Trade and Inequality in Europe and the US

IZA DP No. 14914 DECEMBER 2021 Trade and Inequality in Europe and the US1 The share of low-income countries in global exports nearly tripled between 1990 and 2015, driven largely by the rapid emergence of China as an exporting powerhouse. While research in economics had long acknowledged that trade with lower-income countries could raise income inequality in Europe and the US, empirical estimates indicated only a modest contribution of trade to growing national skill premia. However, if workers are not highly mobile across firms, industries and locations, then the unequal impacts of trade can manifest along different margins. Recent evidence from countries across Europe and the US shows that growing import competition from China differentially reduced earnings and employment rates for workers in more trade-exposed industries, and for the residents of more trade-exposed geographic regions. These adverse impacts were often largest for lower-skilled individuals. We show that domestic manufacturing employment declined much more in countries that saw a large growth of net imports from China (such as the UK and the US), than in countries that maintained relatively balanced trade with China (such as Germany and Switzerland). Drawing on a new analysis for the UK, we further show that trade with China contributed to job loss in manufacturing, but also to substantial declines in consumer prices. However, while the adverse labour market impacts were concentrated on specific groups of workers and regions, the consumer benefits from trade were widely dispersed in the population, and appear similarly large for highincome and low-income households. Globalisation has thus created pockets of losers, and recent evidence indicates that in addition to financial losses, residents of regions with greater exposure to import competition also suffer from higher crime rates, a deterioration of health outcomes, and a dissolution of traditional family structures. We argue that new import tariffs such as those imposed by the US in 2018 and 2019 are unlikely to help the losers from globalisation. Instead, displaced workers may be better supported by a combination of transfers to avert financial hardship, skills training that facilitate reintegration into the labour market, and place-based policies that stimulate job creation in depressed locations. JEL Classification: E31, F13, F14, F16, F23, I14, I38, J21, J23, J31, J61, J62, R11


Introduction
Many economies in Western Europe have experienced a sizeable increase in income inequality since the 1980s, and inequality has grown even more rapidly in the United States. Whereas educated workers in skilled occupations benefited from rising salaries, wages have stagnated for many less educated workers in unskilled occupations. The rising inequality in advanced economies coincided with a period of globalisation that was characterised by rapid growth in international merchandise trade.
Basic economic models predict that trade could contribute to greater inequality in skill-abundant advanced economies, as globalisation leads such countries to specialise in skill-intensive industrial sectors, which raises labour demand for skilled workers but reduces demand for unskilled ones. Yet despite this theoretical link between trade and inequality, empirical analyses long concluded that increased trade was not a major cause of increasing inequality in advanced economies. However, this perspective on trade and inequality has evolved during the decade of the 2010s, as a growing body of empirical research found sizeable impacts of trade shocks on labour markets and inequality. During the same period, international trade has become a more contentious subject in political debate, and a many-decades-old trend towards greater trade liberalisation has been broken by new tariffs that resulted in a 'trade war' between the US and China.
We begin this chapter by describing the remarkable changes in patterns of international trade that have taken place over the last four decades (Section 2), and the contemporaneous conceptual advances in the economic analysis of trade's impact on the labour market and inequality (Section 3). We next discuss recent evidence on the impact of trade on labour markets (Section 4) and consumer prices (Section 5) in advanced economies. The following sections discuss broader implications of adverse trade shocks, including their social and political repercussions (Section 6), as well as options for public policies that seek to support globalisation's losers (Section 7). We conclude with a discussion of our findings (Section 8).

Changing patterns of international trade
We begin by documenting four salient patterns of change in international goods trade that have taken place since 1980. These patterns help in understanding the context and focus of recent empirical analyses of trade's impact on labour markets and inequality.
Pattern 1: Rising world trade in goods. The last four decades have witnessed a large increase in worldwide merchandise trade, whose value grew from $740 billion in 1980 to $19 trillion in 2018. 2 The annual value of goods traded across national boundaries increased considerably faster than worldwide GDP. Panel a of Figure 1 shows that the ratio of world exports to world GDP grew from 16.3% in 1980 to 22.3% in 2018. However, that increase was unevenly distributed over the last four decades. The export-to-GDP ratio actually declined during both the 1980s and the 2010s, and all of the overall expansion occurred in the two intermittent decades of the 1990s and 2000s. 2 World Bank (2021). Figures  Pattern 2: Rising share of low-income countries in world exports. The expansion in trade's overall volume coincided with remarkable changes in the composition of trade flows. One of these changes was the rising importance of goods exports from low-income countries. Panel b of Figure 1 shows that these countries, which had per-capita annual incomes below $480 in 1987, accounted for a mere 3-5% of world exports during the 1980s. The low-income country contribution to world exports slowly increased during the subsequent decade, from 4. 2% in 1990 to 6.7% in 2000. In the following 15 years, low-income countries rapidly rose to prominence as exporters of goods, and their share of global exports nearly tripled to 18.4% in 2015, before levelling off in the most recent years.
Pattern 3: Rising share of global value chains in world trade. Another important change in the composition of world trade comes from the rising importance of global value chains (GVCs), where different stages of production take place in different countries. One measure of this international segmentation of production is the amount of trade that crosses borders multiple times because a country's exports embody inputs sourced from abroad. Panel c of Figure 1 indicates that the fraction of trade associated with GVCs rose from 41. 4% in 1980 to 48.1% in 2015, the last year covered in the underlying World Bank (2019) data series. As with the global export-to-GDP ratio, this growth was concentrated in the middle two decades. The GVC share of world trade grew from 41.6% in 1990 to a peak of 51.8% in 2008, and has declined slightly since.
Pattern 4: Growing trade imbalances. While overall goods trade grew and changed in composition, some countries increasingly specialised as manufacturing hubs whose exports exceeded their goods imports. These countries therefore experienced a growth in their trade surpluses. Other countries instead expanded their imports by much more than their exports, and thus realised large trade deficits. Panel d of Figure 1 shows the combined value of the goods trade deficits of all countries whose trade balance in goods was negative in a given year. Echoing (Dauth, Findeisen and Suedekum, 2014). 4 At a global level, however, the most important transformation was not the opening of Eastern Europe to international trade, but the dramatic economic expansion of the world's most populous country, China. As Figure 2 shows, Chinese merchandise exports to the US and EU15 countries (including the UK) grew from just $4 billion in 1980 to $777 billion in 2019 -an astonishing 194-fold increase. Chinese worldwide exports reached $2.4 trillion in 2019, making it the world's largest goods exporter.
China's rapid expansion had an important impact on all four of the global trade patterns shown in Figure 1. During the 1990s and2000s, China was the largest contributor to the worldwide growth in exports-to-GDP, as well as the largest contributor to rising trade in global value chains. 5 Appendix Figure A1 shows that China's share in worldwide goods exports grew from a mere 0.9% in 1980 to a high of 13.6% in 2015. Over the same period, other low-income countries' share in world exports rose only modestly from 3. 8% in 1980 to 4.8% in 2015, implying  China's rise followed several decades of poor economic performance during the Maoist era, which kept China well inside its production frontier and generated abundant potential for later catch-up growth. Much of that growth was realised during the two decades of the 1990s and 2000s, as China transitioned from an economy with strong central planning and little foreign trade to a market-oriented economy with a sizeable private sector and vibrant trade. China's market opening allowed for a dramatic rise in productivity, with output per worker in manufacturing rising at an annual rate of 8% from 1998 to 2007 (Brandt, Van Biesebroek and Zhang, 2012). As productivity in China surged, manufacturing industries in Europe, North America and elsewhere were increasingly confronted with a 'China shock' of rapidly rising import competition by Chinese manufacturers (Autor, Dorn and Hanson, 2016). However, the growth of Chinese manufacturing exports slowed considerably towards the end of the 2000s. By this time, China had arguably realised much of the catch-up growth that was enabled by its earlier economic reforms. Moreover, the industrial planning of the Chinese government gave stateowned enterprises an increased prominence again, and China's productivity growth slowed down considerably from 2009 onwards (Brandt et al., 2020).

