Costs and Benefits of Trade Shocks: Evidence from Chilean Local Labor Markets

We study Chile’s labor market responses to trade shocks during 1996-2006, exploiting spatial and time variations in trade exposure arising from initial differences in industry specialization across local labor markets and the evolution of shocks across industries. We take advantage of China’s supply and demand’s worldwide shocks to instrument for Chinese import competition and demand for Chilean exports. Our main finding is that increasing manufacturing import competition implied a significant rise in labor informality in more exposed local markets, especially among young and unskilled workers. These groups also suffered significant relative wage losses. Meanwhile, locations that benefited most from the increased demand for primary products experienced a relative increase in employment, particularly among young individuals, and reallocation from self-employment towards salaried jobs in the formal sector, along with relative wage gains among old-age workers. Interestingly, these areas experienced a smaller increase in tertiary education enrollment rates than less exposed areas. JEL Classification: F14, F16, J23, J31, L60, O17, Q02, R12, R23.


I Introduction
The study of labor market responses to changes in trade flows has long been a subject of study in Economics. The literature has placed particular emphasis on the impact of trade shocks on certain labor market variables, such as employment and wages, stressing the differential impact across sectors, occupations and skill groups (Heckscher-Ohlin, Ricardo-Viner). In this paper, we depart from this approach and join a newer line of research that has focused on the effects of trade on individuals living in geographic areas with different patterns of industrial specialization. 1 These papers typically find substantial differences in the spatial distribution of the costs and benefits of trade.
We study the causal effect of trade shocks on Chilean local labor markets, exploiting spatial and time variations in trade exposure arising from initial differences in industry specialization across geographic locations and the evolution of shocks across industries.
Besides considering typical outcomes such as employment, unemployment and wages, we focus on labor informality, a distinctive feature of labor markets in developing countries.
We build upon recent research that finds that labor informality is a relevant margin of adjustment to trade shocks. 2 3 We also explore the impact of trade shocks on tertiary schooling, in line with the literature that finds that trade liberalization amplifies initial differences in regional factor endowments over the long-run through its effects on human capital formation. 4 Our main findings suggest that the costs and benefits of trade shocks depend on the local employment structure and vary considerably across sub-population groups. Locations with greater exposure to increasing import competition from China experienced a relative decline in manufacturing employment and wage losses for unskilled and young workers (mainly, in the service sector). Interestingly, labor informality plays a key role in the responses to trade shocks. In particular, we find that rising import competition implied a significant relative rise in labor informality, mainly among male, young and unskilled workers. On the benefits side, we find that locations specialized in the production and processing of primary products experienced a relative increase in employment, driven by the expansion of Chinese demand for Chilean exports. Employment growth mainly benefited young and male individuals. Interestingly, these areas also experienced a slower increase in tertiary education enrollment rates as compared to areas with less exposure.
These groups experienced a relative decline in manufacturing employment and wages, and a relative increase in labor informality (unregistered jobs with no access to social security). This is consistent with unskilled (mostly young) workers displaced by import competition relocating to new jobs, often in the service sector and under conditions of labor informality. Meanwhile, unskilled and older workers that remained employed in manufacturing experienced wage losses and rising rates of informality. Informality allows firms to reduce tax and other labor costs at the risk of being caught and forced to pay a fine. Facing stronger competition from imports, some firms may take that risk.
Firms may also replace permanent employees with part-time workers and apprentices, and subcontract some production activities with other plants in the informal sector. On the other hand, workers may be more likely to accept an informal labor arrangement in the context of a weaker labor market. Additionally, we find that increasing import competition is associated with a moderate relative increase in the employment rate of young people with some tertiary education which, we argue, might be partially explained by the "added worker effect" hypothesis: as a result of the worsening labor conditions for some members of the household (e.g. old-age workers), secondary workers (e.g. children studying at university or a tertiary institute) may enter the labor market in order to ameliorate negative shocks. In fact, we find that locations facing larger import shocks experienced a relative decline in the fraction of individuals enrolled in tertiary education.
On the other hand, the main benefit from trade was a relative increase in employment in locations specialized in the production and processing of primary products, especially among young workers and males (both skilled and unskilled). Our estimates suggest that job opportunities for young individuals in the primary sector took the form of temporary contracts (presumably short-term or apprentices) instead of permanent wage jobs with social security contributions. In contrast, old-age workers experienced relative wage gains and reallocation from self-employment towards salaried jobs in the formal sector.
Finally, locations favored by the demand shock experienced a lower relative rise in tertiary education enrollment, which may hinder local development through its effects on human capital formation. Overall, our findings reveal considerable heterogeneity in the spatial distribution of the costs and benefits of trade across sub-population groups.
Our research contributes to a vast literature that examines the impact of Chinese import competition on domestic workers, firms, and markets. Most papers in this field emphasize the negative effects of Chinese imports on employment and wages, which generally focus on the most vulnerable firms, workers, and in particular occupations. 11 The deterioration of labor market conditions driven by import competition can lead to worse health outcomes, an increase in crime rates and lower provisions of public goods in affected locations. 12 On the benefits side, manufacturing productivity may rise in response 11 See Autor et al. (2013); Autor et al. (2014); Utar (2018);and Cabral, Martins, Pereira dos Santos, and Tavares (2021). For evidence at the firm-level, see Alvarez and Claro (2009), Pierce and Schott (2018), César and Falcone (2020). 12 Lang, McManus, and Schaur (2016), Feler and Senses (2017), Dix-Carneiro, Soares, and Ulyssea (2018), Che, Xu, and Zhang (2018), Colantone, Crino, and Ogliari (2019), Autor, Dorn, and Hanson (2019), 4 to import competition through within-firm innovation and input reallocation towards more efficient plants (Utar and Torrez Diaz (2013), Bloom, Draca, and Van Reenen (2015)).
Some workers might attempt to return to their education after losing their job to import competition (Utar (2018)) and young individuals may be more willing to finish secondary education (Greenland and Lopresti (2016)). Generally, consumers gain from trade due to lower prices (Traiberman (2019), Jaravel and Sager (2019)).
Our paper is most closely related to recent works that study the impact of both supply and demand shocks on local labor markets. Dauth et al. (2014Dauth et al. ( , 2017 document that regions in Germany specialized in export-oriented industries experienced employment gains that outweighed the employment losses of regions specialized in import-competing industries (but net effects were driven by the rise of Eastern Europe rather than China) and conclude that globalization fostered manufacturing employment in Germany. Egger, Kaynak, and Zoller-Rydzek (2020) document that Turkish regions more exposed to increasing demand for exports from China experienced relative employment gains, while import-competing regions encountered very small effects on the labor market. Costa et al.
(2016) suggest that Brazilian regions facing stronger import competition exhibited relative slower growth in manufacturing wages than less exposed areas, while regions specialized in commodities had the benefit of relative wage gains and shifts towards formal employment.
We contribute to this literature by providing causal evidence on the heterogeneous effects of trade shocks across sub-population groups by exploiting spatial and time variations in trade exposure across local labor markets in a small and open developing economy. Our results confirm previous findings in the literature and add some novel contributions regarding the roles played by labor informality and labor supply among secondary household members in response to import competition. In particular, we find that the informal sector acts as a buffer against a negative shock on the local labor market, and that secondary workers may enter the labor market to help ameliorate the shock.
The rest of the paper is organized as follows. Section II comments on the recent economic history and reforms of Chile and China and presents a brief theoretical discussion.
We discuss the identification problem and the empirical strategy in Section III, and include a detailed description of the data and summary statistics. All empirical findings are presented and discussed in Section IV. We finish with some concluding remarks in Section V. Additional discussions, figures and tables are included in the Appendix. The Chilean case provides a nice scenario in which to study the causal impact of trade shocks on the labor market.
On the other hand, China carried out a broad set of structural reforms beginning in the 1980s, which transformed its agrarian structure into a modern industrialized economy and a world-leading producer of manufactures. The main trade reforms pursued a dualistic regime characterized by import-substitution and export-promotion policies (Naughton, 1996). Alongside these reforms promoting economic growth and trade, the country's Much of China's growth was driven by massive migration from rural to urban 13 While some tariffs exceeded 100 percent in 1974, five years later they were reduced to a uniform ad-valorem tariff of 10 percent. Tariffs increased 35 percent during the recession of 1982-84, but declined to 20 percent in 1985. NTBs were not applied during this transitory period. See Levinsohn (1999) and Pavcnik (2002). 14 The World Bank, World Development Indicators database. 6 regions, strong investments in infrastructure, increased access to foreign technologies, intermediate inputs and capital goods, a massive inflow of foreign direct investment, and a stunning increase in total factor productivity (TFP). According to Brandt et al. (2012), China had an average annual growth in manufacturing TFP of 8 percent over the period 1998-2007. The growth in exports driven by these factors inherent to Chinese economic forces and institutions, provides a potential exogenous shock for firms and workers all over the world. Relatedly, China's income growth, urbanization, and increasing economic integration with the rest of the world, among other factors, brought about a sharp increase in Chinese demand for commodities (e.g. crude petroleum, soybeans, iron and copper), natural resource-intensive manufactures (refined fuels, refined metals and processed food products) and intermediate inputs and capital goods to supply its increasing manufacturing production. Overall, China has become one of the leading-sources of global supply and demand in many key sectors of international trade. Given that China mainly exports labor-intensive low-price consumer products, Chinese imports represent a competitive pressure for domestic producers. César and Falcone (2020) document that around 70 percent of Chinese exports to Chile during 1996-2006 were final consumption goods, whose import prices were on average significantly lower than those from other countries within detailed product categories. The authors find that Chilean manufacturing plants in industries more exposed to Chinese import competition exhibit a relative decline in revenues, employment and physical capital, and face a higher probability of exiting the market as compared to manufacturing plants in less exposed industries (and that all these effects were stronger for plants with low initial levels of productivity). We use this piece of evidence as a motivation to study whether Chinese import competition (and Chinese demand for Chilean exports) have affected Chilean local labor markets differently. By assuming that labor is not mobile across LLMs, we are able to measure general equilibrium effects at the local level, considering that workers might move across firms, occupations, industries and formal-informal jobs in response to import supply and export demand shocks.

