Industry competitiveness and migration flows Elena Gentili January, 2018 Draft version Abstract This paper investigates how the competitive structure of an industry influences different types of migration inflows. The theoretical model combines an industrial monopolistic competitive structure with labor market frictions and workers who are heterogeneous in their reservation wages. If on average migrant workers have lower reservation wages than non-migrant workers and there are no labor shortages, more competitive industries tend to disproportionally hire more migrant workers to decrease their marginal costs. The empirical analysis is conducted at industry level with US manufacturing industry data. Industry price cost margins are instrumented through a structural change in trade policy that differentially impacted industry competitiveness: the entry of China in the WTO. Results confirm the theoretical predictions, with more competitive industries attracting more international migrants, either high or low educated, and low educated internal migrants. Keywords: Migration, Industry competitiveness. JEL codes: F16, F22, J23, J60, L13. Corresponding author: Elena Gentili, Institute of Economics, via Buffi 13, 6904 Lugano, Switzerland (elena.gentili@usi.ch). Institute of Economics (IdEP), Università della Svizzera Italiana (USI), Switzerland.
1 Introduction This paper analyzes the relationship between the competitive structure of an industry and the employment of migrant workers. Particularly, while a large fraction of the recent migration literature focuses on the impact of immigration on several economic outcomes (Dustmann et al., 2017; Ottaviano and Peri, 2012; Borjas and Katz, 2007; Borjas, 2003; Card, 2001 among the others), the aim of this paper is to understand whether the degree of competitiveness of an industry influences the demand for migrant workers. The literature about the role of labor demand in determining migration inflows can be traced back to Piore (1979). However, while descriptively examining how labor demand can influence migration inflows, this literature does not attempt to understand the relationship between the market conditions faced by the firms and the demand for migrant workers. Conversely, this paper attempts to quantify this relationship providing both a theoretical model and some empirical evidence about the impact of industry competition on migration inflows. The theoretical model combines the model of monopolistic competition developed by Melitz and Ottaviano (2008) with classical search models as discussed in Eckstein and Van den Berg (2007). Particularly, I consider an economy where many firms produce different varieties of the same good, workers are heterogeneous in their reservation wages and there are search frictions in the labor market. The optimal number of firms in the market, as well as firm profits and markups are endogenously determined by the distribution of reservation wages. The aim of the model is to show that in the presence of workers that are homogeneous in their productivity level and heterogeneous in their reservation wages, firms that work in lower markup industries have a stronger incentive to depress their marginal costs and, in turn, the wages of workers. On the other hand, workers may accept wage offers below their marginal productivity because of labor market frictions. Assuming that immigrant workers have lower reservation wages, stronger pressures for lower marginal costs induce firms to disproportionally hire more immigrant workers. To empirically test the theoretical model s predictions, I follow Aghion et al. (2005) in measuring the extent of monopolistic competition through a Lerner index at industry level. However, since the composition and the extent of migration flows may influence the competitiveness of industries, reverse causality may bias the OLS estimates. In particular, I need an instrumental variable which is correlated with the competitiveness of industries but unrelated with migration flows. For this reason, I instrument the Lerner index with a structural break in trade policy that changed the competitive pressures in some industries but not in others: the official entry of China in the WTO agreements. Prior to entrance in the WTO, China was already granted a Normal Trade Relationship (NTR) status to trade with the US. This NTR status entailed the application of the Most Favored 1
Nation (MFN) tariffs, which were substantially more favorable with respect to the tariffs applied to non-ntr countries. However, this NTR status had to be renewed every year and especially in the early 90s the Congress attempted to withdraw this preferential treatment several times. Thus, the entrance of China in the WTO did not change the actual tariffs applied, but ended the uncertainty on their implementation. Pierce and Schott (2016) implement a diff-in-diff approach to exert the impact of this trade policy on US employment in the manufacturing sector. Instead, I am using their diff-in-diff interaction term as an instrument for industry competitiveness. Even though increasing demand for cheap labor inputs is just a way that firms have to overcome competitive pressures (firms can also offshore or