100 percent by any measure, since one job can be highly intensive in more than any measure, since one job can be highly intensive in more than one task category. one task category. FIGURE 25: Share of Malaysian citizens employed FIGURE 26: Share of non-Malaysian citizens in jobs with high task importance by sector, 2021 employed in jobs with high task importance by (%) sector, 2021 (%) 100 100 100 100 (%) (%) employment (%) employment (%) of employment of employment 80 80 80 80 60 60 60 60 40 40 40 40 Share of Share of 20 20 Share Share 20 20 0 0 0 0 n n g g e e ess ess g g ion ion ing ing urre urre ing ing ttio ttio ice ice n n n n ltu ltu vic vic urri urri ini ini ucc ucc ult ult n n cttu cttu errv errv icu icu Mi rru Mi rru M M c c Se Se fac fac t t nsst nsst grri grri ufa ufa S S Ag Ag on on nu nu A A Co Co an an C C Ma Ma M M analytical Non-routine analytical Non-routine interpersonal Non-routine interpersonal Non-routine analytical Non-routine analytical Non-routine interpersonal Non-routine interpersonal Non-routine physical Non-routine physical Non-routine cognitive Routine cognitive Routine physical Non-routine physical Non-routine cognitive Routine cognitive Routine manual Routine manual Routine manual Routine manual Routine Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Note: A job that requires high importance in a task category is Note: A job that requires high importance in a task category is that with a task importance score within the top three deciles of that with a task importance score within the top three deciles of the overall distribution of task importance scores in the country the overall distribution of task importance scores in the country within a given year. The shares do not sum to 100 percent by within a given year. The shares do not sum to 100 percent by any measure, since one job can be highly intensive in more than any measure, since one job can be highly intensive in more than one task category. one task category. 26 Migration, Automation, and the Malaysian Labor Market: A summary of findings The shares of Malaysians and migrant workers share of Malaysians employed as professionals (24.4 employed in jobs with high task importance by percent) and technicians and associate professionals sector provides further insight on the structure (20.2 percent) in the mining sector. The mining of employment. High shares of both Malaysian and sector is unique in that it is a capital-intensive sector migrant workers (more than 70 percent) working that employs a small share of labor; the share of in the agriculture sector are employed in jobs with employment in the mining sector was 0.5 percent of high task importance in the non-routine physical and total employment in 2021. Almost half of all workers routine manual tasks, although the shares of migrant employed in the mining sector are employed in high workers are higher (see Figure 25 and Figure 26). skilled occupations. High shares of migrant workers The mining sector employs a very high share of employed in the manufacturing and construction migrant workers in jobs with high importance in non- sectors work in jobs with high importance in routine routine analytical tasks (97.4 percent vs 59.7 percent manual tasks, compared to the relevant shares of of Malaysian workers). This is because the majority Malaysian workers. Finally, a relatively high share (66.6 percent) of migrant workers employed in of Malaysian workers employed in the services the mining sector work in high-skilled occupations sector work in jobs with high importance in routine as professionals, while 12.8 percent work as cognitive tasks (21 percent), compared to migrant technicians and associate professionals. Together, workers (9.4 percent). the sum of these shares is higher than the total FIGURE 27: Share of Malaysian citizens employed in FIGURE 28: Share of non-Malaysian citizens jobs with high task importance by education level, employed in jobs with high task importance by 2021 (%) education level, 2021 (%) 100 100 100 100 (%) employment (%) Share of employment (%) Share of employment (%) ofemployment 80 80 80 80 60 60 60 60 40 40 40 40 Shareof 20 20 Share 20 20 00 00 ry ry l ry aa ry ry ry l aa ry ry aa rm ry aia aa dd ry al rm dd ry ry ay im ay al nn rtrit ay nn ay fo ar rm rirm ar co nd rtair ar fo ee co fom nd im co nd co nd PP TT oo rti Pirm ofor se co ee Te co NN e eo SS Te co -ts- Pr -s c Se No sts s-tse Se N oo PP st Po Po Non-routine analytical Non-routine interpersonal analytical Non-routineanalytical Non-routine Non-routine interpersonal Non-routine interpersonal Non-routine analytical Non-routine interpersonal Non-routine physical Routine cognitive Non-routine physical Non-routine physical Routine cognitive Routine cognitive Non-routine physical Routine cognitive Routine manual Routine manual Routine manual Routine manual Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) (DOSM) Note: A job that requires high importance in a task category is Note: A job that requires high importance in a task category is that with a task importance score within the top three deciles of that with a task importance score within the top three deciles of the overall distribution of task importance scores in the country the overall distribution of task importance scores in the country within a given year. The shares do not sum to 100 percent by within a given year. The shares do not sum to 100 percent by any measure, since one job can be highly intensive in more than any measure, since one job can be highly intensive in more than one task category. one task category. 27 Migration, Automation, and the Malaysian Labor Market: A summary of findings There is evidence that Malaysians with lower in low-skilled jobs with high importance in routine levels of educational attainment have more manual tasks (about 730,000 migrant workers vs opportunities than equally qualified migrant about 643,000 Malaysian workers). Taken together, workers to perform jobs with high importance these numbers suggest that more Malaysians are in non-routine analytical and non-routine employed in jobs – and particularly mid-skilled jobs interpersonal tasks. To elaborate, 40 percent of – with high importance in routine tasks that are more Malaysian workers with secondary education are easily automatable. employed in jobs with high importance in non- routine analytical tasks, compared to 16.3 percent of equally qualified migrant workers (see Figure 5.2 Analyzing the routine task 27 and Figure 28). In parallel, 41.3 percent, 31.1 percent and 28.7 percent of Malaysians workers intensity of jobs based on with no formal education, primary, and secondary Lewandowski, Park and education respectively are employed in jobs with Schotte’s (2023) routine task high importance in non-routine interpersonal tasks, intensity scores compared to 15.9 percent, 11.5 percent and 17.2 percent of migrant workers. At the same time, Consistent with the findings from Section 5.1, higher shares of migrant workers with primary and migrant workers are more likely to be employed in secondary education are employed in jobs with high jobs with higher RTI when routine task content is importance in routine manual tasks, compared to empirically estimated for Malaysia. To first give an Malaysians. Further, a relatively high share of migrant overview of the Malaysian economy, data show that workers with primary and secondary education are more skilled work is associated with a lower RTI (or employed in jobs with high importance