Figure 1 Evolution of China's participation in Chilean trade
Notes. China's import (export) share measured as the value of Chilean imports from (exports to) China divided by Chile's total imports (exports). Source. BACI.

II.2 Theoretical channels
China's supply and demand shocks in Chile are big enough to expect significant effects on the Chilean economy and, in particular, on its labor market. In this section we discuss along which dimensions China's rise might affect Chilean labor markets, considering potential effects on wages and employment, but also alternative adjustment channels, such as the informal sector. Theoretical predictions depend on certain assumptions about labor and capital mobility across firms, sectors and regions, which are key in shaping transitional dynamics between the short-and long-run. Empirical findings and the relevant mechanisms involved are context-specific, and depend on the nature of trade policy changes and its interaction with labor market institutions (Goldberg and Pavcnik (2003), Bacchetta, Ernst, andBustamante (2009), Pavcnik (2017)).
The traditional comparative advantage model (Heckscher-Ohlin) highlights the potential disruptiveness of trade liberalization, assuming that labor is perfectly mobile across sectors. Countries' relative endowments of productive factors translate to the factor 8 content of traded goods and determine relative prices. Assuming that China is relatively abundant in unskilled labor and scarce in natural resources, its opening to trade represents an increase in world's supply of unskilled labor and a rise in world's demand for natural resources. Then, international prices should decline (increase) in unskilled labor (natural resources) intensive industries. 15 However, the impact of trade can be heterogeneous across firms even if they belong to the same industry (Melitz (2003), Bernard, Eaton, Jensen, and Kortum (2003)). While some plants contract and exit the market in response to trade liberalization, others may grow substantially as trade fosters within firm productivity through several channels, such as access to intermediate inputs, capital goods and technologies from abroad, changes in the task content of production, product switching, quality-upgrading, skill-upgrading, scale economies and organizational changes. 16 In this context, trade impact is expected to be heterogeneous across workers as well, depending on their characteristics, those of the firm in which they work, and their joint adaptability to changes in the trade environment.
Additionally, if there are frictions in the labor market, trade may affect the equilibrium levels of employment and unemployment (Davis andHarrigan (2011), Helpman, Itskhoki, andRedding (2010), Helpman and Itskhoki (2010)). Frictions to labor mobility may delay the adjustment process, reducing the potential welfare gains from trade, while frictions to capital mobility and agglomeration economies may amplify the short-term effects of trade over the long-run (Dix-Carneiro (2014), Dix-Carneiro and Kovak (2017)).
In the short-run, import shocks may force local firms to become more competitive in order to survive. Firms can potentially diminish labor costs by cutting workers' benefits, replacing permanent with part-time labor and subcontracting with other plants in the informal sector. 17 Some of these firms may exit the market (presumably, those with lower initial levels of productivity). If contracting plants are more intensive in the use of informal workers than surviving and expanding firms, trade may even reduce the share of labor informality. Paz (2014) develops a small open economy framework with heterogeneous firms in which tariffs affect firm's payroll-tax compliance decisions, and show that trade liberalization has an ambiguous effect on the employment share of informal workers, that depends mainly on initial labor market conditions (i.e. the lower the initial share of labor informality the more likely that trade liberalization increases labor informality).
Recent search and matching models of informality allow for worker and firm heterogeneity, imperfect substitutability across workers within the firm, decreasing returns to scale, and convex vacancy-posting costs, and suggest that the interaction between the skill composition of labor and productivity may have first-order implications for the evolution of labor informality (Haanwinckel and Soares (2021)).
In this paper we do not aim to test the particular predictions of any of these models.
Naturally, we expect that export demand shocks led to higher employment and wages in exposed locations, while import supply shocks led to an opposite effect, recognizing that the impact may vary across different sub-population groups and expecting that younger and more educated individuals adjust more easily to trade shocks. We also recognize that regions face both (trade-induced) supply and demand shocks, so the final effect of trade on local labor markets will depend on local workforce composition, firms' responses, patterns of labor and capital re-allocation and general equilibrium effects (e.g. sectoral linkages, demand multipliers) that, ultimately, will affect firms and workers differently depending on several unobserved economic primitives.