outsource, see for example Ottaviano et al. (2013) and Grossman and Rossi-Hansberg (2012) for offshoring and Antras and Helpman (2004) and Grossman and Helpman (2002) for outsourcing), I find that a one standard deviation increase in the price cost margins (i.e., a one standard deviation decrease in competitiveness) decreases the employment of newly arrived foreign-born international migrants of 45% the mean of the dependent variable. This result holds for both low and highly educated migrants, even though it is stronger among the low educated. Considering different types of migration, i.e. adding foreign-born workers that migrated within the US to international migrants or considering both native and foreign internal migration, an increase in competition increases the employment of foreign-born workers, but decreases the employment of highly educated workers. All the results hold controlling for the technological content of industries. Notice that the results on highly educated internal migrants can be better understood recalling that the model assumes homogenous productivity of workers. Considering workers with different skills as different segments of the labor market, the model still applies to each segment. Less competitive industries may need highly educated migrant workers to expand production. In this case, the internal flow of highly educated migrants is triggered by skill shortages on the labor market rather than considerations about marginal costs. This paper adds to several strands of the literature. First of all, it adds to the literature on migration, providing evidence of a possible channel through which pull factors determine migration flows. Second, it adds to the literature on industrial organization, providing evidence about the effects that different degrees of competition may have on employment decisions. Third, it indirectly adds to the trade literature and how changes in trade policy can impact the labor market structure of the countries involved. The remaining of the paper is structured as follows. The following section introduces the theoretical model. Then, the third and the fourth sections describe the data and explain in detail the empirical strategy adopted. Finally, Section 5 presents the results and Section 6 concludes. 2
2 Theoretical model The main goal of this section is to uncover the relationship between industry price cost margins, i.e. industry markups, and variations in the average wages paid by the firms. In particular, I am going to show that firms in industries with lower markups are upward constrained in setting wages. The mechanism works as follows. The model encompasses a labor market with search frictions, workers who are heterogeneous in their value of home production and a monopolistic competitive structure of the final good markets. Final good markets represent different industries and each final good market comprises different varieties of the final good. Given the monopolistic competitive setting, the maximum price each firm can charge to stay on the market depends on some structural characteristics, such as barriers to entry, bounds of the average wages paid by the firms, and degree of differentiation of good varieties within each market. These structural characteristics are the same for all the firms within the market. The heterogeneity in workers reservation wages induces wage dispersion in each market, and firms pay different average wages. Overall, in addition to the wage dispertion observed in each market, firms in markets with higher threshold prices can afford paying higher wages to their workers. Thus, if immigrant workers have lower values of home production, industries with lower markups will disproportionately hire more immigrant workers. The following subsection describes the equilibrium on the labor market, while the second subsection describes the equilibrium in the final good market. 2.1 Labor market Supply side Each individual participates in the labor market and provides either 1 unit of labor, if employed, or 0 unit of labor, if unemployed. For simplicity, I assume that workers are homogeneous in their productivity and each worker produces only one unit of output. Employed workers earn a wage ω c, while unemployed workers earn their value of home production v c. Accepted wages are distributed according to F (w c ), with w c [ w, while home production values across unemployed workers are w] distributed according to G(v c ) with v c [ v, v]. Notice that the distributions of accpeted wages and home production values may overlap. In the labor market there are search frictions. In every period, unemployed workers receive wage offers at arrival rate λ, and a fraction δ of employed workers dies and exits the labor market. Moreover, wage offers are distributed according to a generic distribution H(w), with w [ w, w]. If the worker accepts the wage offer, then she works with the same firm until she exits the market. If the worker rejects the wage offer, she keeps on searching for wage offers on the market. 3