China's prominent contribution to changing patterns of world trade during the globalisation wave of the 1990s and 2000s explains why many recent empirical analyses of trade's impact on inequality in Europe and the US focus on the 'China shock'. However, the one-time event of the former communist bloc's integration into the worldwide market economy and international trade systems also coincided with important declines in trade and communication costs, which further contributed to the expansion of world trade.
The rapid improvement of computer technology and in particular the spread of the internet facilitated communication even over long distances. Malgouyres, Mayer and Mazet-Sonilhac (2021) show that French firms expanded their international goods imports by 25% once their geographic location got access to broadband internet between 1997 and 2007. Furthermore, the increased use of computer-aided design and manufacturing facilitated offshoring, as it became easier for firms to transmit exact production specifications to suppliers abroad (Fort, 2017). Technological improvements also contributed to a gradual decline in the costs of international air and sea freight shipping during the 1990s and 2000s (Hummels, 2007).
The 1990s and 2000s also witnessed a continued decline in global tariff rates. The worldwide average tariff for goods shipments declined from 13. 6% in 19866% in to 7.5% in 20086% in (Antràs, 2020. This period covers the establishment of the World Trade Organization (WTO) in 1995, and the creation of new regional free-trade zones such as in South America (Mercosur in 1991), Southeast Asia (ASEAN Free Trade Area in 1992) and North America (NAFTA in 1994).
The global decline in tariffs resulted primarily from a reduction of tariff rates applied by developing countries. The average tariff applied by developed countries, which had already fallen to 6.2% by 1990, declined only a little further to 3.8% by 2016 (World Bank, 2019). As a consequence, the rapid growth of imports from China into Europe and the US was arguably not primarily a consequence of trade policy. 6 However, both Western Europe and the US were involved in the expansion or creation of their own regional free-trade zones. The European Union (EU) negotiated association agreements with various Eastern European countries during the 1990s, whose provisions included reductions in tariff rates. These so-called 'Europe Agreements' were followed by the admission of eight Eastern European countries into the EU in 2004. The economic integration between Europe's east and west was characterised by rapidly growing Eastern European exports to the west, and rising Western European investment in the east (Barysch, 2006). In North America, the 1994 NAFTA treaty between Canada, the US and Mexico facilitated trade between the higher-income countries Canada and the US and their substantially poorer southern neighbour Mexico.

Conceptual advances in analysing the labour market impacts of trade
The classical textbook model in economics that links trade to labour market inequality is the Heckscher-Ohlin factor proportions model (Heckscher, 1919;Ohlin, 1933), whose implications for inequality were studied by Stolper and Samuelson (1941). The Stolper-Samuelson theorem posits that a reduction in the relative price of low-skill intensive products will lower the real wage of unskilled workers but raise the real wage of high-skilled workers. A country in Europe or North America with a relatively well-educated workforce would thus experience rising wage inequality following trade integration with developing countries, where a relative abundance of unskilled labour allows for cheaper production of low-skill intensive products. 6 In 2000, the US granted China the status of Permanent Normal Trade Relations, after which the tariff rates applied to Chinese goods were no longer subject to annual re-approval by US Congress. While this policy change did not directly affect tariff rates, it reduced the threat of a later tariff hike (Pierce and Schott, 2016;Handley and Limão, 2017 Harrigan and Barrows, 2009;Khandelwal, Schott and Wei, 2013;Utar, 2014).
While this theoretical link between trade and inequality had long been known, empirical research connecting trade to the labour market gained momentum in the 1990s, when economists debated the causes of the rapidly rising wage differential between college-educated and highschool-educated workers in the US and elsewhere. Empirical studies soon confirmed that international trade did contribute to the rising skill premium in advanced economies, but also found that the magnitude of trade's labour market effect was modest. Krugman (1995)  There was also reason to expect that the impact of low-income country imports on inequality in high-income countries may become even smaller in future years. While Wood (1995)  One intuitive candidate explanation for the small impact of trade on wage inequality in early empirical studies is their reliance on data from the 1980s and early 1990s, a period when both overall trade and exports by low-income countries barely increased (see panels a and b in Figure  1 above). 8 The view that trade had little consequence in the labour market took hold precisely at a time when patterns of trade had started to change rapidly. By the late 2000s, Krugman (2008) warned that 'the consensus that trade has only modest effects on inequality rests on relatively old data'.
However, updating earlier calculations with newer data did not change the conclusions regarding trade's impact on inequality. Irwin (2008) notes that a simple factor proportions model proposed by Krugman (1995), which explained 11% of the rise in the US skill premium from 1979 to 1992, still only explained 11% of the growing skill premium when Bivens (2007) Leamer (1996) Krugman, 1995), or they used trade volumes directly to analyse trade's impact on the labour market by measuring the labour content embodied in traded goods (e.g. Wood, 1994;Borjas et al., 1997). Dauth, Findeisen and Suedekum (2014) similarly argue that the fall of the Iron Curtain in Europe and the transition of 9 Trade shocks also have a larger impact on the average wage differential between skilled and unskilled workers when labour market frictions hinder the reallocation of workers towards sectors with rising labour demand (Burstein and Vogel, 2017).
Eastern European countries to market-based economies during the 1990s created a trade shock for countries in Western Europe such as Germany.
Instead of analysing trade shocks that resulted from a bundle of political and economic reforms in transition economies, other research estimates causal effects of trade shocks based on a single trade policy reform, including reforms originating in the country whose labour market outcomes are the object of interest. Well-studied episodes of trade policy reform include large import tariff reductions in emerging economies such as Brazil (e.g. Kovak, 2013;Dix-Carneiro and Kovak, 2017;Helpman et al., 2017) or India (e.g. Topalova, 2010) during the early 1990s, the reduction of trade policy uncertainty following the US's decision to grant Permanent Normal Trade Relations to China in 2000 (Pierce andSchott, 2016;Handley and Limão, 2017), the phaseout of textile import quotas in the EU and US under the Multifibre Arrangement in the early 2000s, and the US-China 'tariff war' in 2018(Flaaen and Pierce, 2020Autor et al., 2021). A further approach to identify causal impacts of trade relies on large and persistent changes in exchange rates, such as the devaluation of the British pound following the Brexit referendum (e.g. Breinlich et al., 2019;Costa, Dhingra and Machin, 2019) or the appreciation of the Swiss franc following the removal of its peg to the euro in 2015 (Kaufmann and Renkin, 2017;Auer, Burstein and Lein, 2021