in non- involve less routine tasks) in Malaysia. The RTI for routine physical tasks compared to Malaysians. managers in Malaysia is -0.7, compared with 1.1 for Altogether, these findings suggest that Malaysians workers in elementary occupations (see Figure 29). are more likely to work in jobs that are less Analyzing the RTI across countries, Lewandowski susceptible to automation, even when compared et al. (2023) show that this association between to migrant workers with comparable educational occupational skill level and RTI is seen globally. attainment. When the average RTI is calculated by citizenship, it is clear that migrant workers in Malaysia are more In terms of absolute numbers, more Malaysians likely to be employed in more routine jobs, and have are employed in jobs with high importance in a higher average RTI (see Figure 30), suggesting routine tasks that are more easily automatable. that their jobs are at greater risk of automation. This Using data from the 2021 LFS, 1.5 million and 2.3 is also seen across all sectors, with the exception million Malaysians are estimated to be employed of the mining sector (see Figure 31). The majority in mid-skilled jobs with high importance in routine of migrant workers employed in the mining sector cognitive or routine manual tasks respectively.25 In (66.6 percent) work as professionals, which explains comparison, about 37,000 and 276,000 migrant their relatively low RTI for the sector. Given that the workers are estimated to be employed in mid-skilled RTI is calculated at the one-digit occupation level jobs with high importance in routine cognitive or – meaning that a Malaysian and a migrant worker routine manual tasks respectively. There are also employed in the same broad occupation would more Malaysian workers employed in low-skilled have the same RTI – this reflects the relatively jobs with high importance in routine cognitive tasks high share of employment in routine jobs among (about 43,000 vs about 8,500 migrant workers). migrant workers. There are, however, more migrant workers employed 25 It should be noted that a job can be highly intensive in more than one type of task. 28 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 29: RTI by occupation, 2021 FIGURE 30: RTI by citizenship, 2021 1.5 1.5 11 Routine task intensity Routine task intensity 11 0.5 0.8 0.8 0.5 Routine task intensity Routine task intensity 00 0.6 0.6 -0.5 -0.5 -1 -1 0.4 0.4 w erss cu p rrss or rss w llss riiccu es w rss nss r l s ke r er n a pr o na ls n e e rs at fe ss erss tu ra rrkke e ls o io na ar e w r s e al orrke an porrt na r occc rkke n d ra ft rke io pr io na tio ge em e c hi orkke k or rv ic u p ssssio or at o oc Prrof ag o e ss i o ulltu wo ryy o wo ra o pa an a C lw d tw M an w u E d ma ft w rv ll ssu offe l 0.2 e 0.2 M e a nt hin r o p a g sa s c P p an C ill e d n d C le te Ellem ma ta r C ci a Skkil es a i en ag Se rica s le ric so e 00 As s ta ed As c an t i Pllan l Citizens Citizens Non-citizens Non-citizens Se S P Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) and Lewandowski et al. (2023) (DOSM) and Lewandowski et al. (2023) Note: Higher values of the RTI indicate a higher degree of Note: Higher values of the RTI indicate a higher degree of routineness in the tasks performed. routineness in the tasks performed. Analysis using the RTI shows that Malaysians education (see Figure 32). However, migrant workers with up to secondary education have more with a post-secondary or tertiary education are opportunities for employment in non-routine more likely to be employed in less routine jobs. LFS jobs, even when compared to migrant workers data show that this is because higher shares migrant with the same level of educational attainment. workers with a post-secondary or tertiary education This is consistent with findings from the Acemoglu are employed in high-skilled jobs as managers (21.8 and Autor (2011) approach presented in Section percent compared to 7.1 percent of Malaysians) 5.2. Malaysians without any formal education, and and technicians and associate professionals (26.2 those with a primary or secondary education, are percent compared to 17.8 percent of Malaysians). more likely to be employed in jobs that are less In comparison, higher shares of Malaysians with the routine jobs that are less susceptible to automation same level of education are employed in mid- and compared to migrant workers with similar levels of low-skilled jobs. 29 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 31: RTI by sector and citizenship, 2021 FIGURE 32: RTI by educational attainment and citizenship, 2021 11 11 Routine task intensity Routine task intensity Routine task intensity Routine task intensity 0.5 0.5 0.5 0.5 00 00 -0.5 -0.5 -0.5 -0.5 y ar y r m al ry ct o n tu e y rin g al ar y ar y ar ice s n in ng re nd ar g y y s g rtiiar rv ice i m ar ulltur fo rm tu rin da io on d in ru t i co nd M ini Prrim ec on rt st ruc ac tu o o Se rv u co Te Nof M Te ric u f ac Se -s c ric P on st st -se Se N Se an uf Ag C n Ag o M an Po st C Po M Citizens Non-citizens Citizens Non-citizens Citizens Non-citizens Citizens Non-citizens Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) and Lewandowski et al. (2023) (DOSM) and Lewandowski et al. (2023) Note: Higher values of the RTI indicate a higher degree of Note: Higher values of the RTI indicate a higher degree of routineness in the tasks performed. routineness in the tasks performed. All in all, these findings are broadly consistent of automation following the occupation-based with those made using the Acemoglu and Autor approach. It should be noted that this approach has (2011) approach. The advantage of this analysis been criticized in favor of the task-based approach, is that it provides a sound empirical estimate for and its limitations have been discussed in Section 5. identifying the routineness of jobs in Malaysia, This finding is based on the Frey and Osborne (2017) instead of utilizing data from the United States. computerization scores, assigned to occupations The findings are similar, and illustrates that migrant in Malaysia using the Labour Force Survey. In the workers are generally more likely to be employed in same year, 26 percent of workers in Malaysia were routine jobs that are more easily automatable. That employed in jobs that have a medium likelihood said, the relatively low shares of migrant workers – of automation and 22 percent were employed in even in the most routine jobs – suggests that the jobs that have a low likelihood of automation. This automation of such jobs will also have an impact composition has been consistent between 2016 on Malaysians workers. Finally, the relatively high to 2021 (see Figure 33), largely because the same share of employment of highly educated migrant measure for the risk of automation used for all years. workers in less routine, high-skilled jobs may be an Therefore, any changes in the share of jobs by risk indication of skills gaps in Malaysia. of automation is due to changes in the composition of jobs in Malaysia, for which change is slow. This is also regionally in line with other countries in ASEAN, when analyzed using the Frey and Osborne (2017) 5.3 Analyzing the automatability method. In 2016, 50 percent of workers in Malaysia of jobs based on Frey were employed in jobs with high likelihood of and Osborne’s (2017) automation – similar to Philippines (49 percent) and computerization scores the United States (47 percent), lower than Indonesia (56 percent), Germany (59 percent), Vietnam (70 In 2021, 52 percent of the employed in Malaysia percent), Cambodia (57 percent) and higher than were employed in jobs that have high likelihood Thailand (44 percent) (ILO 2016) (see Figure 34). 