To model workers search behavior, I refer to the baseline model of search frictions by Eckstein and Van den Berg (2007) with no search on the job. Each worker compares the present value of being employed, V e, with the present value of being unemployed, V u, and stops searching when the first value outweighs the second. Thus, the worker is indifferent between accepting or refusing the wage offer when: { } ρv u = v c + λ w max[0, V e (w) V u ]dh(w) δ w where ρ is a discount factor, ρv u is the present value of unemployment, and the RHS is the present value of search. The worker stops searching when she finds an offer w which is larger than her reservation wage w. Whenever she rejects the offer she obtains the value of home production, v c. Since the expected value of being employed is V e (w) = 0 e (ρ+δ)t wdt = w/(ρ + δ), the value of search for the worker can be stated as: { } w = v c + λ w [w w ]dh(w) (2) ρ + δ w Demand side The hiring process of the firm can be modeled as a first price sealed bid auction with secret reservation price (Elyakime et al., 1994). (1) In making a wage offer, the firm does not know the reservation wage of the unemployed worker w, nor the potential wage offers of the other firms. The value of the worker for the firm is given by the value of its marginal product. However, since the marginal product of each worker is one, the value of the worker for the firm is given by the price charged for the additional unit produced, i.e. p(ŵ). As will become clearer in the next section, the price set by the firm is a function of the average wages paid and of the maximum price it can charge to stay on the market. Thus, the heterogeneity of workers reservation wages translates in price dispersion within each market, and the structural differences across markets further reinforce this price dispersion. Since wages are distributed according to a cumulative density function F (w), I index with F(ŵ) the distribution of average wages across firms, with ŵ [ŵ m, ŵ M ]. Similarly, the distribution of reservation wages across unemployed workers is a monotonic transformation of the distribution of values for home production G(v). The distributions F(ŵ) and G(w ) are common knowledge. Let s(p(ŵ)) be the optimal strategy of the firm as a function of the worker s value of marginal product. Since firms do not know the potential offers of the other firms, I follow the literature in considering symmetric equilibria and assume that all the firms implement the same optimal strategy. Indexing with j a random firm in the market, the probability that the worker accepts the offer of firm j is the 4
probability of p(ŵ) being the highest offer made to the worker times the probability that s(p(ŵ)) exceeds the reservation wage w of the worker. 1 If s(.) is differentiable and strictly monotone, the expected payoff from hiring an additional worker can be written as: Π(p(ŵ), s(p(ŵ))) = [p(ŵ) s(p(ŵ))]f N 1 (p(ŵ))g(s(p(ŵ))) (3) From this payoff function, following Elyakime et al. (1994), it is possible to show that the optimal wage offer for the firm is: s (p(ŵ)) = p(ŵ) p(ŵ) 0 F N 1 (x)g(s (x))dx F N 1 (p(ŵ))g(s (p(ŵ))) (4) The proof is left to Appendix A. Thus, the optimal wage offer is a mark down strategy, which consists in offering the value of the worker s marginal product minus a mark down. Two remarks are worth mentioning here. The first one is that firms always find it optimal to offer a wage that is lower than the value of worker s marginal product. Thus, firms are always willing to hire an additional worker. Second, this equilibrium strategy is implicitely defined, because firms do not know the reservation wages of the workers. 2 To sum up, Equations (2) and (4) describe the optimal strategies for the workers and the firms. The wage offer distribution H(w) can be obtained from this latter optimality condition and can be defined in the support of the average wage distribution [ŵ m, ŵ M ]. 2.2 Final good market Preferences and output demand As in Melitz and Ottaviano (2008), consider an industry with many firms producing differentiated varieties i Ω of the same good. For simplicity, I focus on a single industry, even though the following theoretical framework may be extended to encompass all the industries in the economy. Consumers share the same utility function over preferences for varieties and leisure, that takes the form: U c = l c + α qi c di 1 i Ω 2 γ (qi c ) 2 di 1 ( 2 i Ω 2 η qi di) c (5) i Ω where l c is the consumption of leisure, the numeraire good, and q i is the consumption of other varieties. The parameters α and η are demand shifters with respect to leisure, while the parameter γ indexes the degree of differentiation between varieties. Normalizing the maximum working time 1 For the equilibrium, it is not necessary having other firms making offers to the workers at the same point in time. The possibility that they do so is sufficient to obtain this result. 2 See Elyakime et al. (1994) for further details on the optimality of a secret reservation price with respect to a public reservation price. 5