III.1 Approach and identification
Our empirical approach exploits cross-local labor market variation in the evolution of Chinese import and export exposure stemming from the interaction of initial differences in local employment composition by industry and the temporal evolution of Chinese import competition and demand for exports across industries. LLMs with a higher initial participation of workers in largely exposed manufacturing industries such as textiles and toys/related products will be more exposed to the import competition shock, while LLMs with a higher share of workers employed in the mining of metal ores industry will be more exposed to the export demand shock. Patterns of industry specialization are constructed with data for 1992, some years before the period under study (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) to take into account that trade shocks may affect the employment distribution across industries. 18 In order to study the effect of Chinese import competition and demand for exports on Chilean LLMs we estimate the following regression equation: where Y is an outcome of interest such as employment or informality rates and r and t index LLMs and time, respectively. CIC is a measure of Chinese import competition computed as the total value of imports from China divided by domestic absorption (production minus net exports). At the industry-level, it is computed as: where Q jt , M jt and X jt are the value of production, imports, and exports, respectively, for industry j in year t.
CDE is a measure of Chinese demand for exports computed as the total value of exports to China divided by domestic production. At the industry-level, it is computed as: To construct Chinese import competition and Chinese demand for exports at the LLM-level as required in equation (1), we interact the initial pattern of industrial composition in each LLM, as given by the local employment share in 1992 (w 1992 rj ), with the value of CIC and CDE at the industry-level: Dependent variables (Y rt ) are labor market outcomes such as the employment and unemployment rates, labor informality rates, (log) average hourly wages, (log) average hours worked, (log) average monthly wages, and the employment share of broad economic sectors (primary, manufacturing and services). To account for potential heterogeneous effects, we run separate regressions for different sub-population groups according to skill, age, gender and sector of employment. In all cases we include LLM (α r ) and year (δ t ) fixed effects and therefore exploit within LLM variation in the exposure to CIC and CDE over time. The preferred specification controls also for three types of preexisting trends: (i) the corresponding outcome variable, to address the concern that any result might represent the continuation of local trends beginning some years before the period under study; (ii) the sum of industry employment shares in the primary and manufacturing sectors, which were used to construct the shock variables (as recommended by Borusyak, Hull, and Jaravel (2020)); 19 and (iii) female labor force participation (LFP). The regional balance test (presented in the Appendix) suggests that LLMs more exposed to CIC exhibit higher initial levels of female labor force participation (LFP), so we decided to account for these differences in the preferred specification. 20 Preexisting trends are constructed as the value of each variable in 1992 interacted with year fixed effects.
We weight each observation by the 1992 LLM share of national labor force. This estimation strategy provides average treatment effects that are weighted by workers instead of LLMs. 21 A potential concern is that labor mobility across LLMs could bias our estimates if workers migrate across locations in response to trade shocks. 22  To study the long-term effect of Chinese import competition and demand for exports on Chilean LLMs, we estimate regressions in changes (1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012), and both changes together-stacked changes). The regression equation is the following: ∆Y r,t:t−1 = β 0 + β 1 ∆CIC r,t:t−1 + β 2 ∆CDE r,t:t−1 + X r,1992 + ε r where ∆Y r,t:t−1 is the change in the corresponding outcome variable at the LLM r (employment and unemployment rates, share of salaried workers, share of self-employment, share of population of working age attending tertiary education) and X r,1992 are control variables measured at the initial year. The preferred specification controls for: (i) employment shares in manufacturing and primary sectors (used to construct the shock variables) and (ii) female LFP. Regressions in stacked changes control also for a period dummy. The construction of ∆CIC r,t:t−1 and ∆CDE r,t:t−1 follows the same logic of equations (4) and (5)  To account for endogeneity concerns, we apply an instrumental variable strategy that has been used in other related studies (e.g. Autor et al. (2013Autor et al. ( , 2014 specific manufacturing and primary industries over time. We use these variables to predict 24 Another potential confound that is expected to be more serious over the long-run refers to compositional changes in local workforce. 25 To categorize countries as middle-income or high-income we follow World Bank classifications.

13
CIC and CDE in Chilean industries. These variables are exogenous as they do not depend on the decisions of Chilean firms or consumers.
To construct both instrumental variables at the LLM-level, we calculate: In the long-term analysis, we compute the instrumental variables in changes (1992-2002 and 2002-2012) (2020)). Borusyak et al. (2020) develop an econometric framework in which identification follows from the quasi-random assignment of shocks while exposure shares are allowed to be endogenous. This framework seems appropriate in our context if we assume that industry shocks (i.e. China's supply and demand shocks to global trade) are exogenous to Chilean industries while local industry employment shares are not because unobserved industry shocks may affect regional outcomes through the same combination of exposure shares. 26 We perform balance tests at the industry and region levels and falsification tests that corroborate the plausibility of the identifying assumptions (see section A.2). Additionally, we conduct a robustness exercise by applying the econometric framework developed by Borusyak et al. (2020).

III.2 Data and descriptives
In order to implement the strategy outlined in the previous section, we combine different data sets: household surveys, census data, firm surveys and trade data.
The Chilean national household survey (Encuesta de Caracterización Socioeconómica Nacional-CASEN) provides information on Chilean workers located throughout the country. It is conducted by Chile's Ministry of Planning (MIDEPLAN) every two or three years. We base our analysis on the 1992, 1996, 1998, 2000, 2003, and 2006  To construct CDE we take production for primary industries from National Accounts and manufacturing from ENIA.
27 Both measures of labor informality are highly correlated at the LLM-level (the correlation coefficient is .94) and somewhat less correlated at the individual level (.75). 28 The authors created a data set classifying 302 municipalities into 62 LLMs by means of an optimization algorithm based on evolutionary computation using commuting data from the Chilean Internal Migration Database (CHIM) and Censuses data for 1982, 1992 and 2002. 29 The advantage of using municipalities is that we increase the number of observations. However, it comes at the cost of augmenting the standard errors of our estimates because municipalities are by definition smaller than LLMs and, in many cases, are part of deeply integrated economic clusters with high mobility of workers across neighboring municipalities. 30 To perform the matching between household and census data we created a concordance table that assigns each three-digit industry reported in the 1992 Census to a single two-digit industry of the International Standard Industrial Classification (Rev. 4). Figure 2 plots the evolution of CIC (top panel) and CDE (bottom panel). CIC was fully concentrated in manufacturing. 31 Industries most exposed to CIC were leather and related products, clothing and apparel, toys and other miscellaneous items, and textiles.

Trade variables
The average annual growth in CIC across these industries range from 3.4 p.p. to 4.5 p.p..
Industries that were mildly exposed to Chinese competition were electrical equipment, machinery, computer, electronic and optical products, furniture and wood manufactures, fabricated metals, and rubber and plastic products. The average annual growth in CIC across these industries range from 0.6 p.p. to 1.8 p.p. The rest of the manufacturing industries were minimally exposed to Chinese import competition.
On the other side, Chile exports both commodities and processed primary products (which are included in manufacturing) to China. Noticeably, Chilean exports to China exhibit a high degree of concentration in a small number of products, mainly from the mining sector. On average during 1996-2006 one third of Chilean exports to China were primary products (within this group, almost 90 percent involved mining of non-ferrous metal ores such as copper) and two thirds were manufacturing products (57 percent are manufactures of basic precious and non-ferrous metals-mainly processed refined copper, 22 percent are products made of pulp, paper and paperboard and 11 percent are processed and preserved fish products). Costa et al. (2016) show that this kind of concentration is a common pattern of exports to China among developing economies. In this line, Figure   2 shows that industries most exposed to China's increasing demand for Chilean exports were mining of metal ores, manufacturing of refined basic metals, and manufacturing of paper products. On average the annual growth of CDE across these industries ranges from 0.55 p.p. to 0.23 p.p. In terms of economic magnitude, China's demand shock in Chile was far less pronounced than the import competition shock. Notably, the period under analysis includes the first years of the commodity boom. Although the international price of copper was highly stable until 2003, it largely increased thereafter (between 2003 and 2006 copper price doubled). Most of our results are robust to deflate export values using industry-specific international price deflators and, also, to shorten the sample period to 1996-2003. Figure 3 plots the spatial distribution of the average growth in CIC and CDE during 1996-2006 across Chilean LLMs. Darker colors correspond to LLMs that present a higher value of ∆CIC and ∆CDE and, therefore, are more exposed to each of these shocks. Note that we separate LLMs' exposition using different percentiles of the ∆CIC and ∆CDE distribution as thresholds (p5, p10, p25, p50, p75, p95). Chinese import competition was slightly more dispersed across the country than Chinese demand for exports, but concentrates in regions initially specialized in manufacturing production such as Santiago, Melipilla and La Ligua (in the Central region), Arica and Iquique (extreme 31 On average during the period under study 99.7 percent of Chilean imports from China belong to manufacturing industries.

16
North) and Talca, Valdivia and Temuco (in the Central-South). On the other hand, LLMs more exposed to increasing demand from China are concentrated in regions specialized in mining; these are located along the Central-North axis, such as Tocopilla and Calama (North), Chañaral, Copiapo and Vallenar (Central-North) and Illapel, Rancagua and Cabildo (Central). The map also shows that regions more exposed to CIC are different than those that are more exposed to CDE. In fact, the unconditional correlation between CIC and CDE across LLMs is -0.05.
Pre-shock patterns of industry specialization, as given by initial employment shares w 1992 rj , explain .66 (.85) of the variation in CIC rt (CDE rt ) while the remaining .34 (.15) is given by the variation in CIC jt (CDE jt ) across industries over time.