30 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 33: Share of employment by probability of FIGURE 34: Share of employment with a high automation, 2016-2021 (%) probability of automation by country (%) 100 100 100 100 Share of employment (%) Share of employment (%) 80 80 80 80 Share of jobs (%) Share of jobs (%) 52 60 60 60 60 52 40 40 40 40 2626 20 20 20 20 2222 00 00 G b o d ia M Ph ilip ate s ay sia in s In do 16)) ia G erm dia d y Th lan d M la ilip pi s na m sia (2 es C m si a In 20 16 d nd Vi etn y Un h ail d Ph St te al y p ne m m an an n n V i an es et a o a Fiinla ( 0 C ne ite aila 2016 2017 2016 2018 2019 2017 2018 2020 2021 2019 2020 d St do n am b 2021 er F a T ite Un High High Medium Medium Low Low a Source: World Bank staff calculations using data from LFS Source: Pajarinen and Rouvinen (2014), Brzeski and Burk (2015), (DOSM) and Frey and Osborne (2017) and ILO (2016), Frey and Osborne (2017) Low-skilled jobs, which are more likely to be held have jobs with high likelihood of automation. Given by migrant workers, are estimated to have the the relative shares of employment of Malaysian and highest likelihood of automation. Data show that migrant workers in high-, mid-, and low-skilled jobs 80.5 percent of workers in low-skilled occupations (see Figure 6), this translates into a relatively high have jobs with high likelihood of automation (see share of migrant workers employed in jobs with Figure 35). This is followed by workers in mid-skilled high likelihood of automation. About 75.6 percent occupations, 67 percent of whom have jobs with of migrant workers are employed in jobs with high high likelihood of automation. In comparison, only likelihood of automation, compared to 48 percent 9 percent of workers with high-skilled occupations of Malaysian workers (see Figure 36). FIGURE 35: Share of employment by probability of FIGURE 36: Share of employment by citizenship automation and occupational skill level, 2021 (%) and probability of automation, 2021 (%) 100 100 100.0 100.0 (%) employment (%) (%) employment (%) 80 80 80.0 80.0 of employment of employment 60 60 60.0 60.0 40 40 40.0 40.0 Share of Share of Share Share 20 20 20.0 20.0 00 0.0 0.0 High-skilled Mid-skilled Low-skilled Citizens Citizens Non-citizens Non-citizens High-skilled Mid-skilled Low-skilled Highprobability High probability Mediumprobability Medium probability Lowprobability Low probability Highprobability High probability Mediumprobability Medium probability Low Lowprobability probability Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) and Frey and Osborne (2017) (DOSM) and Frey and Osborne (2017) 31 Migration, Automation, and the Malaysian Labor Market: A summary of findings FIGURE 37: Share of employment of Malaysians by FIGURE 38: Share of employment of non-Malaysians probability of automation and education level (%) by probability of automation and education level (%) 100 100 100 100 Share of employment (%) Share of employment (%) Share of employment (%) Share of employment (%) 80 80 80 80 60 60 60 60 40 40 40 40 20 20 20 20 00 00 ryy ryy arry arry all all y y ar y ar y arry arry rm a rm a daar daar y y y y rttiiar rttiiar a da fo rm fo rm im a im a nd nd nd nd nd Prrim Prrim co n o fo o fo on on r r ec o ec o co Te co Te co Te Te -s ec -s ec P P No No Se Se N N Se Se stt-s stt-s Po s Po s Po Po High Highprobability probability Medium Medium probability Low probability Lowprobability probability High High probability Medium probability Mediumprobability probability Low Lowprobability probability Source: World Bank staff calculations using data from LFS Source: World Bank staff calculations using data from LFS (DOSM) and Frey and Osborne (2017) (DOSM) and Frey and Osborne (2017) Although higher levels of educational attainment of migrant workers. Similarly, lower shares of are generally associated with employment in jobs Malaysian workers with primary, secondary, post- with lower likelihood of automation, Malaysians secondary, and tertiary education are employed in are more likely to be employed in jobs that do not jobs with high likelihood of automation compared have high likelihood of automation, even when to migrant workers. compared to equally qualified migrant workers. This is similar to the findings made in Section 5.1, Given the relative shares of employment of in which Malaysians with lower levels of educational Malaysian and migrant workers, in absolute attainment have more opportunities than equally numbers, there are more Malaysians that qualified migrant workers to perform jobs that are are employed in jobs with high likelihood of intensive in non-routine analytical and non-routine automation based on the Frey and Osborne (2017) interpersonal tasks. Figure 37 and Figure 38 show methodology. Using the numbers of Malaysians that both Malaysian and migrant workers with post- and migrant workers reported in LFS 2021, about secondary and tertiary education are least likely 6.3 million Malaysian workers are estimated to be to be employed in jobs with high likelihood of employed in jobs with high likelihood of automation. automation. However, across all levels of education, This is more than four times more than the 1.4 million migrant workers are more likely to be employed migrant workers that are estimated to be employed in jobs with high likelihood of automation. For in jobs with high likelihood of automation. While instance, 47 percent of Malaysians without any this is an important estimation, the caveats of this formal education are employed in jobs with high approach (discussed above) should be noted. likelihood of automation, compared to 75.3 percent 32 Migration, Automation, and the Malaysian Labor Market: A summary of findings 6. Summary of findings and concluding remarks Migrant workers have long played a prominent than the number of migrant workers, simply due to role in the Malaysian economy, and empirical the relative size of the Malaysian workforce. analysis shows that there is a generally positive It is important to note that the automation of relationship between the employment of migrant tasks does not necessarily imply a decrease in the workers and the employment outcomes of need for labor, foreign or otherwise. As previously Malaysian workers. More specifically, this paper discussed, automation changes the task content finds that the employment of migrant workers being performed by labor, and can displace labor is associated with greater employment and less (through a “displacement effect”) as capital takes unemployment among Malaysians. The positive over these tasks (Acemoglu and Restrepo 2019). relationship between immigration and employment However, automation can also increase productivity, is larger among those who are aged 30 and older and subsequently increase the demand for labor and among those with a secondary education, in non-automated tasks (through a “productivity suggesting complementarity between migrant effect” (Acemoglu and Restrepo 2019). Hence, workers and Malaysian workers. Indeed, increasingly the net impact of automation on labor demand – large shares of migrant workers hold low-skilled and subsequently wages – depends on “how the jobs, barring which may induce labor gaps due to displacement and productivity effects weigh