Labour market impacts of trade during the 1990-2010 globalisation
We next discuss recent empirical evidence on the impact of trade on labour markets in Europe and the US. 10 As we emphasised in Section 2, international trade expanded rapidly during the 1990-2010 period (panel a in Figure 1), and included a rising fraction of exports from low-income countries (panel b in Figure 1) Pavcnik (2017). 11 Autor, Dorn and Hanson (2016)  apparel (Autor, Dorn and Hanson, 2016). Empirical research has used this cross-industry variation to measure the impact of import competition on manufacturing firms in Europe and the US. Bloom, Draca and Van Reenen (2016) show that in a sample of firms from 12 European countries, those who operated in industries with larger growth of Chinese imports between 1995 and 2017 experienced falling employment and a greater likelihood of firm exit, with more adverse effects in less innovative firms. Several studies from individual European countries confirm that Chinese import competition reduced firms' employment growth and probability of survival (e.g. Mion and Zhu (2013) for Belgium, Branstetter et al. (2019) for Portugal and De Lyon and Pessoa (2020) for the UK). These European results echo early findings from the US, where exposure to low-income country imports was associated with higher rates of plant exit during the 1977-97 period (Bernard, Jensen and Schott, 2006). From 1991 to 2007, Chinese import competition reduced sales, employment, capital stocks, and market valuations of publicly traded US firms (Autor, Dorn, Hanson, Pisano and Shu, 2020 (Bloom, Draca and Van Reenen, 2016), but more recent work shows an inconclusive picture (Campbell and Mau, 2019), while innovation effects in the US appear negative (Autor, Dorn, Hanson, Pisano and Shu, 2020). See Shu and Steinwender (2018) for a literature survey. 13 Over a longer period of 1992 to 2012, which spans the Great Recession, firms in import-competing sectors experienced a greater decline in value added than in payroll, and the Chinese import shock thus increased the labour income share.
that there appear to be substantial industry mobility frictions which prevent workers from costlessly or rapidly moving across sectors following trade shocks. As a consequence, the workers of declining import-competing industries will obtain lower earnings than their peers in other sectors. 14 The structural analyses of sectorial moving costs are complemented by reduced-form studies that trace the differential labour market trajectories of workers in response to the import competition shocks faced by the industries that initially employ them. Studying longitudinal worker data from the UK, De Lyon and Pessoa (2020) find that manufacturing workers were more likely to move across firms and industries if their initial industry was exposed to greater growth in Chinese import competition from 2000 to 2007. However, despite this increased mobility, workers from more import-exposed industries accumulated lower earnings. Comparing a manufacturing worker at the 75 th percentile of Chinese import exposure with one at the 25 th percentile, the former accumulated 3.1% lower annual earnings during the years 2000-07, conditional on pre-period earnings and other worker and industry characteristics. These UK estimates are remarkably close to US results by Autor et al. (2014), who find that a manufacturing worker at the 75 th percentile of exposure to Chinese import competition accumulated 2.9% lower earnings per year between 1992 and 2007 compared with a worker at the 25 th percentile of exposure. Other studies also find adverse impacts of Chinese import competition on the earnings trajectories of workers in Finland (Nilsson Hakkala and Huttunen, 2016), and of imports from Eastern Europe and China on the earnings of workers in Germany (Dauth, Findeisen and Suedekum, 2021). 15 These results indicate that trade shocks can create earnings differentials between workers who have similar skill levels but are employed in industries with different trade exposure. However, the reduced-form estimates additionally show that the earnings effects of import competition are considerably more adverse for workers from lower income strata. Both De Lyon and Pessoa (2020) and Autor et al. (2014) classify workers into three income terciles based on their position in the earnings distribution of workers with the same birth year that was observed prior to the outcome period. 16 In both the UK and the US, the proportional loss in annual earnings as a result of a unit import competition shock is more than 60% larger for workers in the bottom earnings tercile than for those in the middle tercile. The workers who initially belong to the highest-paid tercile of their birth cohort suffer the smallest earnings loss. 17 These results are also consistent with findings by Utar (2018), who studied the 2002-10 adjustment of workers from the Danish textile sector to the phase-out of import quotas under the Multifibre Arrangement. She finds sharp declines both in cumulative earnings and in cumulative work hours for employees who possess at most a high-school degree, but no significant changes for those with college education. Workers whose initial firm lost protection through import quotas were also more likely to experience unemployment spells. That unemployment effect was largest for individuals 14 Artuç and McLaren (2015) estimate that in addition to switching costs across sectors, US workers also face similar costs for moving across occupations, while Traiberman (2020) finds that frictions for occupational switching dominate in Denmark. 15 Studying an earlier episode of trade liberalisation, Devlin, Kovak and Morrow (2020) find that the 1988 Canada-US Free Trade Agreement had little impact on the earnings trajectories of workers in Canada. 16 The classification of workers based on within-cohort earnings rank can be interpreted as a proxy for categorising workers by education. The underlying UK data from the Annual Survey of Hours and Earnings and US data from Social Security Administration records do not provide direct information on education. 17 In an analysis for Germany, Dauth, Findeisen and Suedekum (2021) show that exposure to import competition has the most adverse impact on earnings for workers who have a low worker fixed effect in a panel wage regression that controls for both worker and firm fixed effects. A low worker fixed effect can result from a low education or skill level based on which a worker obtains relatively low earnings with any employer.
with manufacturing-specific vocational degrees, who presumably struggle most to find alternative jobs that fit their skills.
A remarkable finding of the aforementioned analyses for the UK and the US is that greater exposure of a worker's initial industry to Chinese import competition causes only greater subsequent worker mobility across firms and industries, but not greater worker mobility across local labour markets. Both De Lyon and Pessoa (2020) and Autor et al. (2014) find that workers' exposure to import shocks has a small and insignificant negative effect on the number of employment years that a worker spends outside their initial region during the outcome period. 18 These limited spatial mobility responses suggest that an important part of labour market adjustment to import competition shocks takes place within geographically defined local labour markets, which we discuss next. Local labour markets are differentially exposed to import competition based on the extent to which they are specialised in import-competing industries. Analyses of labour market outcomes across regions with differential trade exposure do not simply provide a geographic aggregate of the impact of industry-level trade shocks on the employees of exposed industries. Instead, such analyses also capture all spillover effects of trade shocks that take place at a local level. Such spillovers can result from local demand multipliers, from the propagation of industry-level shocks along local supply chains, or from worker reallocation across industries within a region. Consistent with such spillover effects, both the UK analysis by De Lyon and Pessoa (2020) and the US analysis by Autor et al. (2014) show that individual workers acquire lower earnings not only when their own industry of employment is exposed to greater import competition from China, but also when they live in a local labour market where all other workers on average face a large import exposure shock. Even workers who are not themselves employed in import-competing firms may be indirectly affected when the contraction of such firms creates negative local demand spillovers or reduces job opportunities in their local labour market.
Reduced-form analyses of geographic regions have been used to study the labour market impacts of Chinese import competition in many developed economies, including the UK (Foliano and Riley, 2017), Germany (Dauth, Findeisen and Suedekum, 2014), France (Malgouyres, 2017a), Italy (Citino and Linarello, 2019), Spain (Donoso, Martin and Minondo, 2015), Norway (Balsvik, Jensen and Salvanes, 2015) and the US (Autor, Dorn and Hanson, 2013a). These studies produce three common findings. First, regions with greater exposure to Chinese import competition experienced a statistically significant differential decline in manufacturing employment in all 18 In the case of Germany, Dauth, Findeisen and Suedekum (2014) find some evidence for small positive migration responses of workers whose industries were exposed to rising import competition from Eastern Europe and China. 19 In Third, the subset of studies that also analyse changes in local wage rates all find a negative but often small local wage effect of import competition. However, the overall effect of import competition on households' wage and salary incomes operates not only through the wages of those who have a job, but also through job loss. Autor, Dorn and Hanson (2013a) calculate that two-thirds of the decline in average household earnings in import-competing US local labour markets is explained by lower employment, while only one-third of the income decline is due to lower wage rates for those who stay employed.
A key insight from these analyses is that Chinese import competition shocks had sizeable effects on inequality across local labour markets, as the adverse consequences of these shocks were concentrated in the regions where import-exposed industries had clustered. While reduced-form analyses establish that import-exposed local labour markets experienced lower income growth than less exposed locations, they are not immediately indicative about changes in regions' absolute level of economic well-being. In particular, aggregate gains from trade in the form of lower consumer prices (which we study in Section 5) cannot be identified from a comparison of local labour markets when price changes are uniform across space. Several recent papers thus use reduced-form empirical results to calibrate quantitative trade models that estimate the welfare effects of Chinese imports across local labour markets in the US (Caliendo, Dvorkin and Parro, 2019;Rodríguez-Clare, Ulate and Vásquez, 2019;Adao, Arkolakis and Esposito, 2020;Galle, Rodríguez-Clare and Yi, 2020). Using different theoretical assumptions on the nature and extent of spatial mobility frictions, most of these papers estimate that the average welfare gain of US residents due to Chinese imports was of the order of 0.2-0.3 percentage points. These small average gains combine with a sizeable variation of gains across local labour markets, such that some localities experienced an overall decline in welfare due to the trade shock. For instance, Galle, Rodríguez-Clare and Yi (2020) estimate that the welfare change was as high as plus 1.64 percentage points in the US local labour market that gained most from trade with China, but as low as minus 1.42 percentage points in the location that was most negatively affected. Reducedform analyses indicate that trade's impact on the income dispersion across local labour markets is even substantially larger than suggested by quantitative trade models, which implies that a substantial minority of the US population live in regions where the absolute impact of China's trade on welfare has been negative (Autor, Dorn and Hanson, 2021).

Analyses of average outcomes in local labour markets of course mask the fact that not every resident of a city is equally affected by a trade shock. Indeed, local labour market analyses also
show that Chinese import competition increased inequality within local labour markets, as lesseducated, lower-wage workers were more adversely affected by import shocks than their bettereducated and higher-paid peers. For instance, while import competition led to similar declines in manufacturing employment among college-and non-college-educated workers in Norway and the US, in both countries the increase in unemployment was much larger among the non-college-educated group (Balsvik, Jensen and Salvanes, 2015;Autor, Dorn and Hanson, 2013a). 20 Moreover, the China shock increased wage inequality within US commuting zones. Chetverikov, Larsen and Palmer (2016) show that while the trade shock reduced wages at all percentiles of the local labour market wage distribution, the declines were larger at lower percentiles, with the 20 th wage percentile declining twice as much as the 80 th wage percentile.
While most of the recent literature on the local labour market impacts of trade in Europe and the US focuses on trade with China, there is also a growing body of evidence on the regional impacts of trade integration within continents. Dauth, Findeisen and Suedekum (2014) show that rising imports from Eastern Europe were associated with declining manufacturing employment in exposed German regions, and that German employment reacted more strongly to growing trade with Eastern Europe than to trade with China. Hakobyan and McLaren (2016) study effects of the 1993 North American Free Trade Agreement (NAFTA) on US regions. Their estimates indicate that greater exposure to import competition from Mexico reduced wages in more exposed US localities. As in the analyses for the China import shock, the most adverse outcomes were found for workers with lower education levels. Choi et al. (2021) additionally show that the US locations that were most exposed to rising Mexican import competition experienced more sluggish employment growth following NAFTA.
The impact of trade shocks on local labour markets can be remarkably persistent. Autor, Dorn and Hanson (2021) analyse outcomes in US regions nearly one decade after the China import shock plateaued around 2010. In the year 2018, local labour markets that had been exposed to greater import competition still suffered from reduced employment rates and income levels, and there is little evidence for a recovery from the past trade shock. While these localities eventually experienced some differential net outmigration, such migration occurred only with a substantial time lag (Greenland, Lopresti and McHenry, 2019) and was limited to younger and foreign-born workers (Autor, Dorn and Hanson, 2021). 21 Offshoring and propagation of import shocks along the supply chain International trade increasingly occurs within global value chains, as shown in panel c of Figure 1 above.  Eriksson et al. (2019) additionally show that US local labour markets with less-educated workforces, as well as those with higher wages and manufacturing industries towards the end of a product cycle, experienced the largest employment declines as a result of rising Chinese import competition. 21 A study from Brazil indicates that following a one-time trade liberalisation, regional earnings even continued to differentially decline in localities that experienced a larger surge in import competition (Dix-Carneiro and Kovak, 2017). 22 The term 'offshoring' is used variably in the literature. We adopt the definition proposed in a literature review by Hummels, Munch and Xiang (2018), who characterise offshoring as an international import of intermediate goods by firms that themselves could plausibly have produced these goods.
input, it remains difficult to determine whether its import constitutes offshoring, where a firm relocates actual or potential in-house production to a foreign supplier or subsidiary, or whether the firm obtains inputs from abroad instead of buying them from a domestic supplier. Early work on offshoring by Feenstra and Hanson (1999) argues that firms could more likely have produced goods in-house if those goods are closer to the firms' own outputs in the product space. They propose the term 'narrow offshoring' for imports of goods from the same or similar industries and the term 'broad offshoring' for imports from substantially different industries.
It is instructive to consider how studies of import competition, such as those discussed in the previous two subsections, take imports of intermediate goods and offshoring into account. These analyses construct measures of import exposure that assign all goods to the industries that produce them. Car imports thus provide import competition for car manufacturers, while imports of car parts generate competition for domestic parts producers. At the same time, inputoutput tables indicate that many industries obtain substantial amounts of intermediate inputs from other firms of the same narrow industry. A firm may import car parts and, after a few production steps, sell them as car parts again. Therefore, industry-level studies of import competition actually estimate the combined effects of import competition and narrow offshoring within industries. Moreover, studies of import competition at the level of local labour markets capture not only trade's effect on the directly import-competing industries, but also the spillover effects along industry supply chains that operate within local markets. Biscourp and Kramarz's (2006) analysis of goods imports by French manufacturing firms during the period 1986-92. The authors observe that greater imports were correlated with lower firm-level employment growth. This negative association was particularly strong for narrow offshoring, i.e. for goods that a firm could plausibly have produced itself but whose production it instead chose to offshore. The relative employment losses related to offshoring were largest for less-skilled production workers. Mion and Zhu (2013) similarly note that Belgian manufacturing firms' imports of Chinese goods between 1996 and 2013 were associated with weaker overall employment growth and a falling employment share of production workers in these firms. Hummels et al.'s (2014) 23 Over the two years following the referendum, workers in industries that were more exposed to higher input prices experienced lower wage growth. Consistent with the view that offshored inputs complement high-skilled workers, wage impacts were more negative for graduates than for non-graduates.