Figure 2 Evolution of China's trade shocks across industries
Notes. Chinese import competition (CIC) measured as the value of imports from China divided by domestic absorption (production minus net exports). Chinese demand for exports (CDE) measured as the value of exports to China divided by domestic production. CIC and CDE vary at the two-digit ISIC, rev. 4. Industry average annual change in brackets. Sources. ENIA, National Accounts, BACI.  Table 1 presents labor market statistics over time at the national-level, for all workers and separately for different sub-population groups according to skill level, age and gender.

Labor market variables
A worker is considered skilled if she has completed at least one year of university or tertiary education, and unskilled if her highest educational achievement is a high-school diploma or below.
Similarly to most countries, the structure of the Chilean labor force has moved away from primary and manufacturing sectors and towards the service sector: the share of employment in services increased 3.6 p.p. between 1996 (68.6 percent) and 2006 (72.2 percent).
As in most countries in Latin America, there is a clear pattern in Chilean labor markets: they were weaker between 1996 and 2000, and increasingly stronger between 2000 and 2006. The employment rate fell 1.9 p.p in the first period, and increased 4.6 p.p in the second. The unemployment rate changed accordingly: it climbed 4.7 p.p in the late 1990s and fell 3.1 p.p in the early 2000s. The contrast also shows up for informality (increased 1.6 p.p in the first period and fell 3 p.p in the second when using the pension definition -panel D), and for real wages (stagnant in the 1990s, and increasing in the 2000s), but not for hours of work, which decreased over the whole period.
There are gaps in favor of skilled workers in all labor market variables. These gaps widened in the late 1990s and shrunk in the early 2000s. Compared to unskilled workers, in 2006 the employment rate was 5 p.p higher, the unemployment rate 0.3 p.p lower, the informality rate (panel D) 10 p.p lower, and real wages 2.7 times higher among the group of skilled workers.
The labor market outcomes are weaker for younger workers (18-39) than for other segments. The gaps widened over the period under study in terms of employment, unemployment, and hours of work, but shrank in terms of informality (panel D) and real wages, in particular in the 2000s. Gender gaps are large in Chile: the employment rate for women is 32 p.p lower than for men, the informality rate (panel D) 7.5 p.p higher, and wages 14 percent lower. Some of these gaps have been shrinking over the period under study (e.g. employment), but some others are still growing. For instance the gap in informality between female and male workers grew from 5.1 p.p in 1996 to 7.5 p.p in

2006.
Table A4 in the Appendix shows the structure of the population of working age.
Most workers in Chile are unskilled (74.3 percent), young (51 percent) and male (61.1 percent). However, the share of these groups have been falling over time, in line with rises in education levels, population age and the increasing labor force participation of women. Unskilled individuals are over-represented in the primary and the manufacturing sectors, young workers are slightly over-represented in manufacturing, and female workers are disproportionately employed in the service sector.

20
Finally, Table A5 presents descriptive statistics for the most relevant variables at the LLM level. There is considerable variability across LLM-year combinations for all variables, which is important since this is the variation we exploit in the main regression analysis. For instance, in the case of labor informality (pension definition) the inter-quartile range reaches 5.1 p.p., and it is higher for unskilled (6 p.p.) and young (6 p.p.) workers. The last panel shows statistics for trade shocks at the LLM level. For instance, a LLM-year in the first quartile of the CIC distribution presents an exposure of 0.38; that value rises to 2.02 in the third quartile (see Figure A1 in the Appendix for an illustration of the whole distribution). Chinese demand for exports has a mean of 0.31 and a standard deviation of 0.34. A LLM-year in percentile 25 presents an exposure of 0.13, a value that increases to 0.36 in percentile 75.

IV.1.1 Traditional shift-share design
We begin this section by discussing the main results of the (relative) impact of trade shocks on local labor market outcomes. The first column in Table 2 presents OLS estimates, and columns (2) to (5) report 2SLS. The specifications in columns (3) to (5) control for a preexisting trend in the corresponding outcome variable. Columns (4) and (5) include a preexisting trend in local employment shares in manufacturing and primary sectors, and the preferred specification in column (5) also controls for a preexisting trend in female LFP. 32 The results of the first-stage regressions imply a strong predictive power of both instrumental variables ( Results reveal that the main margin of labor market adjustment to the import competition shock was labor informality. LLMs more exposed to growing CIC experienced a relative increase in the share of individuals working under informal labor arrangements. 2SLS are larger than OLS estimates, which is consistent with the existence of a negative correlation between Chile's import demand shocks and local labor demand for informal workers (that biases OLS estimates towards zero). Accounting for preexisting differences in labor informality across locations reduces the point estimates, but the coefficients remain statistically significant. Controlling for preexisting differences in female LFP across LLMs augments the magnitude of the estimated coefficients considerably, which suggests that labor informality has increased relatively more in exposed locations with lower initial levels of female LFP. 33 The point estimates in column 5 imply that a 0.41 p.p. increase in CIC (i.e. the median growth in CIC across LLMs) raises labor informality by 0.62 p.p., ceteris paribus. The relative increase is 0.68 p.p. under the contractual definition of labor informality, and 0.53 p.p. if we include non-professional self-employed as informal workers. 34 As discussed previously, heightened import competition may induce some firms 32 These controls take into account that the shift-share variables were calculated under the "incomplete shares" scenario, as defined by Borusyak et al. (2020), and that the regional balance test suggests that locations exposed to larger import shocks have on average a higher rate of female LFP. 33 Accounting for initial differences in per capita income across LLMs (or in the share of migrants or employment rate) produces a similar increase in estimated coefficient for CIC. This led us to think that there might be a positive correlation between local economic conditions and import shocks, which, in any case, generate a downward bias on our estimates.
to replace formal with informal workers (at the risk of being caught and fined), or permanent laborers-with part-time employees and apprentices, and to reduce their costs by subcontracting some production activities with other firms in the informal sector. 35 At the same time, workers displaced by import competition may be more prone to accept a (new) job in the informal sector. Also, individuals joining the labor force in exposed locations may be more likely to accept an informal labor arrangement in the context of a weaker labor market. Our estimates suggest that these mechanisms may have been present in the Chilean case.
Growing Chinese import competition did not affect the relative employment and unemployment rates in more exposed locations. These findings suggest that workers displaced by import competition probably found new jobs, likely in the informal sector.
From the results in Table 2 CIC does not seem to have reduced the relative average hourly or monthly wages significantly (these results are not robust to other specificationssee more on this later). Finally, more exposed locations experienced a relatively lower reduction in the average hours worked, which, as we discuss in the next section, may be related to the deterioration in labor conditions after the trade shock.
Our estimates suggest that China's increasing demand for Chilean exports implied a relative rise in the employment rate in more exposed LLMs. In particular, a 0.07 increase in CDE (i.e. the median growth in CDE across LLMs) augments the employment rate by 0.21 p.p., ceteris paribus. Local markets more exposed to CDE also experienced a relative increase in the average hours worked: an increase in CDE of 0.07 p.p. increases worked hours by 0.53 p.p., on average. Finally, export shocks are associated with a relative increase in the average monthly wage in exposed locations, although this effect turns out to be statistically significant only after accounting for initial differences in female LFP rates (or per capita income) across LLMs.
All of these findings prove to be highly robust across several exercises: excluding