against a rise in educational attainment among Malaysians. each other” (Acemoglu and Restrepo 2019, p. 4). Migrant workers are more likely to perform Thus, it should not be assumed that an increased tasks that are at high risk of automation, but utilization of automation technologies will decrease the relatively large Malaysian workforce means the net demand for migrant workers. that the adoption of automation technologies There are other reasons to be cautious about the can potentially have a greater impact on the potential of automation to decrease the need Malaysian workforce. Based on the analysis on for foreign workers. First, automation is likely the task content of jobs using the Acemoglu and to happen gradually. The existence of relevant Autor (2011) approach, this paper finds that both technologies is insufficient for their effective Malaysians and migrant workers are exposed to adoption by all firms at the same time, given the jobs that are intensive in routine tasks that are more costs and benefits of automation, as well as the easily automatable. A higher share of Malaysians role of laws and regulations in governing the is employed in jobs with high task importance adoption of new technologies (World Bank 2020b). in routine cognitive tasks, while a higher share of Globally, the adoption of automation technologies migrant workers is employed in jobs with high task has been slower than expected (World Economic importance in routine manual tasks. Analysis using Forum 2023). Second, as mentioned above, even the RTI data from Lewandowski et al. (2023) and the though automation may decrease the demand for Frey and Osborne (2017) approach, however, clearly some jobs, it is likely to increase the demand for illustrate that migrant workers are more likely to be others, and particularly those that can complement employed in jobs that are routine and with high automation technologies. Third, Malaysia is rapidly likelihood of automation respectively.26 Overall, aging. Migrant workers, who are younger than Malaysians do indeed have greater access to jobs Malaysian workers on average, will be an increasingly with tasks that are less susceptible to automation, important resource as Malaysia’s working age even when compared to migrant workers with similar population continues to shrink. Taking all of these qualifications. That said, in absolute terms, the into consideration, the net effect of automation on number of Malaysian workers who can be affected the demand for foreign workers in the near future is by automation technologies is substantially higher unclear. This is especially the case when one considers 26 The limitations of the Frey and Osborne (2017) approach which have been discussed in Section 5 should be noted. 33 Migration, Automation, and the Malaysian Labor Market: A summary of findings the growth of jobs that can complement automation (World Bank 2020b). One of the ways to do this is technologies and jobs that are not automatable, to implement the planned multi-tiered levy system, such as jobs in the care sector. Importantly, studies while ensuring that the setting and adjusting of on the barriers to mechanization and automation in these levies are transparent, evidence-based, and Malaysia find that among the main barriers include frequently reviewed. The levy would also increase the prohibitive costs of acquiring and maintaining the cost of hiring migrant workers, which can address such technologies, especially for small and medium the concern that the availability of low-skilled and enterprises. These would need to be addressed low-wage migrant workers is impeding automation. to increase technology adoption, independent of Strengthening the foreign worker management migration policy. system opens the door to economic migration that can be a “win-win” for both sending and receiving Given that adopting automation technologies will countries. change or displace more jobs held by Malaysians, encouraging the adoption of automation A better understanding of the perception towards technologies should be accompanied by active migration and migrant workers can inform the labor market policies that can support workers social and political feasibility of migration. The in a changing world of work. Strengthening hiring of foreign workers in Malaysia has long been lifelong learning systems, including the provision a contentious issue, despite evidence on the lack of training for skills that can complement advanced of adverse impacts on Malaysian workers. Ensuring technologies such as socioemotional skills and that migration is demand driven may address the advanced digital skills will be crucial (World Bank economic feasibility of migration, but addressing its 2020b). Equally important is ensuring that access social and political feasibility will require addressing to training is available to all workers, including public perception towards migration. The first step informally employed workers who are less likely would be to ensure that empirical findings on the to benefit from training conducted by employers. relationship between the hiring of migrant workers Relatedly, an enhanced social insurance system and the labor market outcomes of Malaysians are which includes informally employed workers can known, both within government and by the public. support workers who are displaced in the process The narrative that migrant workers decrease job of technology adoption. Additionally, a better opportunities for Malaysians is pervasive and should understanding of the barriers to implementing new be countered by evidence, even if the evidence technologies faced by firms can support efforts goes against perception. More research will be towards increasing technology adoption. required to understand the perceived costs of migration, but data from the World Values Survey Moving forward, it would be worthwhile to provides some insight. In the most recent wave of strengthen the foreign worker management the World Values Survey, conducted between 2017 system to ensure that it is demand driven. and 2019, about 51 percent of Malaysians aged 15 While investments in automation technologies and older mentioned that they would not like to may change the type of migrant workers required have immigrants or foreign workers as neighbors. by Malaysia in the future, it will not eliminate the This illustrates a negative sentiment towards demand for migrant workers. For example, in the migrant workers than extends beyond economic life. context of rapid aging, South Korea is among Understanding the reasons behind this sentiment the most automated economies globally, with its will be important to be able to address concerns manufacturing industry having the highest robot regarding migration and migrant workers. density in the manufacturing sector in the world in 2022 (IFR 2023). In parallel, it has been making There are other avenues for future research reforms to attract more migrants (Chung 2021). that can deepen understanding on the complex Therefore, it is important for Malaysia to ensure that interplay between migration and automation. This its foreign worker management system is responsive includes conducting a firm technology adoption to economic needs. At present, Dependency survey that aims to directly explore questions ceilings and levies are not well-suited to ensuring related to the substitutability or complementarity the admittance of the right number of foreign between the adoption of automation technologies workers, or foreign workers with the right skills and the hiring of migrant workers. Alternatively, 34 Migration, Automation, and the Malaysian Labor Market: A summary of findings the relationship can be studied through qualitative as well as the potential impacts of aging on the sectoral analysis on subsectors that are of particular labor market, including both native and migrant interest given the relative concentration of migrant workers. Such evidence can support policymakers workers. Relatedly, there is room to further in navigating the changing employment landscape, understand the potential impacts of automation, towards a more productive Malaysia. 