By contrast, analyses of manufacturing firms' imports of goods allow the effects of narrow offshoring to be isolated, and often combined with the impacts of broader offshoring, i.e. firms' imports of goods from different industries. An early application of this approach is
Analyses of offshoring are interested in imports that occur upstream from a domestic industry within a supply chain. While narrow offshoring can involve the import of intermediate goods from other firms of the same industry or from foreign affiliates of MNEs, a more broadly defined offshoring will involve imports from different supplier industries. A domestic producer of tyres may, for instance, be affected by the availability of cheaper imports of synthetic rubber and steel fibre, which allows the firm to offshore its own production of these inputs or to switch from domestic suppliers to foreign suppliers of such goods. However, import shocks can propagate along supply chains not only downstream from supplier industries to their customers, but also upstream from customers to suppliers. An increased import of foreign cars may, for instance, reduce the ability of domestic tyre producers to sell their products to domestic car manufacturers.

Acemoglu et al. (2016) investigate both the downstream and upstream propagation of Chinese import shocks across US industries based on national input-output linkages between industries. For each industry, they measure a direct exposure to Chinese import competition based on imports of the industry's own products, a downstream-propagating exposure based on imports at the level of an industry's supplier industries, and an upstream-propagating exposure based on imports by an industry's customer industries.
Their results indicate that direct exposure to Chinese import competition caused significant job losses in US manufacturing industries over the 23 There was no significant impact of the pound's devaluation on the prices of UK exports.

1991-2011 outcome period. Greater import exposure at the level of supplier industries, which includes broadly defined offshoring, created a positive but statistically insignificant employment effect in manufacturing industries and an insignificant negative effect in non-manufacturing industries. Studying data from France over the period 2000-07, Aghion et al. (2021) similarly observe a positive but insignificant effect of imported inputs from China on employment in
manufacturing firms. While some firms will benefit from access to cheaper foreign inputs, others may face disruptions when their domestic suppliers succumb to import competition. Drawing on coarser data from the World Input-Output Database, Kiyota, Maruyama and Taniguchi (2021) show that supplier exposure to Chinese import competition from 2000 to 2014 caused only modest employment changes not only in US industries, but also in their counterparts in the UK and France. Instead, job gains in Germany were significant. 24 Adao, Arkolakis and Esposito (2020) integrate input-output linkages into a local labour market analysis. They find that US regions whose industries buy more inputs from supplier industries that are exposed to Chinese import competition did not experience differential changes in employment or wages due to this indirect trade exposure.  Autor, Dorn and Hanson (2013a) to identify the exposure of German local labour markets from 1988 to 2008 not only to import competition, but also to export demand shocks in the industries that are present in a region. They find that growing exports to Eastern Europe and China are associated with both rising employment and rising median wages in local labour markets.
The notion that rising exports are associated with gains for workers is also supported by firmlevel evidence. Biscourp and Kramarz (2006) observe that growing exports by French firms to the rest of the world during the late 1980s and early 1990s were correlated with higher employment growth in these firms, especially for skilled production workers. Moreover, greater demand for the exports of Danish firms during the 1990s and 2000s had positive wage effects for the employees of these firms (Hummels et al., 2014). Different from offshoring, which raised skill premiums within firms, exports led to a roughly similar proportional increase in wages for workers with and without tertiary education.
The local labour market analysis by Dauth, Findeisen and Suedekum (2014)  Why were patterns of aggregate import growth more similar across countries than patterns of export growth? One potential explanation is that many imports from China are consumer goods such as apparel, shoes, toys, home electronics or furniture that are demanded to a similar extent by consumers in the US and in European countries. Autor, Dorn and Hanson (2016) (Dadush, Domínguez-Jiménez and Bruegel, 2020 We find evidence consistent with that hypothesis in Figure 4, which shows a simple correlation between OECD member countries' growth in net goods imports from China per manufacturing worker between 1999 and 2007, and the percentage change in their country's manufacturing jobs during the same period. 29 The figure shows a statistically significant negative relationship between the growth of a country's net imports from China and the change in its manufacturing employment. 30 Indeed, three of the five countries with the largest increase in net Chinese imports per manufacturing worker are also among the five countries with the largest contraction in domestic manufacturing employment: the UK, the US and Iceland. Conversely, the only two European countries whose goods exports to China increased by more than their imports either experienced a modest growth in manufacturing jobs (Switzerland) or saw no change (Luxembourg).
While Figure 4 shows a significant negative correlation between net import growth from China and growth of domestic manufacturing employment in developed countries, it is noteworthy that employment changes can differ substantially across countries with similar China exposure. For 29 Following Autor, Dorn and Hanson (2013a), we scale the dollar change in a country's net imports from China by the country's manufacturing employment at the start of the period. By focusing on net Chinese import exposure (i.e. imports minus exports), one implicitly assumes that the association between export growth and changing manufacturing employment has the opposite sign but otherwise the same magnitude as the association between imports and manufacturing employment. Dauth, Findeisen and Suedekum (2014) show that imports from and exports to Eastern Europe indeed had such symmetric employment effects in German local labour markets. 30 We exclude the Netherlands and Belgium whose trade statistics are sometimes considered problematic. To investigate the reaction of a country's labour market to a trade shock of a given magnitude more systematically, we next study results from comparable local labour market analyses from the UK (Foliano and Riley, 2017), Germany (Dauth, Findeisen and Suedekum, 2014), Spain (Donoso, Martin and Minondo, 2015), Norway (Balsvik, Jensen and Salvanes, 2015) and the US (Autor, Dorn and Hanson, 2013a). A common feature of these studies is that they estimate the impact of Chinese import competition (measured in thousand dollars of imports per worker) on the share of a country's working-age population that is employed in manufacturing. While all of these studies use an empirical design similar to that of Autor, Dorn and Hanson (2013a), the countryspecific regression analyses differ in terms of the periods they cover, the geographic concepts of local labour markets they use and the control variables they incorporate. All of these features can potentially influence estimated effect sizes. 31