IV.1.2 Exposure-weighted industry-level design (SSIV)
Recent contributions discuss the validity of Bartik instruments in different research designs. 36 We assess the plausibility of our assumptions conducting balance tests and falsification exercises (see section A.2). We argue that the exogeneity of industry shocks is very plausible in our context, while that of (pre-shock) local industry employment composition is a priori more difficult to justify. Borusyak et al. (2020) recently developed an econometric approach (SSIV) in which identification follows from the quasi-random assignment of shocks, while exposure shares are allowed to be endogenous. SSIV regression coefficients are obtained from an IV regression estimated at the level of the identifying variation (i.e. the industry shocks). Standard errors are asymptotically valid as long as the identification condition holds at the industry-level. 37 In this section we implement the SSIV approach. The regressions are run at the industry-level (there are 155 observations: 31 industries along 5 years). 38 Results are shown in Table 3, which has the same structure as Table 2. All specifications control for preexisting trends in (i) the sum of employment shares of exposed industries in the traded sector (primary and manufacturing) in 1992, and (ii) the corresponding dependent variable. We control for a preexisting trend in female LFP from column (3) onwards.
Columns (4) and (5) Table 3 with those in column 5 of Table 2. In line with our main estimates in Table 2, we find that regions exposed to rising import competition experienced a relative increase in the informality rate. This result holds across all specifications and definitions.
The magnitude of the estimated coefficients is somewhat larger in the SSIV than in the traditional design (30 percent higher in the case of the pension definition of informality).
The results in Table 3 reveal a small positive effect of the import shock on employment (possibly due to an "added worker effect" that we discuss later), and a relative decline in hourly and monthly wages. 39 In regard to the impact of CIC on hours worked, the sign 36 See Adão, Kolesar, and Morales (2019), Borusyak, Hull, and Jaravel (2020) and Goldsmith-Pinkham, Sorkin, and Swift(2020). 37 Shock-level aggregates are obtained by averaging the outcome and treatment variables using exposure shares as weights. Borusyak et al. (2020)'s framework is motivated by an equivalence result: the orthogonality between a shift-share instrument and an unobserved residual can be represented as the orthogonality between the underlying shocks and an unobservable shock-level. 38 The sectoral classification of the 31 industries is as follows: manufacturing (22), primary (8) and services (1). There is no variation within services because this sector is not subject to China's trade shocks (at least directly). We include the service sector because shock variables were constructed using industry shares based on 1992 total employment. Most of our results are robust to the exclusion of the service sector and also without controls for exposure shares (available upon request). 39 Estimated coefficients for the effect of CIC on wages were negative but not statistically significant in 26 and magnitude of the estimated coefficients are very similar under both frameworks.
Secondly, the effect of the export shock on employment is somewhat higher in the SSIV than in the traditional design (the coefficient increases by 22.7 percent), while the estimated coefficient for CDE on unemployment becomes statistically significant, which suggests that the export shock causes a relative reduction of unemployment in exposed locations. Notably, the magnitude of this coefficient is lower than that of the employment rate, suggesting that the export shock may drive people that were outside the labor force (inactive or studying) into employment. Estimates in Table 3 imply that regions exposed to increasing demand for exports from China experienced a relatively large increase in hours worked and wages. Wage effects are statistically significant after controlling for a preexisting trend in female LFP and, in the case of hourly wages, after considering initial differences in the share of unskilled individuals. The effect of export shocks on informality is not robust neither across specifications nor across definitions.
the traditional estimation, and the magnitude of estimated SSIV coefficients increases considerably as compared to traditional estimates (by 2.7 to 3 orders of magnitude).  (1) run by OLS and those in columns (2) to (5) run by 2SLS. All regressions control for industry and year fixed effects. To run these regressions, original variables measured at the LLM-level are converted to exposure weighted industry-level variables applying the method developed by Borusyak, Hull and Jaravel (2020). *** p<0.01, ** p<0.05, * p<0.1.

IV.1.3 Heterogeneity across sub-population groups
To shed more light on the previous findings, we estimate equation (1) for different sub-population groups defined in terms of skill level, age and gender. Table 4 presents the estimates for our preferred specification, i.e. column 5 in Table 2 which includes LLM and year FE and controls for preexisting trends in the corresponding outcome variable, exposure shares and female LFP. 40 The impact of rising import competition on labor informality was especially strong among unskilled, young and male workers. Point estimates indicate that a 0.41 p.p. rise in CIC increases the share of unskilled workers with no right to receive a pension when retired by 0.61 p.p., ceteris paribus. In the case of young workers this effect climbs to 0.74 p.p. These findings are consistent with the idea that young and unskilled workers are more prone to accept an informal arrangement than more educated and experienced workers in a context of weaker labor markets (e.g. they may have less bargaining power, less savings and/or more urgency to work). Notably, in the case of young workers, point estimates increase when we include non-professional self-employed individuals as informal workers (Panel E compared to C and D), which suggests that unskilled young individuals might be shifting from salaried jobs in the formal sector towards self-employment.
Increasing CIC generated a small relative decline in the employment rate of old-age workers in more exposed LLMs, and an even smaller relative increase in the unemployment rate, which suggests that a fraction of the old-age workers who lost their jobs to import competition left the labor market. On the contrary, rising CIC is associated with a moderate relative increase in the employment rate of young, skilled individuals (i.e. those with at least one year of tertiary education). These results are consistent with the "added worker effect" mechanism: when labor conditions for some members of the household (e.g. old-age displaced workers) worsen, secondary workers (e.g. children studying at university or tertiary institutes) may enter the labor market to ameliorate the negative shock. 41 Our long-term estimates, conducted using census data, are also in line with this hypothesis (see section IV.2.2). Finally, Table 4 also reveals that locations with greater exposure to CIC experienced a relatively lower reduction in the average hours worked by all groups, and a relative decline in hourly and monthly wages among unskilled and young workers.
The gains from China's increasing demand for exports in terms of employment and wages were also asymmetric across sub-population groups. Gains in relative employment were limited to young workers and males (both skilled and unskilled). Specifically, a 0.07 p.p. rise in CDE augments the relative employment rate of workers aged 18 to 39 by 0.36 p.p., ceteris paribus. In contrast, the increase in monthly wages in regions exposed to growing CDE took place especially among old-age workers: a 0.07 p.p. rise 40 Most of these results are strongly robust to the exclusion of preexisting trends, and also to the inclusion of pre-trends in many other control variables. We discuss some of these exercises in the robustness section. 41 Recent evidence for Argentina suggests that job losses by the male household head lead to a significant increase in the labor force participation of female partners and sons (Ciaschi (2020)).

29
in CDE increases monthly wages by 0.92 p.p. on average, and 1.36 p.p. for old-age workers. 42 Relatedly, more exposed locations benefited from a larger decline in labor informality of old-age workers and males, but only when considering the non-professional self-employed as informal (Panel E), which suggests that some of these workers might have moved from self-employment towards salaried jobs in the formal sector in response to new job opportunities created by the export shocks. Estimates in the next section suggest that this transition may have occurred primarily in the service sector.
42 In line with the general result for all workers, these effects turn out to be statistically significant after controlling for a preexisting trend in female LFP (or in per capita income).