35 Migration, Automation, and the Malaysian Labor Market: A summary of findings References Abdur Rahman, A., & Schmillen, A. (2023). From Farms to Factories and Firms—Structural Transformation and Labor Productivity Growth in Malaysia. Asian Economic Papers, 22(1), 82-111. Accetturo, A., Bugamelli, M. & Lamorgese, A. (2012) Welcome to the machine: firms’ reaction to low-skilled immigration. Bank of Italy Temi di Discussione (Working Paper) No. 846. Acemoglu, D. & Autor, D. (2011). Skills, tasks and technologies: Implications for employment and earnings. In Handbook of Labor Economics (Vol. 4, pp. 1043-1171). Elsevier. Acemoglu, D. & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2), 3-30. Acemoglu, D. & Restrepo, P. (2020). Robots and jobs: Evidence from US labor markets. Journal of Political Economy, 128(6), 2188-2244. Acemoglu, D., Koster, H. R., & Ozgen, C. (2023). Robots and workers: Evidence from the Netherlands (No. w31009). National Bureau of Economic Research. Ahsan, A., Abella, M., Beath, A., Huang, Y., Luthria, M., Nguyen, T. V. (2014). International Migration and Development in East Asia and the Pacific. Washington, DC: World Bank. Aghion, P., Antonin, C., Bunel, S., & Jaravel, X. (2020). What Are the Labor and Product Market Effects of Automation? New Evidence from France. CEPR Discussion Paper DP14443. Paris. Ali Ahmad, Z., Simler, K. & Yi, S. (2020). Using Remittance Transaction Data for Timely Estimation of the Foreign Worker Population in Malaysia. World Bank Policy Research Paper 9331. Washington, DC: The World Bank. Ang, J. W., Murugasu, A. & Chai, Y. W. (2018). Low-Skilled Foreign Workers’ Distortions to the Economy. Kuala Lumpur: Bank Negara Malaysia. Anwar Ibrahim. (2023). Ekonomi Madani: Memperkasa Rakyat. Transcript of speech delivered at Securities Commission Malaysia, July 27, 2023. https://malaysiamadani.gov.my/wp-content/uploads/2023/07/FINAL-ENGLISH- Translation-Ucapan-Kerangka-Ekonomi-MADANI-3.pdf Arntz, M., Gregory, T., & Zierahn, U. The Risk of Automation for Jobs in OECD Countries: A Comparative Analysis. OECD Social, Employment and Migration Working Papers No. 189. Athukorala, P. C., & Devadason, E. S. (2012). The impact of foreign labor on host country wages: The experience of a southern host, Malaysia. World Development, 40(8), 1497-1510. Autor, D. H., Levy, F., & Murnane, R. J. (2003). The skill content of recent technological change: An empirical exploration. The Quarterly Journal of Economics, 118(4), 1279-1333. Azizah Kassim (2009). Filipino Refugees in Sabah: State Responses, Public Stereotypes and the Dilemma Over Their Future. Southeast Asian Studies, 47(1): 52–88. Azizah Kassim (2014). Recent Trends in Transnational Population Inflows into Malaysia: Policy, Issues and Challenges. Malaysian Journal of Economic Studies, 51(1), 9–28. Beylis, G., Jeff, R. F. Morris, M., Sebastian, A. R. & Sinha, R. (2020). Going Viral: COVID-19 and the Accelerated Transformation of Jobs in Latin America and the Caribbean. Washington, DC: World Bank. BNM (2021). Economic and Monetary Review 2020. Kuala Lumpur: Bank Negara Malaysia. Brzeski, C. & Burk, I. (2015). The robots come. Consequences of automation for the German labour market. ING DiBa Economic Research. Chung, E. A. (2021). How South Korean Demographics Are Affecting Immigration and Social Change, in: Lee, C. M. & Botto, K., Demographics and the Future of South Korea. Washington, DC: Carnegie Endowment for International Peace. Coelli, M. & Borland, J. Behind the headline number: Why not to rely on Frey and Osborne’s predictions of potential job loss from automation. Melbourne Institute Working Paper No. 10/19. Melbourne: Melbourne Institute. Dustmann, C., Fabbri, F. & Preston, I. (2005). The impact of immigration on the British labour market. The Economic Journal, 115(507), F324-F341. EPU (2021). Twelfth Malaysia Plan 2021-2025. Putrajaya: Economic Planning Unit. Frey, C. B., & Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerization? Technological forecasting and social change, 114, 254-280. Graetz, G., & Michaels, G. (2018). Robots at work. Review of Economics and Statistics, 100(5), 753-768. 36 Migration, Automation, and the Malaysian Labor Market: A summary of findings Humlum, A. (2019). Robot adoption and labor market dynamics. Princeton University. IFR (2022). World Robotics: Industrial Robots 2021. Frankfurt: International Federation of Robotics. IFR (2023). World Robotics: Industrial Robots 2022. Frankfurt: International Federation of Robotics. ILMIA (2018). Developing Multi-Tier Levy Malaysian Case Study. Selangor: Institute of Labour Market Information and Analysis. ILO (2012). International Standard Classification of Occupations: Structure, group definitions and correspondence tables. Geneva: International Labour Office. Kamaruddin, S. S., Mohammad, M. F., & Mahbub, R. (2016). Barriers and impact of echanization and automation in construction to achieve better quality products. Procedia-Social and Behavioral Sciences, 222, 111-120. Koch, M., Manuylov, I., & Smolka, M. (2021). Robots and firms. The Economic Journal, 131(638), 2553-2584. KRI (2018). The Unsung Labour: Care Migration in Malaysia. Discussion Paper 1/18. Kuala Lumpur: Khazanah Research Institute. Lewandowski, P., Park, A., Hardy, W., Du, Y., & Wu, S. (2022). Technology, skills, and globalization: Explaining international differences in routine and nonroutine work using survey data. The World Bank Economic Review, 36(3), 687-708. Lewandowski, P., Park, A. & Schotte, S. (2023). The global divergence in the routinization of jobs, in: Gradín C., Lewandowski P., Schotte S., & Sen K., Tasks, Skills, and Institutions - The Changing Nature of Work and Inequality. Oxford: Oxford University Press. Lewis, E. (2011). Immigration, skill mix, and capital skill complementarity. The Quarterly Journal of Economics, 126(2), 1029-1069. Mahbub, R. (2012). Readiness of a developing nation in implementing automation and robotics technologies in construction: A case study of Malaysia. Journal of Civil Engineering and Architecture, 6(7), 858-866. Mann, K. & Pozzoli, D. (2022). Automation and Low-Skill Labor. IZA Discussion Paper No. 15791. Marcolin, L., Miroudot, S., & Squicciarini, M. (2019). To be (routine) or not to be (routine), that is the question: a cross- country task-based answer. Industrial and Corporate Change, 28(3), 477-501. MIDA (n.d.). How AI is revolutionizing the industry. Putrajaya: Malaysian Investment Development Authority. Accessed from: https://www.mida.gov.my/how-ai-is-revolutionising-the-industry/ MOSTI (2021). The Malaysia Robotic Roadmap 2021-2025. Putrajaya: Ministry of Science, Technology and Innovation. Oxford Insights (2023). Government AI Readiness Index 2023. London: Oxford Insights. Özden, Ç. & Wagner, M. (2014). Immigrant versus natives? Displacement and job