Figure 5. Effect of a unit increase in Chinese import competition on local manufacturing employment across countries
Note: Based on authors' calculations from studies of Chinese import competition from the US (Autor, Dorn and Hanson, 2013a), Germany (Dauth, Findeisen and Suedekum, 2014), Norway (Balsvik, Jensen and Salvanes, 2015), Spain (Donoso, Martin and Minondo, 2015) and the UK (Foliano and Riley, 2017 Having this caveat in mind, we show in Figure 5 the decline in the percentage of the working-age population employed in manufacturing that is associated with a $1,000 per worker increase in annual imports from China. The bars in the figure indicate regression coefficient estimates, while whiskers provide 95% confidence intervals for the regression estimates. The figure shows that all studies consistently find a statistically significant negative impact of Chinese import competition on the fraction of the local working-age population that is employed in manufacturing. 32 However, 31 Appendix Section B details the differences across local labour market studies from different countries. 32 The figure omits studies by Malgouyres (2017a) for France and Citino and Linarello (2019)  What could explain the sizeable international differences in the elasticity of manufacturing employment to import competition shocks shown in Figure 5? We first note that the particularly adverse employment effect of import competition in Spain appears consistent with the observation that the Spanish labour market also tends to adjust to adverse cyclical shocks with job losses that are much larger than in other European countries (Bentolila, Cahuc, Dolado and Le Barbanchon, 2012). Dauth, Findeisen and Suedekum (2014) further argue that employment effects in Germany may have been smaller than in the US because Germany invests more in active labour market policies that help displaced workers to transition to other jobs. Appendix Table A1 provides OECD measures of spending on active labour market policies (including benefit administration, employment incentives, training and direct job creation). Of the countries considered in our Figure 5 comparison, Norway spent the most on such policies relative to its unemployment stock in 2007, followed by Germany. The UK and especially the US spent the least on such policies. While it is difficult to draw firm conclusions from a comparison of a small number of countries, the Table A1 data provide some support that countries with greater emphasis on active labour market policies were more successful at containing the job loss due to rising import competition. 35 in France than in the US, while import-exposed local labour markets in France additionally experienced a large decline in non-manufacturing jobs (Appendix Table B4). In Italy, the trade-induced decline in the manufacturing-to-population ratio as measured by the product of the average growth of import exposure and the estimated employment decline per unit of exposure is twice as large as in Norway and slightly smaller than in Germany (Appendix Table B3). 33 We convert results to be expressed in thousands of 2007 US$, except for Spain where results are expressed in terms of current dollars (from 1999 to 2007) as we could not convert them using the figures reported in Donoso, Martin and Minondo (2015). 34 See Appendix Table B3 for details of these calculations. 35 Worker-level evidence from the UK (De Lyon and Pessoa, 2020) and the US (Autor et al., 2014) indicates that workers who find a job outside their initial import-exposed industry are more likely to move to another manufacturing industry rather than the non-traded sector. To the extent that active labour market policies facilitate a reallocation of workers within the manufacturing sector of a local labour market, one would expect to see smaller coefficient estimates in Figure 5 for those countries that invest more in such policies.

Import competition and consumer prices
Research on trade and inequality often emphasises the differential impacts of trade shocks on the labour market outcomes of different groups of workers. A less analysed channel through which trade can impact inequality is consumer prices. Basic models of international trade posit that trade is welfare-enhancing because consumers gain access to lower-priced imported goods. Moreover, consumers may benefit from access to a broader variety of goods (e.g. Broda and Weinstein, 2006), and import competition can induce domestic producers to reduce their prices (e.g. Feenstra and Weinstein, 2017). While the impact of trade on average consumer prices in a country is interesting in its own right, a price shift will generate inequality in society only when different population groups experience differential price changes for their consumption bundles.  Import penetration grew most for textile and apparel products, for furniture, and for consumer 36 Our analysis excludes CPI categories for services, which are not directly exposed to goods trade. 37 We map import changes and employment changes to consumer product codes using a conversion based on the shares of consumption spending devoted to different industries' output of different products in 2010. We use the same mapping to estimate the number of workers employed in the production of different consumer products (after an adjustment for spending on imported goods). See Appendix Section C for more details.
electronics products and appliances. The product categories with greater import growth generally experienced both larger price and employment declines in the UK.
The price data include two sets of goods that are notable outliers. First, the prices of goods related to fuel and heating increased much faster than the prices of all other goods from 1999 to 2007, with increases ranging from 34% for vehicle fuel to 89% for gas and 161% for liquid fuel (these goods are omitted from Figure 6 to improve visibility). Other than energy products, the good with the largest price increase was tobacco, which shares with fuel products a high level of product taxes. Since fuel and tobacco prices often exhibit large changes that are unrelated to international trade, our subsequent analysis probes the robustness of its results to the exclusion of these goods. 38 Second, the largest price declines by far were observed for technology products such as IT equipment (-82%), photography equipment (-65%) and audio-visual equipment (-61%). While these goods benefit from low-cost assembly in China, their sharp price decline may primarily be a consequence of rapid technological improvement and the associated introduction of better product varieties (Houseman, Bartik and Sturgeon, 2014), rather than changing patterns of trade. We thus also analyse the robustness of our results to excluding these technology goods.  Fajgelbaum and Khandelwal (2016) show that spending shares on food and manufactured goods are smaller in higher-income countries. To the degree that this pattern also holds within countries, one may expect that low-income households benefit more from declining goods prices than more affluent households.

We use data from the 2001 Living Costs and Food Survey to investigate spending patterns across income strata in the UK. The relationship between income and spending shares on goods is weakly negative. Households in the lowest income quintile spent 57% of their expenditure on goods, while those in the highest income quintile spent 52% on goods. However, not all goods were disproportionately consumed by lower-income households. Figure 7 plots, for each good in the UK CPI, the share of this good in total spending by bottom quintile households, minus the corresponding spending share for top income quintile households in 2001, and shows this against the growth in Chinese import penetration from 1999 to 2007. Most of the goods on which the low-income households spend a larger fraction of their expenditure compared with high-income
households are food products such as meat, bread and dairy, which face no meaningful import competition from China. The high-income households instead outspend the poor on cars and motor fuels, which again have little Chinese trade exposure. Among the goods that face a significant growth in import competition, the poor spend relatively more on shoes and appliances, but the rich spend relatively more on a sizeable number of import-exposed products such as computers, furniture and garments. As a consequence, the regression slope in Figure 7 is weakly negative, and further calculation reveals that the average import exposure of the top income quintile's goods basket is about 16% larger than the exposure of the bottom quintile's 42 We calculate ! !"#$% × ∑ (Δ&ℎ()*+, !   Table 1. This follows a similar calculation by Jaravel and Sager (2019), who obtain a much larger benefit for consumers of $288,000 per displaced worker in the US when they use a comparable specification and instruments. In their US estimates, the effect of the import shock is larger for prices and smaller for employment than in the UK, such that the ratio & !"#$% &%&! is two to four times larger in the US. Moreover, Jaravel and Sager scale the percentage change in consumer prices not with household consumption, but with domestic absorption which also includes the purchases by firms and government of goods whose prices may not be accurately captured in the CPI. basket. While low-income households thus spend a larger fraction of their overall expenditure on goods, the high-income households spend relatively more on the import-competing goods that stand to benefit most from price declines. These two offsetting effects are nearly equally large in magnitude, such that the import exposure of the entire expenditure basket of goods and services is almost equivalent for both groups. Therefore, our analysis suggests that trade-induced price declines equally benefit low-income and high-income households, and thus have little impact on inequality between these groups.  Figure 7 captures the rich and the poor's expenditure shares on goods regardless of whether these goods have been domestically produced or imported. If the low-income households were more likely to consume cheap imported products while affluent households more often buy domestic products whose prices change little, then our analysis may underestimate the relative benefit to the poor in terms of lower prices. 45 However, Levell, O'Connell and Smith (2017) show for the case of food products that UK households of different income levels do not systematically differ in their relative purchases of domestic versus imported goods. Borusyak and Jaravel (2018) Borusyak and Jaravel (2018), Hottman and Monarch (2018) and Bai and 45 The distinction between spending on domestically produced goods and spending on imported goods will be less relevant if import competition induces domestic producers to substantially reduce their prices, as suggested by Jaravel and Sager (2019).

Stumpner (2019) all find that rising Chinese imports to the US reduced consumer prices roughly evenly for households of different education groups or income strata. 46
In addition to the literature on the price effects of Chinese import competition, recent empirical research has also analysed the response of domestic prices in Europe and the US to sudden large changes in exchange rates or changes in tariffs. 47 Given the above observation that the expenditure shares on goods or imported goods do not vary greatly across income groups within the UK or the US, one might expect that changes in goods prices do not create substantial inflation inequality, unless a price change -for instance due to tariffs -differentially affects goods whose expenditure shares vary strongly across income groups. Breinlich et al. (2019) indeed estimate that the sudden depreciation of the British pound following the 2016 Brexit vote increased prices for consumers in the UK by an average of 2.9%, but had no differential impact on households of different income levels.
While Section 4 above emphasised the differential impact of trade shocks on labour market outcomes across regions, there is much less evidence on their regional price effects. It is generally assumed that the prices of traded goods do not vary greatly across space within a country. However, the price effects of an import shock could still vary across regions if there is variation in regional consumption bundles. Breinlich et al. (2019) provide some evidence for such spatial effects in their analysis of changes in UK consumer prices following the sterling depreciation of 2016. They estimate that the resulting increase in inflation was 0.7 percentage points lower in the London area, where housing costs account for a relatively large share of overall expenditure, than in Northern Ireland, where consumers spend a relatively high proportion of their income on such goods as food, clothing and fuel. In the case of the Chinese import competition shock in the US, Bai and Stumpner (2019) instead find no evidence for differential price effects across major regions of the country.
In sum, we conclude that the impact of trade shocks on prices appears to be quite uniform across income groups and geographic regions. While shocks such as rising import competition contributed to greater inequality in labour market outcomes, such inequality does not appear to be substantially offset or exacerbated by differential changes in prices.