to 1, the value of consumption and leisure is equal to ω c (1 l c ) for the employed and to v c (1 l c ) for the unemployed. If each consumer demands a positive quantity of leisure, the inverse demand for each variety i is given by p i = α γq c i ηq c (6) Focusing on the subset of varieties i with positive consumption, the individual demands for variety i are: q c i = α ηn + γ 1 γw c p i + ηn ηn + γ 1 p (7) γwc where N are the consumed varieties and p = (1/N) i Ω p i di is the average price. For unemployed consumers the wage w c is replaced by the value of home production v c. The aggregate demand for variety i in the economy can be derived integrating the individual demands over F (w c ) and G(v c ): q i = α ηn + γ φ ψ γ p i + ηn ψ p (8) ηn + γ γ where φ = w c F (w) df (wc )+ v c G(v) dg(vc ) and ψ = w c F (w) (1/wc )df (w c )+ v c G(v) (1/vc )dg(v c ). The threshold price p M above which the demand for variety i goes to 0 can be found setting the previous equation equal to 0 and solving for p i : p M = γα ψ(ηn + γ) φ + ηn p ηn + γ Thus, the threshold price is a function of the overall number of firms in the market, the average price, and the distribution of wealth across consumers. (9) Production Consider a production structure in which labor is the only production factor and there is only one type of labor required in the production process. To enter the market, firms face an entry cost f E for research and development. Thus, only firms able to cover f E decide to enter the market. After entering the market, firms hire workers for production. To keep the model tractable, recall that every worker can only produce one unit of output. produced is the marginal wage of the last worker w. Thus, the marginal cost of each unit Each firm considers the number of firms and the average price in the market as given and maximizes its profits according to the residual good demand in Equation (8). Equating marginal revenues to marginal costs, the optimal price and the optimal quantity must satisfy: where ŵ = wdw/q(ŵ) is the average wage paid by the firm. q(ŵ) = ψ [p(ŵ) ŵ] (10) γ Notice that since every worker produces only one unit of output, the number of workers in the firm l(ŵ) coincides with the output 6
produced, that is l(ŵ) = q(ŵ). Now, it is possible to write the optimal price, quantity, profits and markups as functions of average wages and threshold prices: p(ŵ) = 1 2 (p M + ŵ) (11) q(ŵ) = ψ 2γ (p M ŵ) (12) π(ŵ) = ψ 4γ (p M ŵ) 2 (13) µ(ŵ) = 1 2 (p M ŵ) (14) where the markup is µ(ŵ) = p(ŵ) ŵ. Thus, an increase in the threshold price p M has a positive effect on all the four variables under consideration. On the other hand, an increase in the average wage paid by the firm has a positive effect on price (that is, p (ŵ) > 0) and a negative effect on quantity produced (q (ŵ) < 0). 3 Notice that if the average wage increases above the threshold price p M, the markup becomes negative and the firm is forced to exit the market. Thus, in markets where threshold prices are structurally quite low, markups are also small and firms make lower wage offers to workers. Equilibrium As noted before, under free entry, only the firms able to cover their initial costs enter the market. Notice that from the maximum price condition in (9), replacing p = (p M + w)/2, it is possible to infer the maximum number of firms on the market: N = 2γ η α p M p M w (15) where w = p M 0 wdf (w)/f (p M ) is the average wage in the economy (not just within the firm). Only the firms with profits greater than the fixed costs f E enter the market. This means that in the aggregate a zero profit condition has the form: pm 0 π(ŵ)df(ŵ) = ψ 4γ pm 0 (p M ŵ) 2 df(ŵ) = f E (16) 3 Notice that the wage of the marginal worker plays a pivotal role in determining the optimal number of workers hired in equilibrium. Indeed, an increase in the marginal wage of the last worker increases the average wage of the firm, decreasing the optimal quantity produced. This corresponds to a decrease in the optimal number of workers in equilibrium, which apparently contradicts the hiring of an additional worker. However, this may happen for specific values of the parameters, as long as the decrease in quantity produced is less than one. Moreover, the impact of an increase in the marginal wage on average wages should be smaller the larger the number of workers in the firm. On the contrary, firms always find it optimal to hire workers at wages lower than the average, since this results in a decrease in price and an increase of the optimal quantity produced. 7
As in Melitz and Ottaviano (2008), to proceed with the analysis, I assume a specific distribution for F(ŵ). For simiplicity, I consider the uniform distribution. As long as the maximum value of the average wage distribution ŵ M is larger than the threshold price, the threshold price is: [ 12γ(ŵM ŵ m )f E p M = ψ ] 1 3 (17) Using Equations (14) and (17), the average markup in the market can be written as: µ = 1 4(ŵ M ŵ m ) [ 12γ(ŵM ŵ m )f E ψ ] 2 3 (18) Thus, the average markup is increasing in the initial research and development cost f E, the degree of product differentiation γ and the upper bound of the average wage distribution ŵ M, and is decreasing in the extremes values of F (w c ) and G(v c ). Discussion This setting predicts that firms in industries with larger barriers to entry and more differentiated varieties can charge higher markups after entering the final good market. This translates in the possibility of offering higher wages to workers. Given the search frictions in the labor market