IV.1.4 Heterogeneity across sub-population groups and economic sectors
This section extends the analysis by exploring heterogeneities across sectors -primary, manufacturing and services. Given that we lose statistical power when looking at specific groups within each sector at the local-level, these estimates should be read with some caution. Tables A7, A8 and A9 report the results for each of these sectors. Since the sample is restricted to employed individuals, we include regressions for the absolute (log) level of employment in each sector, instead of regressions for employment and unemployment rates. Table A8 shows that increasing CIC caused a relative decline in the number of workers employed in manufacturing. Job losses were especially large for unskilled and young individuals, and somewhat larger for females than males. 43 Also, more exposed locations experienced a relative increase in labor informality in manufacturing, mainly among unskilled and old-age workers. Notably, in the contractual definition of informality, this effect is driven mainly by old-age and female workers, which might be connected with the subcontracting/outsourcing hypothesis. 44 We find a relative increase in labor informality in the primary (mainly for skilled, old-age and male workers) and service sectors (especially for skilled, young and male workers).
Locations exposed to increased import competition suffered a relative decline in the average wage of old-age workers employed in manufacturing (which is consistent with the relative increase in labor informality for this group). By contrast, they also experienced a relative increment in the average manufacturing wage of young and female workers, which may be the result of a compositional change. 45 Finally, we find a relative wage decline in the service sector (especially for young workers), which could be related to at least two hypothesis: the "added worker effect" and reallocation dynamics from manufacturing towards services. 46 On the other hand, LLMs exposed to increased demand for exports experienced a relative rise in employment among young workers in the primary sector, in manufacturing employment (with a similar magnitude across different groups but statistically significant for males only) and in the number of old-age workers employed in services (Tables A7). Estimates for labor informality suggest that (new) job opportunities for young individuals in the primary sector took the form of temporary contracts (presumably short-term or apprentices) instead of permanent salaried jobs with social security contributions. On the contrary, formal service jobs may have been taken mainly by unskilled, old-age and male 43 One reason is that women have a greater employment participation in exposed than unexposed manufacturing industries (e.g. wearing apparel, textiles, leather products). 44 For instance, a large textile company might focus on management, design and commerce, while outsourcing to small and medium plants some stages of the production process (e.g. cutting, sewing, embroidery). As we previously argued, this is common in Chile (and even more so in manufacturing) and the intensity of outsourcing increased over the period under study. 45 I.e. manufacturing workers that survived import competition might have on average higher skills and/or belong to more productive plants than displaced individuals. 46 Papers that track workers' trajectories over time have found this kind of reallocation pattern in response to import competition or trade liberalization episodes (Utar (2018), Dix-Carneiro and Kovak (2019)).
workers that were previously self-employed. Finally, our estimates suggest that areas with greater exposure benefited from relative wage gains in the three sectors.

IV.1.5 Local employment structure
This section aims to further understand the short-term mechanisms of labor reallocation within LLMs. We start in Table A10 by exploring the impact of China's supply and demand shocks on the local employment structure across sectors. 47 The first column corresponds to OLS, while columns (2) to (4)  which reinforces the idea that labor reallocation may go from manufacturing towards services.
To assess whether the relative employment contraction occurs mainly in exposed industries, we run separate regressions for exposed and unexposed industries within each sector. 48 The service sector was disaggregated into construction, trade (including transport and storage) and the rest of services. We present these estimates in Table   A11. In this table columns refer to dependent variables (employment share in each group) and panels correspond to different specifications. As expected, the results confirm that manufacturing industries more exposed to increasing import competition suffer a relative employment contraction. This result is highly robust across specifications. Panel A shows that these locations experienced a relative increase in the fraction of workers employed in services such as construction and trade. Controlling for initial differences in female LFP across LLMs increases the magnitude of the estimated coefficient for the trade sub-sector (in line with the fact that females are more likely to work in this sector).
In contrast, controlling for a preexisting trend in the corresponding outcome variable, but not in female LFP, turns the coefficient for the construction sector (where most workers are men) statistically significant. Overall, in the preferred specification (Panel D), the coefficients are statistically indistinguishable from zero, with the only exception of the negative coefficient for the employment share in exposed manufacturing industries.
We carry out separate regressions by gender (i.e. sectoral employment shares are calculated separately for males and females). This exercise is relevant considering that the period under study witnessed a major increase in the rate of female employment (from 40 percent in 1996 to 47.1 percent in 2006). 49 Results in Table A12 suggest that the employment structure across genders was differently affected by the imports shock. Although increasing CIC caused a relative reduction in the employment share in manufacturing for both genders, its effects on the primary and service sectors differ. In regions more exposed to CIC, the increase in the relative employment share of services was much larger for males than females. The results suggest that, unlike men, some of the displaced women reallocated into the primary sector as well.
Finally, we estimate the impact of trade shocks on the employment shares by labor relationship: waged employees, self-employed and employers. We present these estimates for all workers and separately by gender in Table A13. Even columns correspond to the preferred specification and uneven columns exclude the preexisting trend for female LFP.
The results suggest that regions more exposed to import competition experienced a relative contraction in the fraction of waged employees, and that this effect concentrates mainly on males. Estimates in Panels B and C suggest that men losing their jobs as wage earners became self-employed and even employers. In this sense, there would be some potential for creative destruction in the face of import competition. 50 On the export side, regions more exposed to increasing demand from China experienced a relative increment in the fraction of employers. This effect concentrates on males also. Export shocks may generate business opportunities (both in the traded and non-traded sectors) that entail the creation of new firms (presumably small at the beginning) but do not substantially modify the local employment structure, at least in the short-run. Some of these companies may thrive and become job creators as they grow. In fact, in what follows, we show that the export shock has increased the employment share of salaried jobs over the long-run.
49 Also, Table A4 highlights that females are less under represented in the service sector than in the manufacturing and primary sectors. On average, during 1996-2006 females represented 42.8 percent of employment in services, 28.5 percent in manufacturing and 15.6 percent in the primary sector. The three sectors saw an increase in female participation between 1996 and 2006 (3.4 p.p., 2.6 p.p. and 6.3 p.p., respectively). 50 Additionally, import competition may increase within firm productivity through skill-upgrading and rising innovation in surviving firms, and also because of employment reallocation towards more technologically advanced firms (Utar and Torrez Diaz (2013), Bloom et al. (2015)).

IV.2 Long-term effects
This section explores the long-term impact of China's growing import competition and demand for exports on migration, employment outcomes, and schooling, by exploiting census data for 1992, 2002 and 2012. We study changes over time : 1992-2002 and 2002-2012. We present separate estimates for each sub-period and for both periods together (stacked changes). The first stage has much more statistical power in the first period than in the second (Table A14) 1987-1997 and 1997-2007. 54 Note that the second period almost coincides with the one we use in the short-term analysis (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006). All specifications control for the employment share in the primary and manufacturing sectors, while even columns include a preexisting trend in female LFP. Regressions in stacked changes control also for a period dummy and the exposure shares in the second period (as recommended by Borusyak et al. (2020)). Table A15 suggest that China's import and export shocks did not affect migration rates differently across LLMs. This is reassuring as it indicates that migration patterns across LLMs should not contaminate our short-term estimates. Table A16 extends the analysis with census data to several labor and education outcomes. Additionally, in Table A17 we explore the existence of heterogeneous effects across different sub-population groups. The results suggest that LLMs exposed to increasing import competition experienced a relatively small rise in employment, along 51 This is in line with the robustness exercise that extends the sample period sequentially until 2015 (see Table A20 discussed in Section IV.3.2). 52 According to the International Federation of Robotics (IFR), Chile started to import robots for manufacturing production in 2006 (6 units). The operational stock of robots grew steadily in the following years reaching 97 robots in 2016. 53 A migrant is an individual that moved from one LLM to another within the periods 1987-1992, 1997-2002 or 2007-2012; it is calculated using a retrospective question about the municipality of residence five years previous. 54 International migrants are excluded from this definition because they were living in a different country in 1987, 1997 or 2007. However, they represent less than 10 percent of Chilean migrants in all these years. In line with our short-term estimates, we find that the imports shock caused a relative increase in the employment rate of skilled and young individuals. These locations also experienced a relatively minor rise in the share of working age individuals attending higher education. 56 We interpret these findings as partial evidence in favor of the "added worker effect" hypothesis. The labor supply of young and skilled individuals increases in the face of worsening labor conditions of other household members (e.g. old-age workers displaced by import competition) so secondary workers (e.g. young individuals studying at university or tertiary institutes) enter the labor market to ameliorate the negative shock. 57

IV.2.2 Employment and education
Regarding the demand shocks, we find that locations exposed to China's increasing demand for exports experienced a relative increase in employment, which holds in both periods, and in the share of salaried workers (which concentrates mainly in 2002-2012). Table A17 suggest that the export shock caused employment gains and shifts towards salaried jobs among all sub-population groups, which were somewhat stronger for young, skilled, male individuals. The magnitude of the estimated coefficients for the young more than doubles that for the old-age group. Similarly, while all groups in exposed locations exhibited a relative decline in unemployment, this effect was much stronger for unskilled individuals. The fact that the magnitude of the estimated coefficients for the impact of growing CDE on unemployment are considerably lower than that for the employment rate (especially in the case of skilled, young and males), indicates that new jobs were taken by individuals that were outside of the labor force (i.e. inactive or studying).