Societal and political repercussions
Fissures in the fabric of society Economic analysis usually evaluates the welfare effects of trade based on its impact on people's consumption possibilities. These are determined both by their labour earnings and by the price and variety of the goods that they can purchase. From this perspective, a loss of well-paid 46 While the rich and poor have roughly equal spending shares for import-exposed products, it could be that for some reason, the goods that are more often purchased by one group may have prices that react more elastically to an import shock of a given magnitude. Jaravel and Sager (2019) Vance's (2016) best-selling memoir that recounts the rise and decline of a manufacturing town in the US Midwest. When the local steel plant downsized during the 1980s, many workers lost not only their jobs but also some of their pride and purpose. Public facilities in the city deteriorated, substance abuse and crime increased, and families broke apart. Vance cites the work of sociologist William Julius Wilson (1996), who observed that 'A neighborhood in which people are poor but employed is different from a neighborhood in which people are poor and jobless. Many of today's problems in the inner-city ghetto neighborhoods-crime, family dissolution, welfare, low levels of social organization, and so on-are fundamentally a consequence of the disappearance of work'. Vance argues that these observations, which originally referred to inner-city ghettos, apply equally to decaying manufacturing towns. (2018) both report increased crime rates in these localities, especially regarding property crime.

Recent research brings more systematic support to the notion that localised impacts of import competition shocks had broader social repercussions beyond changes in individual earnings. Feler and Senses (2017) show that US local labour markets with greater exposure to Chinese competition experienced declines in local tax revenues which in turn led to declines in government expenditures for such purposes as education, parks and natural resources, and transportation. Moreover, they and Che, Xu and Zhang
Import competition shocks have also been linked to a range of adverse health behaviours and outcomes. Drawing on data from the British Household Panel Survey, Colantone, Crinò and Ogliari (2019) show that workers in the UK whose industries become exposed to greater Chinese import competition experience higher rates of anxiety and depression, social dysfunction and loss of confidence. These individuals also become more likely to report alcohol and drug consumption and heavy smoking.  Lang, McManus and Schaur (2018) similarly detect that greater exposure to Chinese import competition increased mental health problems in the local population. Moreover, a greater fraction of individuals in these localities report an inability to afford a doctor visit. In regions that simultaneously faced a large exposure to automation, the import shocks also contributed to a broad range of additional adverse health outcomes, such as diabetes, obesity and pain (Adda and Fawaz, 2020). 50 Perhaps most worrisome, local import exposure induced higher mortality rates, 48 In theoretical models that include a basic labour-leisure trade-off, job loss and unemployment would have a negative effect on workers' utility due to the reduction of income, but greater leisure time of workers who are out of jobs would contribute positively to utility. 49 Colantone, Crinò and Ogliari (2019) Adda and Fawaz (2020) follow Autor and Dorn (2013) in measuring local labour markets' exposure to automation based on their specialisation in routine-task-intensive occupations, which can potentially be substituted by computers and robots. Autor, Dorn and Hanson (2013b) show that the routine-intensity of US commuting zones is not correlated with notably due to drug and alcohol poisoning (Autor, Dorn and Hanson, 2019;Pierce and Schott, 2020).
Trade shocks can also affect family and household structures, with shocks to female and male workers generating substantially different results. Keller and Utar (2018) show that Danish workers whose initial employers faced rising import competition from China during the period 2000-09 spent less time as employees at the initial firm than otherwise-comparable workers whose industries were not exposed to import competition. While the trade-induced decline of employment at the initial firm had a similar magnitude for both women and men, the subsequent career trajectories differed substantially across genders. Men's reduced employment at the initial firm was compensated by correspondingly longer employment spells at other firms, but women's reduced employment at the initial firm was associated with longer periods of nonemployment. Women instead became more likely to marry and to have children, and the import shock thus raised fertility and marriage rates overall. These Danish results partly echo findings for the US. Autor, Dorn and Hanson (2019) split the aggregate import exposure of local labour markets into gender-specific shocks based on the initial gender composition of trade-exposed local industries, and find that an adverse shock to females' labour market opportunities increases marriage rates and fertility. However, shocks that reduce employment and earnings primarily among men have the opposite effects of reducing marriage and childbirth. Since the US manufacturing sector employs substantially more men than women, the adverse effects for men dominate in the aggregate, which means that local labour markets with greater overall import exposure experience larger employment and earnings declines among men than among women, and reductions in fertility, marriage and cohabitation. Moreover, children in these localities become more likely to grow up in poor and single-parent-headed households. 51 Complementing this evidence, the UK study by Colantone, Crinò and Ogliari (2019) finds evidence that children experience adverse outcomes as a consequence of import shocks in their father's industry, while there are no effects of shocks to their mother's industry. In particular, children aged 11-15 whose father's industry faced greater trade shocks were less likely to speak to parents about issues the youngsters cared about, spent longer hours watching TV, had lower self-esteem, and felt greater unhappiness with life.
It is noteworthy that the adverse impacts of trade shocks on public goods provision, crime, health and family structures affect the residents of the same locations, and probably often the same individuals, who also suffer the most negative economic consequences from these shocks.

Public attitudes towards trade and electoral outcomes One might expect the negative consequences of increased import competition for certain groups and regions to be reflected in public attitudes towards trade, electoral outcomes and ultimately countries' willingness to pursue open trade policies.
their exposure to Chinese import competition, which allows separate identification of the impacts of automation and trade on local labour markets (Autor, Dorn and Hanson, 2015). 51 The notion that an adverse local labour market shock may lead not only to economic but also to social decline does not apply to recent trade shocks only. For instance, Black, McKinnish and Sanders (2003) document that the rapid decline of the US coal and steel industries during the 1980s caused an increase in single-parent households, as employment opportunities for men deteriorated in heavily affected areas.

Figure 8. Change in positive attitudes towards trade
Note: Calculations by Davenport, Dorn and Levell (2020) 2002and Spring 2007, 2010, 2011 Davenport, Dorn and Levell (2020) analyse recent international data from the PEW Global Attitudes Survey, which asked respondents about their general attitudes towards trade. Figure 8 shows  Davenport, Dorn and Levell (2020) Davenport, Dorn and Levell (2020)  Research in both economics and political science has explored whether import competition shocks in the 2000s also affected electoral outcomes specifically in those local labour markets that were most exposed to trade. 54 Colantone and Stanig (2018a) relate the Chinese import competition faced by large geographic regions in 15 Western European countries to outcomes in parliamentary elections. They find that rising import competition was associated with electoral gains for far-right parties between 1988, while Milner (2021 shows that these effects persist through 2018. This aggregate result receives support from more detailed local labour market studies in individual countries, which showed an impact of import competition on growing electoral success of far-right fringe parties in Germany parliamentary elections between 1987 and 2009 (Dippel et al., 2021) and of France's National Front party in presidential elections from 1995 to 2012 (Malgouyres, 2017b). Moreover, residents of UK regions that faced a larger growth of import competition during the 1990s and 2000s became more likely to voice nationalist sentiments and lower support for EU membership in the British Household Panel Survey (Harms and Steiner, 2018), and these regions were more likely to vote for Brexit in 2016 (Colantone and Stanig, 2018b).

based on Pew Global Attitudes Surveys conducted in Summer
For the US, Autor, Dorn, Hanson and Majlesi (2020) observe various outcomes that suggest an ideological and electoral shift to the right in local labour markets with greater exposure to Chinese import competition over the period 2000-16. These outcomes include a rising market share of the pro-Republican TV channel Fox News, a greater growth of campaign contributions by conservative donors, a larger probability of electing politicians from the right wing of the Republican party in congressional elections, and greater vote shares for the candidates of the Republican party in presidential elections. These outcomes, however, do not reflect a monotone shift towards more conservativism. Import-exposed locations also saw increased campaign It is not immediately obvious why import competition would lead to electoral gains for populist right-wing parties and politicians. 55 Using a mediation analysis, Dippel et al. (2021) show that the effect of the import shock on vote shares for far-right parties in Germany can be fully explained by the shock's adverse impact on local labour market outcomes. However, this leaves open the question of why a deterioration in local labour market conditions due to trade would favour primarily nationalist right-wing parties, and not left-wing parties that promote greater financial redistribution.
One interpretation is that nationalist right-wing politicians promote protectionist trade policies that shield domestic workers from the effects of import competition. The 2016 election of Donald Trump as president of the US stands out because he spoke far more about trade during campaign events and took more protectionist stances than any other major-party candidate for the US presidency from 2008 to 2016 (Cerrato, Ferrara and Ruggieri, 2018 Another interpretation is that the lasting economic decline of local labour markets due to import competition contributed to a cultural backlash of rising nationalism and support for right-wing parties' anti-immigration stances. This more general backlash may have then persisted even as general attitudes towards trade recovered. Colantone and Stanig (2018b) provide evidence for a connection between trade shocks and anti-immigrant sentiment based on survey data from the 2016 British Election Study. They show that the residents of import-competing regions of the UK were less likely to perceive immigration as good for Britain's economy or cultural life, and were less likely to advocate greater immigration. Cerrato, Ferrara and Ruggieri (2018) Tella and Rodrik (2020) show that support for trade protection rises especially strongly when respondents are confronted with a scenario of job loss due to outsourcing to a developing country, but also to a lower extent when job loss is due to other labour market shocks such as technological change. 57 Trade policy has not always been a strongly partisan topic in the US. In 2009, a survey by the Pew Research Center found that the fraction of registered voters who stated that trade agreements between the US and other countries were bad for the US was about one-third both among self-identified supporters of the Republican party and of the Democratic party (Jones, 2018). By 2018, the fraction of respondents who thought trade agreements were bad had increased to 46% among Republicans, and declined to 19% among Democrats. Feigenbaum and Hall (2015) note that both Republican and Democratic politicians who represent US congressional districts with high exposure to Chinese import competition have become more likely to vote against trade liberalisation.