and the uncertainty about workers reservation wages, firms in industries with larger markups may pay higher wages to their workers in order to secure them. Since workers are still paid less than their marginal productivity, firms are always willing to hire an additional worker who accepts a profitable wage offer. Within this framework, if on average immigrant workers have lower reservation wages than native workers, firms in industries with lower markups will hire more immigrant workers to stay on the market. Before moving to the empirical analysis, two remarks are worth mentioning here. First, in this model workers are assumed to be homogeneous in their productivity level. Translating this assumption in the real world, this means that workers with homogeneous characteristics define different segments of the labor market. Second, the model presumes an excess supply of workers in the labor market (i.e. there is unemployment). In fact, this analysis may not be true in the presence of labor shortages. In this case, the hiring of migrant workers may be due to the lack of specific skills on the labor market rather than to low industry markups. 3 Data To empirically confirm the predictions of the theoretical model above, I use individual level data from the American Community Survey (ACS) supplemented by the 1990 Census. Then, I link 8
the fraction of recent immigrant workers by industry to the price cost margins contained in the NBER-CES Manufacturing Industry Database. I focus on recent immigrants because I assume that changes in the competitive structure affects the hiring of new workers. Moreover, recent migrants may be less assimilated in the American society than non-recent migrants and thus should have lower reservation wages. Finally, I exploit the NBER trade database to obtain data about tariffs with China. The next subsection provides an overview of the datasets used. Then, the second and the third subsections describe the construction of the dependent variables and of the Lerner index to measure price cost margins. 3.1 Data sources ACS data The first data source is the American Community Surviey (ACS), which contains data about the nationality of workers, the type of industry in which they work, their occupations and their annual earned income. Particularly, I use the ACS data between 2000 and 2011, supplemented by the 5% sample of the 1990 Census. In this dataset, the definition of the industry where the individuals work deserves particular attention. From 2000 on, industries are coded according to a variation of the North American Industry Classification System (NAICS). Particularly, some NAICS codes in the ACS do not precisely correspond to a single NAICS category of the official NAICS classification. To overcome this problem, I convert these ACS NAICS codes into official NAICS codes with a lower number of digits. 4 On the other hand, in the 1990 Census industries are not reported according to the NAICS classification system. Thus, I convert 1990 Census industry codes into NAICS codes through the concordance table provided by the Census Bureau. Moreover, since ACS NAICS have a different number of digits according to the precision with which industries are recorded, I convert all the NAICS codes into 4-digit NAICS codes. Particularly, I retained the first 4 digits of 5- and 6-digit NAICS codes, and I randomly assigned 1-, 2-, and 3-digit NAICS codes to 4-digit NAICS subcategories. After harmonizing the definition of industries across ACS and 1990 Census data, I restrict the sample to employed workers aged 18 or more with remunerated labor the week before who are not self-employed. Then, I only retain individuals working in a manufacturing industry, i.e. whose NAICS codes begin with a 3, and drop individuals with missing NAICS. After this process, the overall sample size amounts to 2.3 million of observations. Table 1 shows some descriptive statistics about individuals in this dataset and how the subsample of workers in the manufacturing sector relates to the overall sample of workers. On average, people working in the manufacturing sector 4 NAICS codes vary between 1 and 6 digits according to the precision with which the industry is defined. 9
are more likely to be men, married, less educated and to be employed as operators and laborers. Foreign workers are defined according to the variable Citizen. Following the literature (see for example Ottaviano and Peri, 2012), naturalized citizens and non-citizens are classified as foreign workers, while individuals born abroad of American parents are classified as natives. The variable Migrate1 reports the migration decisions of workers with respect to the previous year, asking whether respondents are staying in the same house as the year before, moved within the state, moved across states or moved from abroad. For 1990, this information is available only with respect to the previous 5 years. This variable is called Migrate5 and is only available for 1990. To impute the yearly inflow of migrants according to this variable, I simply assume that migration flows were uniformly distributed across the previous 5 years. Thus, I assign to 1990 the value of Migrate5 divided by 5. However, in the robustness