Results in
In this line, local markets exposed to larger export shocks experienced a relatively lower increase in the fraction of individuals studying at a tertiary institution or a university. 58 Effects are strong and even larger if we concentrate on young people (ages 18-30). These effects take place in both periods but are somewhat higher in 1992-2002.
Our hypothesis is that export shocks have created new jobs that, in many cases, were taken by individuals that could otherwise be studying at a tertiary institute or university.
In other words, export opportunities created jobs that raised the opportunity cost of schooling (Atkin (2016)). This evidence is consistent with export shocks biased towards 55 Overall, short-term estimates suggest that the impact of CIC on relative employment and unemployment rates were statistically indistinguishable from zero, which we argue was driven mainly by labor reallocation towards informal jobs. 56 This effect is much stronger for individuals aged 18-30 but, in this case, it concentrates in the second period. 57 Greenland and Lopresti (2016) document large increments in U.S. high school graduation rates in LLMs most affected by import competition. While Utar (2018) finds that Danish workers displaced by import competition tried to return to complete education mostly after moving to the service sector. Importantly, workers enrolled in school to increase job prospects receive an education allowance from unemployment insurance. 58 The fraction of the population under age 18-65 (18-30) enrolled in tertiary education increased from 12.9 (15.7) percent in 1992 to 23.4 (29.3) in 2002 and 32.8 (45.1) percent in 2012. (2017), Li (2018)). Overall, these findings deserve a more detailed study that is beyond the scope of this paper.

V Concluding Remarks
This paper studies the causal effect of trade shocks on local labor markets, exploiting spatial and time variations in trade exposure arising from initial differences in industry specialization across Chilean geographic locations and the evolution of shocks across industries. To account for the endogenous nature of trade we instrument Chinese import competition and demand for exports with China's trade patterns in the rest of the world.
This strategy aims to capture supply and demand-driven shocks inherent to China's economic forces and institutions that allowed it to gain global market share within specific industries over time.
Our findings suggest that the local employment structure responds to trade shocks.
Notably, locations with greater exposure to increasing import competition suffered a relative deterioration of labor market conditions, which mainly affected young and unskilled individuals. These groups are the most affected by job losses in manufacturing and, more generally, by a weaker labor market, which leads to relative wage declines and increases in labor informality. A small proportion of unskilled older workers who lost their jobs to import competition left the workforce, while those remaining employed in manufacturing suffered increases in labor informality and relative wage losses. Our evidence is consistent with the "added worker effect" hypothesis by which secondary workers (e.g. children studying at university or a tertiary institute) enter the labor market in order to ameliorate the worsening labor conditions of other members of the household (e.g. old-age workers).
We find that the informal sector acts as a buffer in response to import shocks.
Increasing labor informality seems to be a strategy to cope with rising import competition.
Informality allows firms to reduce taxes and other labor costs (at the risk of being caught and forced to pay a fine). Firms may also replace permanent with part-time employees and apprentices and subcontract some production activities with other plants in the informal sector. On the other hand, workers may be more likely to accept an informal labor arrangement in the context of a weaker labor market. Of course, this finding is not enough to claim optimality for informal labor arrangements, which brings many economic costs and inefficiencies (Dix-Carneiro, Goldberg, Meghir, and Ulyssea (2019), Ulyssea (2020)).
China's global trade expansion, in turn, caused a relative employment increment in local labor markets specialized in the production and processing of primary products.
Employment gains mainly benefited young and male workers. These individuals took advantage of employment opportunities in the primary sector, mainly in the form of temporary contracts (presumably, short-term or apprentices). Old-age workers harnessed formal jobs in the service sector, improving their relative wages. Finally, these locations 37 experienced a relatively lower increase in tertiary education attendance than less-exposed areas, which could delay local development through its effects on human capital formation.
We believe that this topic deserves further attention and represents an interesting avenue for future research.
Our results point out considerable heterogeneity across workers in the spatial distribution of the costs and benefits of trade. These findings are especially relevant for developing countries where exports exhibit a high degree of concentration in a few number of primary products, where informality and tax evasion are key policy issues, and where an important fraction of employment concentrates in low-competitive industries characterized by high informality.

A Appendix
A.1 Data details

Labor market outcomes from household surveys
The employment rate is the proportion of persons employed in relation to the population of working age. Employed individuals are those that have worked during the last week and those who declared that they did not work but actually have a job. The unemployment rate is the fraction of individuals who did not work but have been looking for a job in the last two months in relation to active population. Active population includes both the employed and unemployed individuals.

Migration and employment outcomes from censuses data
The census contains information on individual's current and 5-year previous regions of residence, which allows to construct a measure of migration for 1987-1992, 1997-2002 and 2007-2012, defined  Also, no census includes information on labor incomes, social security benefits or contractual situation, so in this case we can not explore earnings or labor informality.

Trade variables
Information on bilateral trade flows is obtained from the CEPII-BACI database. These data contain annual information on bilateral trade values and quantities for more than 59 The number of hours of work is asked in terms of weeks so we multiply this value by 4.28 to express it in monthly terms. 60 This information is included in the question related to employment status. We compute imports and exports flows at the two-digit industry-year level (according to the ISIC Rev. 4). To construct the measure of Chinese import competition (CIC) we merge the ENIA and BACI datasets. M jt and X jt in equation (2) are obtained by aggregating product-level information from BACI data, while Q jt is measured by adding up plant-level information from INE-ENIA. We convert the information at the six-digit level of the HS to the two-digit level of the ISIC using correspondence tables from United Nations Statistics Division.
Similarly, we merge BACI, ENIA and sectoral GDP datasets to construct a measure of Chinese demand for exports (CDE) at the two-digit industry-year level.

A.2 Balance and pre-trend tests
In the first set of balance tests, we regress potential industry-level confounders directly on the shocks. This test has rarely been used in Bartik-type designs (Borusyak et al. (2020)). The idea is to test if the China's shocks do not predict variables related to employment structure and technology, so we can argue that these shocks are as-good-as-randomly assigned across industries. We obtain industry-level production variables from the 1996 Encuesta Nacional Industrial Anual-ENIA (INE). We compute the log average industry wage, the fraction of production workers, capital intensity (i.e. the stock value of capital divided by the number of workers) and the fraction of machinery in total investment. All these variables are calculated at the 2-digit ISIC Rev. 4 level, for manufacturing industries only. We regress each of these variables on the shocks, which are normalized to have a unit variance. Table A1 shows that industry variables are not significantly correlated with industry-level shocks.
In the second set of balance tests, we regress potential region-level confounders on the shift-share instruments (again, normalized to have a unit variance). These variables are the log average per capita income, the fraction of individuals with secondary education (or below) in population of working age, the share of workers under age 40-65 in population of working age, the fraction of individuals under age 18-30 assisting to tertiary education, the female employment rate, the percentage of migrants in population of working age and the fraction of Chile's population living in each LLM. The idea is to test if the Bartik-shock instruments do not predict LLM outcomes related to local economic and demographic structures, so we can be more confident that the shocks are as-good-as-randomly assigned across locations. These regressions control for the 1992 share of manufacturing and primary employment which were used to construct the shock variables. Estimates in Table   A2 suggest that there is no statistically significant correlation between these variables and the shift-share instruments, with the only exception of the female employment rate.
In particular, locations exposed to larger import shocks tend to have a higher female employment rate in 1992. 62 Moreover, given that Chile experienced a major increase in female LFP during the period under study, this unbalance might led to potential confounding effects if regions were subject to different labor supply dynamics. In this context, it is important to control for initial differences (pre-trends) in female LFP across LLMs. Additionally, we perform separate regressions by gender (section IV.1.3).
Finally, we conduct a pre-trend analysis at the LLM-level to test if past changes in labor market outcomes are correlated with the future changes in China's supply and demand shocks. We regress the average annual change in labor market outcomes during 1990-1994 on the average annual change in CIC andCDE during 1996-2006. We run these regressions by OLS and 2SLS. Results are very similar if we regress these variables directly on the instruments. Results in Table A3 suggest that there is no significant relationship between past changes in local employment, unemployment and informality rates, and future changes in location-specific trade shocks. The only exception is the log average hourly (or monthly) wage. Locations that were subsequently exposed to larger import supply shocks had a greater relative increase in labor incomes in the first years of the 90s. We conduct a robustness exercise (i.e. controlling for initial differences in labor incomes in the main regressions) to address potential endogeneity concerns related to this unbalance. All results remain virtually unchanged.