creation. Policy Research Working Paper 6900. Washington, DC: The World Bank. Pajarinen, M., & Rouvinen, P. (2014). Computerization threatens one third of Finnish employment. Etla Brief, 22(13.1), 2014. Ratnasingam, J., Ab Latib, H., Yi, L. Y., Liat, L. C., & Khoo, A. (2019). Extent of automation and the readiness for industry 4.0 among Malaysian furniture manufacturers. BioResources, 14(3), 7095-7110. Spitz-Oener, A. (2006). Technical change, job tasks, and rising educational demands: Looking outside the wage structure. Journal of Labor Economics, 24(2), 235-270. Tan, T. T. & Ng, A. (2018). The economic impact of foreign workers in Malaysia: A brief econometric investigation. Kuala Lumpur: Khazanah Research Institute. World Bank (2015). Malaysia Economic Monitor: Immigrant Labour. Washington, DC: The World Bank. World Bank (2018). The Automatability of Occupations in Malaysia: Automatability Profiles of Occupations on the 2017/2018 Critical Occupations List. Washington, DC: The World Bank. World Bank (2020a). A Silver Lining: Productive and Inclusive Aging for Malaysia. Washington, DC: The World Bank. World Bank (2020b). Aiming High: Navigating the Next Stage of Malaysia’s Development. Washington, DC: The World Bank. World Bank (2020c). Indonesia’s Occupational Tasks and Skills: From occupational employment demand to tasks and skills requirements. Washington, DC: World Bank. World Economic Forum (2023). Future of Jobs Report 2023. Geneva: World Economic Forum. Wurscher, I. (2018). Refugee and Asylum-Seekers in Malaysia: The Consequences of Invisibility. International Law Review, College of Law, Michigan State University. Yap, J. B. H., Lam, C. G. Y., Skitmore, M., & Talebian, N. (2022). Barriers to the adoption of new safety technologies in construction: A developing country context. Journal of Civil Engineering and Management, 28(2), 120-133. 37 Migration, Automation, and the Malaysian Labor Market: A summary of findings Annex 1: Analyzing the relationship between the share of migrant workers with labor market outcomes of Malaysians TABLE A1: Effects of immigration on labor force participation rate, employment rate, unemployment rate, and wages of Malaysian citizens Instrumental Dependent variable: First OLS variable (IV) Labor market outcomes of Malaysians differences (preferred) 0.11 0.13 0.032 Labor force participation rate (0.12) (0.15) (0.26) 0.23* 0.12 0.21 Employment rate (0.12) (0.13) (0.23) -0.21*** 0.0015 -0.29* Unemployment rate (0.039) (0.059) (0.12) Hourly wages -0.45 0.099 -0.11 (in natural log form) (0.31) (0.61) (0.95) Hourly wages including overtime -0.59* 0.098 0.000094 (in natural log form) (0.27) (0.52) (0.88) Number of observations 192 176 128 Notes: Robust standard errors are in parentheses. *p<0.05, **p<0.01, ***p<0.001. The coefficients presented here are for the independent variable capturing the ratio of non-citizens to citizens. Regressions control for age, educational attainment, state, and year. This analysis follows that from Dustmann, Fabbri The goal of the analysis is to estimate the causal and Preston (2005) which analyzed the impact of effect of a higher concentration of non-citizens on immigration on the British labor market. The model labor market outcomes. The difficulty is that non- can be elaborated as follows: citizens do not necessarily immigrate to a given state it = 2 it + 3 ln ( _ ) it + 4 _ it + 5 _ it + t + i + it where i stands for state and t for year, O randomly, but they are driven by economic conditions is the outcome variable of interest (labor across the states. These are likely correlated with force participation rate, employment rate, the labor market outcomes, leading to an omitted unemployment rate, and hourly wages, all of which variable bias in the ordinary least squares (OLS) are for Malaysian citizens only), fr is the ratio of estimates. The instrumental variable (IV) approach is non-citizens to citizens in a given state, in a given designed to overcome the omitted variable bias by year, skill_ratio is a vector for skills of citizens using the past levels of immigrant concentrations as consisting of three variables, namely the number an instrument. The underlying assumptions are that of citizens with advanced, low-intermediate, these levels are uncorrelated with current economic and high-intermediate education divided by the outcomes but still explain some of the current number of unqualified citizens, age_cit and age_ migration through various cultural linkages. The noncit capture the average ages of citizens and implemented IV analysis therefore uses the lags of non-citizens divided by 100, year and state reflect the levels of the instrumented variables, which are year and state fixed effects. expressed as the first differences. More specifically, 38 Migration, Automation, and the Malaysian Labor Market: A summary of findings the ratio of non-citizens to citizens in a given state in terms of their first differences. The results of the and year and the natural log of the skill ratio of Sargan overidentification test generally suggest citizens (all expressed in terms of first differences) that instruments are valid and are not correlated are instrumented by variables capturing their three- with the error term. However, the test is weak in the and four-year lags (all in terms of their levels). All of cases when the labor force participation rate and the the other variables in the IV analysis are expressed employment rate are the outcome variables. TABLE A2: Effects of immigration on labor force participation rate, employment rate, unemployment rate, and wages of Malaysian citizens, taking into account education level Labor force participation rate Ratio of non-citizens to citizens 0.24*** -0.23 -0.79 (0.034) (0.46) (0.60) Interaction terms Lower-intermediate education × -0.16*** 0.27 0.85 Ratio of non-citizens to citizens (0.032) (0.47) (0.85) Upper-intermediate education × -0.089* 1.50** 2.31** Ratio of non-citizens to citizens (0.042) (0.55) (0.86) Advanced education × -0.14*** -0.34 0.23 Ratio of non-citizens to citizens (0.035) (0.56) (0.86) Employment rate Ratio of non-citizens to citizens 0.20*** -0.16 -0.46 (0.033) (0.42) (0.60) Interaction terms Lower-intermediate education × -0.18*** 0.18 0.69 Ratio of non-citizens to citizens (0.031) (0.44) (0.85) Upper-intermediate education × -0.13** 1.36** 1.98* Ratio of non-citizens to citizens (0.045) (0.52) (0.85) Advanced education × -0.13*** 0.26 -0.063 Ratio of non-citizens to citizens (0.037) (0.59) (0.85) 39 Migration, Automation, and the Malaysian Labor Market: A summary of findings TABLE A2 (cont.): Effects of immigration on labor force participation rate, employment rate, unemployment rate, and wages of Malaysian citizens, taking into account education level Instrumental Dependent variable: First OLS variable (IV) Labor market outcomes of Malaysians differences (preferred) Unemployment rate Ratio of non-citizens to citizens 0.072*** -0.066 -0.60* (0.012) (0.12) (0.28) Interaction terms Lower-intermediate education × 0.017 0.092 0.33 Ratio of non-citizens to citizens (0.012) (0.14) (0.40) Upper-intermediate education × 0.028 0.019 0.40 Ratio of non-citizens to citizens (0.016) (0.21) (0.40) Advanced education × -0.043** 0.081 0.58 Ratio of non-citizens to citizens (0.015) (0.22) (0.40) Hourly wages (in natural log form) Ratio of non-citizens to citizens -0.22 -1.64 -0.28 (0.12) (1.20) (1.77) Interaction terms Lower-intermediate education × 0.32* 1.14 0.44 Ratio of non-citizens to