check section I provide an overview of the results without manipulating this variable. Wages are adjusted for inflation and transformed into logarithms. From the ACS and the 1990 Census, I also derive some additional controls. In particular, I construct occupational dummies according to five different occupation categories: managerial and professional specialty occupations; technical, sales, and administrative support occupations; service occupations; precision production, craft, and repair occupations; operators, fabricators, and laborers. To the same extent, I exploit the Siegel prestige score to measure the occupational standing. Finally, all the estimates are presented distinguishing between highly educated and low educated workers. While in the theoretical model I assume that there is only one type of labor required by the firms, in reality there are different types of labor inputs required. While the theoretical model is still valid if considering these different types of labor inputs as different labor markets, in the econometric analysis distinguishing between low and highly educated workers is a rough approximation of the variety of labor inputs observed in the real world. Highly educated workers are defined as those workers with some college education or more. Low educated workers are defined as those workers with high school diploma or less. NBER-CES Manufacturing Industry Database Data about industries are drawn from the NBER-CES Manufacturing Industry Database by Becker et al. (2013). Industries are defined according to harmonized NAICS codes at 6-digit level and are observed from 1958 to 2011. For every 6-digit NAICS industry in the manufacturing sector, this database contains information about the total value of shipments, the total value added, the total number of workers and the total real capital stock. For the purpose of this paper, I only consider years 1990 and from 2000 onwards. Even though this database is the combination of two databases, 10
the Annual Survey of Manufactures (ASM) and the Economic Census (EC), the authors provide interesting additional information about price deflators of investments, value of shipments and cost of materials. Thus, some of the relevant variables, i.e. the value of shipments, the investments and the value added are converted into real terms deflating the nominal values with the appropriate price deflators. The only missing information from this database is the yearly depreciation of capital. However, it is possible to compute the depreciation rate adopting a simple capital law of motion, i.e. K t = (1 δ t 1 )K t 1 + I t where K t and K t 1 are the real capital stock in years t and t 1, δ t 1 is the depreciation rate in year t 1, and I t is the amount of investments in year t in real terms. Then, the depreciation of capital can be obtained multiplying the depreciation rate for the real capital stock. From this database I also extract two other control variables: the real capital stock normalized by the value of shipments and the share of production workers out of total employment. These two measures are also used by Pierce and Schott (2016) to control for the capital intensity of industries and the skill composition of workforce. Finally, since data from the ACS are converted into 4-digit NAICS codes, I harmonized the 6-digit NAICS codes in this database with ACS data averaging the relevant variables at 4-digit NAICS level. Figure 1 shows the trends in four NBER-CES Manufacturing Industry Database variables between 1990 and 2011. The top-left graph in Figure 1 depicts the evolution of the manufacturing sector total employment. As widely documented by Autor et al. (2014) and by Pierce and Schott (2016), total employment in manufacturing sharply decreased after 2001, because of the increase in import competition from China. The top-right and the bottom-left graphs respectively show the evolution of the average value of shipments and the average value added. These two variables show similar trends, with a small downturn after 2001 and a subsequent increase. In both cases it is possible to observe a second trough after 2007, because of the Great Recession. Finally, the bottom-right graph shows the trend in real capital stock, which is steadly increasing before the early 2000s and stabilizes afterwards. NBER trade database Data on US tariffs are drawn from the NBER trade database by Feenstra et al. (2002). They report ad valorem equivalent tariff rates on traded goods between 1989 and 2001. Traded goods are coded according to the 8-digit Harmonized System (HS). Since I am interested in the difference between non-mfn tariffs and MFN tariffs, I first compute this tariff gap for every product. Then, to link the information on tariffs to the industries where the traded goods are produced, I use the 11