Survey data indicate that voters who supported the UK's 2016 Brexit referendum saw Britain's exit from the European
Union as an opportunity to achieve greater sovereignty and reduced immigration, but also to achieve greater trade openness through new trade agreements (Owen and Walter, 2017). theory provides several explanations for such an impact of adverse economic shocks on nationalism and anti-immigration sentiment. Grossman and Helpman (2018) argue that economic decline can increase people's identification with their own social group, while Gennaioli and Tabellini (2019)

note that shocks can shift group allegiance from economic classes (high-and low-income) to culture-based identity that pitches nationalists against cosmopolitans and immigrants.
Politicians may also strategically emphasise controversial issues in order to strengthen cultural identification and increase turnout among their supporters (Glaeser, Ponzetto and Shapiro, 2005). Indeed, the widely perceived backlash against globalisation, as reflected in many recent political outcomes, may be less a consequence of a swing in public opinion against globalisation than a result of its politicisation (Walter, 2021).

Policies to support globalisation's losers
The realisation that globalisation had adverse economic and social impacts on some segments of the population in high-income countries raises the question of how governments could best support such losers from trade. We discuss below evidence on compensation of earnings losses, programmes that facilitate labour market adjustment, place-based policies for the revival of economically depressed locations, and trade policies that seek to mitigate exposure to trade shocks. A common theme for all of these policy options is that even when they are effective in helping trade's losers, they tend to generate important trade-offs between equity and efficiency.

Compensating income losses
Economists have long realised that trade could be financially beneficial for everyone. 59 For this to be the case, however, it is necessary that the beneficiaries from trade use a part of their gains to provide compensation to trade's losers. In practice, such redistribution occurs primarily through general tax and transfer systems that shift resources from employed and high-income individuals towards unemployed and low-income individuals.
Evidence on government transfers triggered by trade shocks is scarce. Autor, Dorn and Hanson (2013a) show that US households in local labour markets with greater exposure to Chinese import competition report larger receipt of social security and welfare benefits. Comparing local labour markets at the 75 th and 25 th percentiles of import exposure over the 2000-07 period, the social security and welfare payments per working-age adult increased by 2.2% more in the former location. This elasticity is very similar to that of wage and salary incomes, which declined by a differential 2.3%. However, when measured in dollar rather than percentage terms, the estimated increase in social security and welfare income amounts to just an additional $18 per adult and year, which falls far short of an estimated $581 decline in wage and salary income.
In addition to analysing households' self-reported social security and welfare income, Autor, Dorn and Hanson (2013a) studied government data that cover a broader measure of transfer expenditures, which additionally includes unemployment benefits, medical benefits and educational assistance. Some of these transfers, such as the quantitatively important medical assistance, comprise in-kind benefits rather than financial payments to households. All of the aforementioned types of transfers differentially increased in more import-exposed locations. Again comparing local labour markets at the 75 th and 25 th percentiles of import exposure over the 2000-07 period, the former experienced an additional $63 per-capita growth in annual transfer payments, with one-third of that effect coming from medical benefits alone. However, this increase in government transfers still only compensates for a modest portion of the tradeinduced earnings decline. 60 The situation is different in countries with more extensive welfare systems. Utar (2018) finds that Danish workers in firms directly threatened by the expiration of the Multifibre Agreement experienced a 3.7% reduction in annual earnings, but 'personal income' -which includes government transfers, business and self-employment income -only fell by around 1%. However, even in Denmark, workers may not always be well insured against negative income shocks due to trade. Hummels et al. (2014) report that a doubling of Danish firms' offshoring reduced low-skilled workers' labour earnings by 1.5% in the following year, while a measure of 'gross earnings' that also includes unemployment insurance benefits and social assistance income declined only modestly less by 1.3%. , de Gortari and Itskhoki, 2017). However, a lack of adequate income support can have grave consequences for workers whose jobs and communities are exposed to adverse trade shocks, such as the workers in import-competing US localities who report that they are unable to afford a doctor's visit (Lang, McManus and Schaur, 2018).