citizens (0.14) (1.35) (2.51) Upper-intermediate education × 0.36** 1.34 0.20 Ratio of non-citizens to citizens (0.13) (1.47) (2.51) Advanced education × 0.31* 3.05 0.39 Ratio of non-citizens to citizens (0.12) (1.76) (2.51) Hourly wages including overtime (in natural log form) Ratio of non-citizens to citizens -0.24 -2.24 -0.83 (0.13) (1.67) (1.97) Interaction terms Lower-intermediate education × 0.32* 1.92 1.34 Ratio of non-citizens to citizens (0.15) (1.77) (2.78) Upper-intermediate education × 0.36* 2.06 0.95 Ratio of non-citizens to citizens (0.14) (1.90) (2.79) Advanced education × 0.33* 3.46 0.61 Ratio of non-citizens to citizens (0.13) (2.13) (2.79) Ratio of non-citizens to citizens 768 704 512 Notes: Robust standard errors are in parentheses. *p<0.05, **p<0.01, ***p<0.001. The coefficients presented here are for the independent variable capturing the ratio of non-citizens to citizens. Regressions control for age, educational attainment, state, and year. 40 Migration, Automation, and the Malaysian Labor Market: A summary of findings This analysis extends the one above by taking into unqualified citizens, age_cit and age_noncit capture account the education level of workers. The model the average ages of citizens and non-citizens divided can be elaborated as follows: by 100, year reflects year fixed effects. igt = 1+ 2 * it + 3 _ g * it + 4 _ igt + 5 _ it + 6 _ g * t + itg where i stands for state, g for skill groups (based The IV analysis uses the lags of the levels of the on educational attainment), and t for year, O instrumented variables, which are expressed as the is the outcome variable of interest (labor force first differences. More specifically, the ratio of non- participation rate, employment rate, unemployment citizens to citizens in a given state and year and rate, and hourly wages, all of which are for Malaysian its interaction with the skill group (all expressed citizens only), skill_group represents the vector of the in terms of first differences) are instrumented four education groups of citizens (advanced, low- by variables capturing their three- and four-year intermediate, high-intermediate, unqualified) fr is lags (all in terms of their levels). All of the other the ratio of non-citizens to citizens in a given state, in variables in the IV analysis are expressed in terms a given year, skill_ratio is a vector for skills of citizens of their first differences. The results of the Sargan consisting of three variables, namely the number of overidentification test generally suggest that the citizens with advanced, low-intermediate, and high- null hypothesis that instruments are valid and are intermediate education divided by the number of not correlated with the error term. 41 Migration, Automation, and the Malaysian Labor Market: A summary of findings Annex 2: Acemoglu and Autor’s (2011) task-based approach To implement the task-based approach to estimate Where multiple ISCO 2008 occupations map to probabilities of automation for Malaysia, this report a single SOC 2010 occupation, the task score for draws on the framework and methodology proposed the single SOC 2010 occupation is mapped to all by Acemoglu and Autor (2011). The authors identify corresponding ISCO-2008 occupations. There are 16 measures of the importance of different tasks 31 non-military ISCO 2008 occupations do not have in different occupations. These tasks are captured task scores based on this mapping. 28 of these 31 by five categories of routine and non-routine tasks: ISCO 2008 occupations are manually mapped to non-routine cognitive analytical tasks, non-routine SOC 2010 based on the SOC 2010 and ISCO 2008 cognitive interpersonal tasks, and non-routine occupation titles and descriptions. Third, the scores manual physical tasks, which are less susceptible are mapped from ISCO-2008 occupations to MASCO to automation; and routine cognitive and routine 2013’s non-military occupations using a crosswalk manual tasks, which are more susceptible to developed by the World Bank in partnership with automation (see Table A3). TalentCorp based on MASCO 2013 and ISCO 2008 occupation titles and descriptions. Where multiple The importance scores for tasks are obtained from ISCO 2008 occupations map to a single MASCO the online O*NET database and are mapped to 2013 occupation, the closest match from ISCO 2008 the O*NET SOC 2019 occupational classification is selected based on the MASCO 2013 and ISCO scheme. 837 of the 1,016 O*NET SOC 2019 2008 occupation titles and descriptions. Where occupations have data on the task measures multiple MASCO 2013 occupations map to a single included in Acemoglu and Autor (2011). Similar ISCO 2008 occupation, the task scores for the ISCO to the above exercise, to assign the relevant 2008 occupation is mapped to all corresponding occupational task scores, the O*NET SOC 2019 are MASCO 2013 occupations. The O*NET scores for mapped to MASCO 2013. each MASCO 2013 occupation are also matched First, the O*NET SOC 2019 8-digit occupational to the 2010-2021 Malaysian Labour Force Survey classification scheme is mapped to O*NET SOC (LFS) and the 2010-2021 Salary and Wages Survey 2010 (8-digit level)27 using a crosswalk provided (SWS), both of which include 4-digit MASCO 2013 by O*NET.28 Where multiple O*NET SOC 2019 occupation codes. (8-digit level) occupations map to a single SOC Finally, the scores on each of the 16 measures 2010 (6-digit level) occupation,29 the task scores are averaged within the five categories of routine are averaged to obtain a single score. Second, and non-routine tasks to derive a composite score the SOC 2010 occupations are mapped to ISCO for each task category and year, weighted by the 2008 using a crosswalk file provided by the United share of employment in the year. This allows for an States Bureau of Labour Statistics (see Footnote analysis of the relative importance of the different 31). Where multiple SOC 2010 occupations map to task categories, taking into account the share of a single ISCO 2008 occupation, a simple average jobs that involve the different tasks. of the task scores is taken to obtain a single score. 27 This step is performed as the subsequent occupational code transformations to ISCO-08 and MASCO 2013 are linked to SOC 2010. 28 See https://www.onetcenter.org/taxonomy/2019/code_change.html 29 To translate the scores into MASCO 2013, O*NET SOC 2019 need to be translated into O*NET SOC 2010, and then translate to SOC 2010. O*NET SOC 2010 occupational classification scheme is similar to but distinct from the SOC-2010 classification scheme. See https://www.onetcenter.org/taxonomy/2019/soc.html 42 Migration, Automation, and the Malaysian Labor Market: A summary of findings TABLE A3: O*NET task measures used to construct skill categories Task category O*NET task measure Analyzing data or information Non-routine cognitive analytical Thinking creatively Interpreting the meaning of information for others Establishing and maintaining interpersonal relationships Non-routine cognitive Guiding, directing, and motivating subordinates interpersonal Coaching and developing others Manual dexterity Spatial orientation Operating vehicles, mechanized devices, or equipment Spend time using your hands to handle, control, or feel objects, tools or controls Importance of repeating same tasks Routine cognitive Importance of being exact or accurate Structured versus unstructured work Controlling machines and processes Routine manual Pace determined by speed of equipment Spend