concordance table provided by the authors (see Pierce and Schott, 2012, for more details). Finally, I average the tariff gaps at 4-digit NAICS level. 3.2 Dependent variables To measure migration inflows, I consider three different types of migration. 5 First of all, industry competitiveness may trigger international migration. To measure this, I divide the number of foreign-born workers that are working for a specific industry and were abroad the year before by the total number of foreign-born workers in that industry. This measure provides an idea of how fast the stock of immigrant workers is growing in each industry. However, industry competitiveness may not only induce international migration. Immigrant workers living in the US may be more responsive than native workers to changes in the industrial structure and may also migrate within the US. Cadena and Kovak (2016) document this trend showing that when an area is impacted by an adverse economic shock, Mexican immigrant workers respond more strongly than natives geographically relocating to other areas of the US. Thus, the second measure of migration is computed dividing the number of foreign workers in each industry that migrated within the state, across states or from abroad by the number of foreign workers in the industry. This measure gives an idea of the overall migration of foreign-born workers by industry. Finally, changes in industrial competitiveness may also trigger internal migration of native workers. Thus, the third measure is obtained dividing the flow of native and foreign-born workers that moved either within the US or from abroad by the overall number of workers in the manufacturing sector. This measure encompasses all the possible types of migration. Notice that people that moved within the same state are not necessarily migrating. Thus, the second and the third measures of migration may actually overstate the migration patterns. However, the first measure of migration is not affected by this problem and the results presented in the next section are consistent across these three dependent variables. Figure 2 shows the evolution over time of these three dependent variables by education group in absolute numbers. After 2000, the trend is decreasing for all the variables under consideration, even though the decrease seems to be more pronounced among the low educated. In the case of recent foreign-born migrant workers that were abroad the year before, after 2007 the flow of low educated workers becomes even smaller with respect to highly educated workers. The bottom right graph shows the evolution of total employment in the manufacturing sector in absolute numbers. 5 Recall that in our dataset only employed workers in the manufacturing sector have been retained, so that migration flows are computed according to the migratory behavior of individuals working in a specific industry in a given year. For this reason, the migration rates computed here are not comparable to traditional migration rates that are computed at geographical level including the overall population. 12
Even though the scale of the graph does not allow for a better comparison, the trend of employment is similar to the one in Figure 1. Indeed, both employment and foreign workers are decreasing in the period under consideration. At this point, it would be interesting to understand whether there are statistical significant differences between the reservation wages of native and migrant workers. However, information about reservation wages is very difficult to find. Thus, I simply check whether there are differences between the wages of recent migrants and the other workers and more generally if there are differences between the wages of foreign-born and native workers. Table 2 shows the results. In performing these regressions, I focus on individual level ACS and 1990 Census data and I control for basic demographic and socio-economic variables such as age, age squared, gender, marital status, education, race, type of occupation, Siegel prestige score. Row correlations between foreign-born and migration dummies and logarithm of annual income show that foreign-born workers and all the other categories of migrants are paid less than natives or non-migrant workers. Foreign-born recent international migrants seem to be the most penalized category, with wages that are 36% lower than other workers. 3.3 Price cost margins The main explanatory variable is a proxy for the average price cost margin in the industry. Particularly, I compute a Lerner index at industry level in the spirit of Aghion et al. (2005) and Nickell (1996). The larger the index, i.e. the larger the difference between the market price and the marginal cost, the less competitive the market. In the extreme case of perfect competition, this index tends to 0, while in the case of monopoly this index tends to 1. Within this paper, for every industry i and year t, the Lerner index li has the form: ( ) value added depreciation li it = value of shipments In the NBER-CES Manufacturing Industry Database the value added is defined as the difference between the value of shipments and the cost of materials, supplies, containers, fuel, purchased electricity and paid work and is adjusted for inventories and value added from merchandising operations. From this value added, I further subtract the depreciation of real capital and I divide it by the value of shipments. This definition departs from the definition of Aghion et al. (2005) because I cannot observe the financial costs. As a result, I am forced to assume that all the manufacturing industries in my database are incurring in the same financial costs. In addition, since I cannot compute the depreciation rates of capital for 2011, I simply replace the Lerner index it (19) 13