One reason for not implementing a full compensation is that a redistribution of income via taxes and transfers from trade's winners to the losers can generate a sizeable loss of economic efficiency (Antràs
Lowering adjustment costs Adverse labour market impacts of trade shocks such as rising import competition from China are differentially concentrated among the employees of the industries most exposed to such shocks and in the geographic regions where such industries are concentrated. Many workers do not quickly adjust to trade shocks through mobility across sectors and space. Active labour market policies (ALMPs) seek to reduce adjustment costs in the labour market by facilitating workers' re-employment through such measures as job search counselling, retraining, or employment subsidies. 61 Recognising trade's potential to disrupt labour markets, the US set up a trade-specific ALMP called Trade Adjustment Assistance (TAA) in 1962. President Kennedy explained that 'When considerations of national policy make it desirable to avoid higher tariffs, those injured by that competition should not be required to bear the full brunt of the impact. Rather, the burden of economic adjustment should be borne in part by the Federal Government' (Kennedy, 1962). 62 In its current form, the TAA programme primarily covers the costs of retraining unemployed individuals. The majority of workers in TAA-sponsored training enrol in occupational training programmes -for instance, for such activities as computer operators, office clerks, medical assistants or nursing aides (Hyman, 2018). During the TAA-sponsored training, programme 60 Barrot et al. (2017) show that households in more import-competing local labour markets instead accumulate greater debt in the form of home equity extraction and credit card debt. 61 (Re)training can also be offered by firms to their employees. However, Costa, Dhingra and Machin (2019) (Hyman et al., 2021).
Across local labour markets, TAA expenditures grow particularly rapidly in locations that face a greater growth of Chinese import competition (Autor, Dorn and Hanson, 2013a) and that experience greater employment declines (Kondo, 2018). However, the TAA programme is small compared with many other transfer programmes that target the unemployed or low-income households. As a consequence, TAA accounts for less than 1% of the increase in total government transfer spending that results from a local labour market's exposure to import competition (Autor, Dorn and Hanson, 2013a).
Given the small scale of trade-specific programmes such as TAA or the EU's European Globalisation Adjustment Fund (EGF), it is likely that many workers who are displaced by trade 63 Overall, 63% of all petitions are approved. A simple comparison between all accepted and rejected petitioners is complicated by the fact that underlying economic circumstances systematically differ across the two groups, as successful petitioners apparently were able to better document that layoffs occurred due to large trade shocks. However, petitions are judged by individual case investigators who vary in their approval leniency, which creates a source of randomness in the evaluation process where a marginal petition is more likely to be approved when it is assessed by a more lenient investigator. Hyman's (2018) analysis exploits this institutional feature in an instrumental variables design that compares workers with marginal petitions whose acceptance probability was influenced by the discretion of the case investigator. 64 Hyman (2018) assumes that the economic deadweight loss that results from taxation could be anywhere between $0.25 and $0.75 per dollar of tax revenue, which yields an estimated internal rate of return of 9% for the former and 0% for the latter case.
shocks will be supported by general ALMPs that are available to all unemployed workers, rather than the ones directly connected to trade. 65 Indeed, the UK never applied for any EGF funds and instead preferred to fully rely on its own domestic programmes to support redundant workers (UK Parliament, 2016). A recent meta-analysis of several hundred mostly European ALMP evaluation studies shows a large dispersion in effectiveness across such programmes (Card, Kluve and Weber, 2015). About two-thirds of the analysed studies found statistically significant positive effects of ALMPs on employment probabilities two or three years after the end of a programme, with ALMPs supporting human capital accumulation yielding the best outcomes. Most of the programme evaluation studies did not report programme costs, and hence the metaanalysis does not report information on cost-benefit calculations. 66 Overall, the evidence on trade-specific and general ALMPs suggests that such programmes can be helpful at easing displaced workers' labour market adjustment. Different from transfer payments to the unemployed, programmes that help workers to find new jobs mitigate not only displaced workers' financial losses, but also the adverse social outcomes associated with persistent joblessness. However, not all ALMPs achieve their goals to help workers find good jobs, and even when programmes are effective, they do not always perform well in cost-benefit calculations.
Reviving local economic activity An important result of the recent literature on local labour market effects of trade is that adverse shocks do not lead to large and rapid population adjustments. Instead, displaced workers are often spatially immobile, and adverse labour market outcomes such as low employment rates and depressed earnings remain heavily concentrated in the locations whose industries were most exposed to adverse trade shocks, even years after such shocks subside.
An alternative or complement to supporting displaced individuals with financial transfers or active labour market policy programmes is to direct support at depressed regions, cities or neighbourhoods via place-based policies. Such policies can take the form of place-based redistribution, where a system of taxes and transfers is conditioned on individuals' location rather than on their individual income level. Gaubert, Kline and Yagan (2021) show that placebased redistribution can be more efficient than income-based redistribution under some conditions, especially if the poor population is spatially concentrated and if the place-based redistribution does not induce strong migration from thriving to depressed locations. They also provide survey evidence which suggests that there is more popular support for targeting tax credits to poor people who live in distressed places, rather than poor people independent of location.  (Claeys and Sapir, 2018). 66 See Crépon and van den Berg (2016) for another recent review of the literature on ALMPs. cohesion policy that seeks to reduce economic disparities across European regions. An important pillar of this policy is EU Structural Funds that are directed at poorer European regions whose per-capital income is below 75% of the EU average. These funds are often used to finance projects related to transportation and energy infrastructure. Becker, Egger and Ehrlich (2010) estimate that regions just below the 75% income threshold for eligibility experience a 1.6 percentage point greater growth of per-capita GDP over periods of about five years from the 1990s to the early 2000s, compared with regions just above the 75% threshold which are not eligible for this funding. However, funding did not lead to differential regional employment growth. Moreover, the Structural Funds stimulated greater GDP growth only in about a third of the eligible regions, and were ineffective particularly in regions with low human capital levels and poor political governance (Becker, Egger and Ehrlich, 2013).
Governments often use place-based policies that provide subsidies to specific firms in depressed areas. In the UK, the Regional Selective Assistance (RSA) programme provided discretionary grants to firms in regions with low income levels and high unemployment. 67 RSA financed up to 35% of the costs of investment projects that involve capital expenditure on property, plant or machinery, with most funding going to firms in the manufacturing sector. Approved projects were those expected to either generate new employment or protect jobs that might be lost otherwise. Criscuolo et al. (2019) investigate the consequences of a change in regional eligibility for RSA funding that was mandated by the periodic updating of the EU's list of economically disadvantaged regions. They find that locations that were more likely to become eligible for RSA based on their characteristics in the early 1990s experienced a significantly smaller decline in manufacturing employment and a larger decrease in unemployment during the early 2000s. The authors estimate that the direct costs of the RSA programme and its indirect costs via distortionary taxation amount to only about $3,500 per manufacturing job that the programme managed to create or preserve.
Broader reviews of spatially targeted programmes (e.g. Kline and Moretti, 2014;Neumark and Simpson, 2015;Ehrlich and Overman, 2020) (Read, 2005) Jabbour et al. (2019) and Bown et al. (2020) show that the more widely used anti-dumping tariffs, which are justified by other countries' unfair pricing practices, had no positive aggregate effects on employment in France and the US, respectively. Such tariffs did little to shore up domestic employment in protected industries, but were associated with job losses and lower wages in downstream customer industries. 70 The MFA, which was originally instituted in 1974, was gradually phased out from 1995 to 2005. However, the final removal of MFA quotas in 2005 still caused a major shock to the textile and apparel sectors in high-income countries. In the US, prices of previously quota-protected goods experienced a sudden drop of 38% (Harrigan and Barrows, 2009). increased from 3% to 21%, while the average tariff that the Chinese applied to US goods rose from 8% to 22% (Bown, 2021).
The US tariff hikes were widely seen as part of President Trump's agenda to bring back jobs to America (Tankersley, 2019). Yet there is little evidence for employment gains. Flaaen and Pierce (2020) study the impact of the US-China 'trade war' on employment in broad US manufacturing industries, and find that the industries that were protected by higher import tariffs did not add significantly more jobs, while downstream customer industries faced job losses. An analysis of employment across all economic sectors in local labour markets also finds no evidence for substantial job gains in localities that were specialised in tariff-protected industries (Autor et al., 2021). Instead, employment fell in locations whose exporting industries were more exposed to other countries' retaliatory tariffs. In addition to the overall negative impact of the tariffs on US employment, the US import tariffs also contributed to an increase in prices faced by US firms and consumers (Amiti, Redding and Weinstein, 2019;Fajgelbaum et al., 2020;Flaaen, Hortaçsu and Tintelnot, 2020;Cavallo et al., 2021).
While tariffs and quotas are relatively easily measurable thanks to their quantitative nature, there exists less systematic information and thus less research on restrictions of trade through regulatory policies. Countries may, for instance, restrict imports through onerous and internationally incompatible product safety standards. Leonardi and Meschi (2020)
Additional studies for France (Malgouyres, 2017a) and Italy (Citino and Linarello, 2019) also pursue a similar empirical approach but use different measure of import exposure or labour market outcomes. We therefore include them only partially in our analysis below.
There are a number of methodological differences also between the papers included in our main comparison.   Table B1. Table B3 translates the estimated effects of a unit increase in import competition into implied losses in manufacturing employment. We obtain these by multiplying the coefficients in Table B1 by the average change in Chinese imports per worker in each country. We are also able to include in this comparison results for Italy, which are based on a different measure for import competition. The final column of Table B3 indicates the annualised change in the percentage of working-age population working in manufacturing given the product of average import growth in a country and employment change per unit of import growth. The pattern of manufacturing decline across countries remains similar to that in Table B1, with the smallest negative employment effect in Norway and the largest one in Spain. The impact of Chinese import competition on manufacturing employment is similar in the UK and the US, where the US faced a larger growth in Chinese import competition than the UK, while the UK experienced a larger decline of manufacturing per unit of import exposure.
Results for additional local labour market outcomes in Table B4 indicate negative effects of import shocks on overall employment and wages. In this broader set of labour market effects, we are able to include France, where the impact on the log count of manufacturing workers of a $1,000 increase in Chinese imports is slightly larger than in the US, but smaller than in Spain. 72 Results in these papers are reported for local currencies in different years -for instance, in € 2005 in the case of Dauth, Findeisen and Suedekum (2014 Donoso, Martin and Minondo (2015) report results based on nominal rather than real currency units.    Autor, Dorn and Hanson (2013a) (US), Foliano and Riley (2017) (UK), Dauth, Findeisen and Suedekum (2014) (Germany), Balsvik, Jensen and Salvanes (2015) (Norway), Donoso, Martin and Minondo (2015) (Spain) and Malgouyres (2017a) (France), and are scaled to units of $1,000 (2007) per worker (except for Spain).
Controls for US, UK, Germany, Norway and Spain are the same as outlined in Table B1. For France, they are the initial share of college-educated residents, of women in the employed population, foreigners, and production workers in the workforce as well as 22 regional fixed effects, period effects and the log of initial total employment. Non-manufacturing employment for Norway excludes public sector employment. Wage changes are defined differently in different countries: for the US, it is the average log weekly wage; for Germany, it is the log change in regional median wage; for Norway, it is average log earnings; and for Spain, it is the change in the log average wage. Malgouyres (2017a)

C. Analysis of the impact of Chinese import competition on UK consumer prices
Measuring import penetration and employment changes for different products Our price analysis makes use of data on imports and exports of commodities, employment in different industries (taken from Office for National Statistics (2020)) and the prices of consumer products (taken from the Consumer Prices Index).
We take our import and export data from the UN Comtrade database at the level of six-digit Harmonised System product codes. We map these to 836 Classification of Product by Activity We define the growth of import penetration for a product between year ! and year " as Regression sample We exclude services, including the category telephone equipment and services as this is predominantly spending on services. Our price data distinguish 49 goods products, which is more aggregated than the 58 distinct COICOP groups for which we have data on import penetration. For those nine goods where the price data are more aggregated than the import data, we calculate the weighted average of import exposure for industries using spending shares from the Living Costs and Food Survey as weights. Due to data limitations, we also exclude recreational durables during the 1994-98 and 1999-2007 periods, and cars and sports equipment for 1994-98, as price data for these goods do not go back to 1994. This leaves us with a sample of 46 goods for which we have import penetration, price and employment data for the whole period, and 48 goods with complete data from 1999 onwards.
In our regression analysis, we also make use of the change in import penetration across two periods, 1994-98 and 1999-2007. Since the two periods are not of equal length, we use annual average changes in prices, employment and import penetration. There is a rapid and sustained increase in reported UK imports from China between 1999 and 2000, which most likely reflects a change in the treatment of imports from Hong Kong which originated in China (Baranga, 2018). For this reason, we include imports from Hong Kong in our measure of Chinese import penetration for the UK (but not in our measures of Chinese imports into other countries, as they are not affected by this issue).
Alternative employment regressions An alternative way to estimate the impact of Chinese import competition on different industries is to use more tightly defined SIC industry codes, instead of the product groups for which we also have price data. Panel a of Table C1 shows results for the effect of Chinese import competition estimated in the same way as we did in Table 1 but using 362 industries defined at either the four-or three-digit level (depending on whether it is feasible to map commodity codes to fourdigit or only to three-digit industries) rather than at the level of COICOP industries. The employment impacts are in general smaller than those in Table 1, and indicate employment losses of between 0.8% and 1.6% per percentage point increase in import penetration once we control for two-digit industry fixed effects. These figures are similar to those reported by Acemoglu et al. (2016) and Jaravel and Sager (2019) for industry-level employment regressions in the US. Table C1 reports results when we conduct the employment analysis at the level of more aggregate two-digit SIC industries. The employment impacts are considerably larger, suggesting that the level of aggregation can affect the estimated effects. This may partly reflect attenuation biases as it can be difficult to exactly allocate imports to detailed industries that produce similar outputs. In the main text, we report impacts at a level consistent with our consumer goods categories. These estimates lie between the estimates shown in panels a and b of Table C1 once we control for industry fixed effects.