time making repetitive motions Source: O*NET based on Acemoglu and Autor (2011) 43 Migration, Automation, and the Malaysian Labor Market: A summary of findings Annex 3: Schotte, Park and Lewandowski’s (2023) routine task intensity (RTI) scores The content for this annex was adapted from the World Bank’s Skill toward Employment and Lewandowski et al. (2023), which should be referred Productivity (STEP) surveys, and the Chinese Urban to for a comprehensive understanding of the Labor Survey (CULS) to measure the task content methodology behind the RTI. of jobs. Table A4 shows the survey task items from the PIAAC used to calculate task content measures. The RTI was first developed by Lewandowski et Importantly, these task content measures are al. (2022), using survey task items from the PIAAC, consistent with O*NET occupation task measures. TABLE A4: Overview of survey task items from PIAAC, STEP, and CULS selected by Lewandowski et al. (2022) to calculate task content Non-routine Non-routine Task content cognitive cognitive Routine cognitive Manual analytical interpersonal Task items • Solving • Supervising • Changing • Physical tasks problems others order of tasks – reversed (not • Reading news • Making speeches able) (at least once a or giving month) presentations • Filling out forms (any frequency) (at least once a • Reading month) professional journals (at least • Making speeches once a month) or giving presentations – • Programming reversed (never) (any frequency) Source: Adopted from Lewandowski et al. (2023) Notes: The cut-offs for the “yes” dummy are in parentheses. See Lewandowski et al. (2022) for full details on the measurement of the task content. 44 Migration, Automation, and the Malaysian Labor Market: A summary of findings Using these task content measures, Lewandowski For each occupation, the model that fits the data et al. (2022) defined a composite measure of best from a set of seven alternatives that differ in routine task intensity (RTI), which increases with explanatory variables was selected. Lewandowski the importance of routine content of work, and et al. (2023) use leave-one-out cross-validation, and decreases with the importance of non-routine selected models that exhibit the lowest root mean ( ) content of work. The formula for RTI is as follows: square errors, the lowest mean absolute errors, and + = ln ( ) _ ln analytical personal cog 2 where rcog , nranalytical, and nrpersonal are routine cognitive, (with one exception) the highest pseudo-R-squared. non-routine cognitive analytical, and non-routine Specifications that are consistent with the findings cognitive interpersonal tasks levels respectively. For of worker-level regressions in Lewandowski et al. each task, the lowest score in the sample is added (2022) were prioritized. to the scores of all individuals, plus 0.1, to avoid non-positive values in the logarithm. The adjusted R-squared values of the chosen specifications for the correlates of RTI at the The task content of occupations in countries with 1-digit occupation level ranges between 0.128 no available survey data on tasks is estimated with a (for elementary occupations, ISCO 9) to 0.408 set of ordinary least squares (OLS) regressions that (for agricultural workers, ISCO 6). The adjusted relate the RTI of occupation j in country c to four key R-squared values for the regressions at the 2-digit factors: (1) development level, measured by the gross occupation level were not disclosed by Lewandowski domestic product (GDP) per capita (in purchasing et al. (2023). power parity, natural logarithm); (2) technology use (T), approximated by the number of internet uses per 100 inhabitants; (3) globalization (G), quantified by the foreign value added share of domestic output; and (4) supply of skills (S), measured by the average years of schooling. Fixed effects γ kj are added for 2-digit ISCO sub-occupations k that belong to a giving 1-digit occupation j. The regression model is expressed formally as follows: kjc = j0 + j1 c + j2 c + j3 c + j4 c + kj + kjc 45 Migration, Automation, and the Malaysian Labor Market: A summary of findings Annex 4: Frey and Osborne’s (2017) occupation-based approach To implement the occupation-based approach to Following the methodology of World Bank (2018), estimate the automatability of jobs in Malaysia, first, the available 702 SOC-2010 at 6-digit level this report draws on the findings and methodology occupations are mapped to ISCO- 2008 using of Frey and Osborne (2017) and the World Bank a crosswalk file provided by the United States (2018) report “The Automatability of Occupations Bureau of Labour Statistics.30 Where multiple SOC in Malaysia: Automatability Profiles of Occupations 2010 occupations map to a single ISCO 2008 on the 2017/2018 Critical Occupations List”. In occupation, a simple average of the automatability Frey and Osborne (2017), the authors argue that probabilities is taken to obtain a single score. Where automation is constrained in tasks where there are multiple ISCO 2008 occupations map to a single engineering bottlenecks to computerization, such SOC 2010 occupation, the automatability score as those involving perception and manipulation, for the single SOC 2010 occupation is mapped to social and creative intelligence, to computerization. all corresponding ISCO 2008 occupations. Where Utilizing expert opinions, the probability of no probability is calculated from this exercise, the automation of 702 occupations in the United States ISCO and SOC titles are manually examined and is estimated. This approach has been extended in the corrected. abovementioned report in the Malaysian context, by mapping these probabilities of automation or Next, the scores are mapped from ISCO 2008 “computerization” to occupations in Malaysia. As occupations to MASCO 2013’s non-military in Frey and Osborne (2017), and the World Bank occupations using a crosswalk created by the World (2018) report, we consider an occupation at a “low” Bank in partnership with TalentCorp, which is based risk of automation if the probability assigned is 30 on the occupation titles and descriptions of both percent or less, a “high” risk of automation if at 70 MASCO 2013 and ISCO 2008. Where multiple percent or more and “medium” if in between this ISCO 2008 occupations map to a single MASCO range. 2013 occupation, the closest match from ISCO 2008 is selected based on the MASCO 2013 and ISCO Frey and Osborne (2017) provide automation 2008 occupation titles and descriptions. Where probabilities for occupations using the Standard multiple MASCO-2013 occupations map to a single Occupational Classification System (SOC) 2010 ISCO 2008 occupation, the automatability score classification (which is the occupational classification for the single ISCO-2008 occupation is mapped to scheme utilized by federal statistical agencies in the all corresponding MASCO 2013 occupations. The United States) at the 6-digit level. This is mapped to automatability probability for each MASCO 2013 the Malaysia Standard Classification of Occupations occupation is matched to the 2010-2021 Malaysian (MASCO) 2013, which is based on the International Labour Force Survey (LFS) and the 2010-2021 Salary Standard Classification of Occupations (ISCO) with and Wages Survey (SWS), both of which include revisions to meet the specific requirements of the 4-digit MASCO 2013 occupation codes. Malaysian labour market. 30 See https://www.bls.gov/soc/soccrosswalks.htm (last accessed August 9, 2023). 46