A.3 Robustness exercises
In this section we present and discuss the results of several robustness exercises carried out for the main regression analysis (which follows the traditional shift-share research design conducted at the LLM-level).
The first exercise aims to control for a potential bias in estimated coefficients arising from outliers in the shock variables. We order LLMs according to their exposure to Chinese import competition (measured by the average growth in CIC over time) and, separately, using their exposure to Chinese demand for exports. Then, we exclude the 5% extreme values in each of these distributions and run the baseline regressions. This exercise excludes 12 LLMs (the three with the highest and lowest exposures to CIC or CDE) and therefore runs the regressions using 49 instead of 61 LLMs. 63 We present these estimates in Table   62 Results are very similar if we use other years of the period under study (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) exercise suggests that increasing Chinese import competition caused a relative decline in the average hourly and monthly wages of workers living in more exposed locations. In the preferred specification, a 0.13 p.p. increase in CIC reduces the average hourly and monthly wages by about 1 percent, ceteris paribus.
The second robustness exercise consists on including preexisting trends of variables related to the economic and demographic structures of LLMs that might be potentially correlated with future Chinese import competition and demand for exports. These controls partially address a bias in estimated coefficients if they correlate with unobserved shocks affecting both trade shocks and labor market outcomes simultaneously. As before, preexisting trends are constructed as the value of the corresponding variable in 1992 interacted with year fixed effects. We include pre-trends for the average household per capita income (in logs), the fraction of old-age individuals in population of working age, the fraction of unskilled individuals in population of working age, the share of migrants in population of working age, and all of these variables together. Table A19 presents the estimates. Results for the effect of export shocks on the employment rate are robust to controlling for initial differences across locations in all of these variables. Including all preexisting trends together halves the magnitude of estimated coefficient (and the p-value is 1.03). Notably, running the regression with full controls for the employment rate of young workers returns an estimated coefficient of 0.035 with a p-value equal to 2.19. Estimates for the effect of import shocks on labor informality are highly robust to the inclusion of all of these control variables. It is worth mentioning that the economic magnitude of the estimates increases when we control for a preexisting trend in income or in the share of migrants, which might be explained by a positive correlation between local economic conditions and future CIC. In the same spirit, when controlling for these pre-trends, import competition causes a relative reduction in the average monthly wage of greater exposure to Chinese demand for exports are Chañaral (the average growth in CDE is 0.59 p.p.), Cabildo (0.43 p.p.) and Calama (0.38 p.p.).
workers in more exposed LLMs. On the other side, the positive effect of the export shock on relative wages is highly robust to the inclusion of all of these pre-trends.
In the following exercise, we run the main regressions extending the sample period sequentially (up to 2015). Local labor market outcomes were obtained from the Chilean In the next exercise we use different groups of countries to construct the instrumental variables (IVs). These estimates are shown in Table A21. Remember that we compute the IVs using the average participation of China in the imports and exports of all countries in the world, except Chile, because the idea is to capture China's supply and demand-driven shocks (inherent to Chinese economic forces) that allowed it to gain global market share within specific industries over time. We can calculate the average China's industry shares using different groups of countries as well. This exercise might alleviate endogeneity concerns related to a potential correlation between common shocks at the industry-level between Chile and different groups of economies. Using The World Bank's definition, we compute the IVs for middle-income economies (excluding Latin America), high-income and Latin American countries. Interestingly, our estimates exhibit robustness to the use of different sets of countries to compute the IVs. Estimated coefficients for the impact of import competition on labor informality are slightly higher than in the baseline scenario, while those for the effect of export shocks on employment rate are somewhat smaller.
Overall, we believe that using all countries in the world is the preferred option because it implies a lower probability of industry-level common shocks between Chile and the rest of the world. 65 Additionally, we show that our results are robust to changing the unit of analysis.
Instead of LLMs, we calculate all relevant variables of equation 7 at the municipality-level (also known as comunas). Figure A2 in the Appendix presents a map showing the variation in CIC and CDE across municipalities. One advantage of running these regressions is that we count on a higher number of observations and therefore gain some statistical power in this direction. However, it might come at the cost of a larger measurement error because municipalities are on average smaller than LLMs and, in many cases, very spatially and economically integrated implying a high degree of labor mobility across neighboring municipalities. 66 In this case, estimated coefficients may present an attenuation bias. We cluster standard errors at the level of LLMs. Table A22 presents the results. Remarkably, most of our estimates remain robust to this exercise. Generally, estimated coefficients present a lower magnitude than the main estimates, which is consistent with the hypothesis of attenuation bias.
In the following exercise, we estimate an "unweighted-version" of the main regressions of the paper. Table A23 reports the unweighted estimates. In the preferred model, we weight each observation by the LLM share of population of working age in 1992 (as in Autor et al. (2013) and Costa et al. (2016)). This empirical strategy provides average treatment effects (ATE) that are weighted by workers instead of LLMs. While in this exercise we use no weights in the regressions, so that all LLMs weigh the same regardless of the size of their workforce. Results in Table A23 show that the positive employment effects of export shocks are robust not to using working-age population weights in the main regressions. On the other side, estimated coefficients for the effect of import shocks on labor informality are no longer statistically significant. Point estimates are somewhat smaller than in the baseline (weighted) regressions and standard errors rise by more than two orders of magnitude. This result might be explained by the fact that Chinese import competition occurs mainly in densely populated areas with presence of industrial clusters specialized in the production of manufactures (e.g. big cities and their surroundings areas, like Santiago and Arica) so estimates are weaker if we do not take into account differences in workforce size across LLMs. On the other side, estimates for the impact of CDE on labor market outcomes do not exhibit major changes.
Finally, we run the main regressions of the paper using a balanced sample of LLMs (these estimates are shown in Table A24). The exercise restricts the sample to 57 LLMs 65 Common shocks are perhaps more likely when using Latin American countries, given they have closer geographical, economic, commercial and cultural relations with Chile. 66 On average, a LLM is formed by 4.7 municipalities.
48 that are observed in each of the five years (285 observations instead of 296). Estimated coefficients present very little variation compared to the preferred estimates in Table 2.

Figure A2 Spatial distribution of CIC and CDE at the municipality-level
Notes. Chinese import competition-left (Chinese demand for exports-right) across municipalities measured as the interaction of local employment share of each industry in 1992 and the evolution of CIC (CDE) across industries over time. This figure plots the average change in each of these variables across Chilean municipalities over time. Sources. 1992 Chilean Census, ENIA, National Accounts, BACI.    (2) and (3) these variables are instrumented using the average annual change in China's world supply and demand shocks. All regressions weighted by LLM's share of Chile's population of working age in 1992. Robust standard errors clustered by region.                 (5) and (6) control also for a period dummy and for the exposure shares in 2002. *** p<0.01, ** p<0.05, * p<0.1.       1996-2006 1996-2009 1996-2011 1996-2013 1996-2015 (1)