in 2011 with the ratio between the total value added and the total value of shipments. 6 However, in Appendix B I also provide a robustness check without 2011. Figure 3 shows the distribution of the Lerner index computed this way. The distribution has a mean of 0.47 and a standard deviation of 0.12. Notice that very few industries tends to the extremes, i.e. either perfect competition or monopoly. Finally, Figure 4 provides a first grasp of the correlation between the dependent variables and the Lerner index. It focuses on two of the three dependent variables of interest, i.e. the share of recent foreign-born workers who were abroad the year before out of foreign-born workers and the share of recent native and foreign migrants out of total employment, and shows the trends of these variables for the most competitive quartile of industries and for the least competitive quartile. More competitive industries seem to attract more foreign recent international migrants, either high or low educated. Moreover, while the trend of recent foreign international migrants in industries in the least competitive quartile is steadily declining over time, the same trend for the most comptitive industries is converging over time, but started at a much higher level in early 2000s. For what concerns overall migration, concerning both internal and international, native and foreign migrants, trends between the least and the most competitive quartiles seem to be much more similar. However, plotting the trends for different education types, the most competitive industries attract more low educated migrants, while the least competitive industries asttract more highly educated migrants. 4 Empirical strategy and identification 4.1 OLS specification Empirical specification The baseline econometric specification is Y it = α + βli it + ΓX it + κ t + ε it (20) where i indexes each manufacturing industry and t indexes each year. Y it is one of the three dependent variables described above. li it is the Lerner index at industry level, κ t represents time fixed effects, and X it is a set of controls. In the baseline regression, this set of controls only includes the logarithm of industrial capital intensity, that is the stock of real capital normalized by the value of shipments. In a robustness check in Appendix B, I also include a measure of skill composition of employees, i.e. the number of production workers normalized by the total number of workers employed. Indeed, as shown by Lewis (2011), industries located in areas with larger inflows of low 6 As explained in Section 3.1, I infer the depreciation rates from a capital law of motion that encompasses the investments of the following year. 14
educated foreign workers tend to adopt new technologies at a slower pace. Reverting this argument, the average capital intensity of an industry may also influence the migration flows composition, since industries that are more capital intensive may require less low educated labor force and more highly educated labor force. Thus, the capital intensity variable should account for these effects and is included in the baseline specification. In the Appendix, I also provide a robustness check controlling for the skill composition of the industry. 4.2 IV approach However, since migration flows may also affect the price cost margins of an industry, to correctly identify the impact of competition on migration flows, I adopt an instrumental variable approach. Particularly, a compelling instrument should be correlated to the price cost margins of an industry but unrelated to migration inflows. The IV strategy comes from Pierce and Schott (2016). In their paper, these authors investigate whether the granting of the Permanent Normal Trade Relations status to China, i.e. the entry of China in the WTO, is directly related to the decrease in manufacturing employment observed in the early 2000s in the US. Indeed, in trading with the US, China was already benefiting from the most favorite nation (MFN) tariffs before WTO entrance. However, these MFN tariffs were renewed on a yearly basis and especially in early 90s the Congress attempted to revoke this preferential treatment several times. Thus, the entry of China in the WTO did not actually decrease the tariffs faced by US importers, but ended the uncertainty related to the tariff treatment. On this ground, Pierce and Schott (2016) exploit the difference in the tariffs imposed to countries not included in the MFN agreements and the tariffs imposed to countries within MFN agreements, and interact this difference with the time of entry of China in the WTO. Indeed, industries that were exposted to the risk of higher tariffs should have faced a larger increase in competition from China. While they exploit this diff-in-diff strategy to directly infer the impact of this trade policy on employment, I use this interacted variable to instrument the Lerner index at industry level. This is because I want to understand which is the effect on migration of a decrease in price cost margins. Indeed, the end of the uncertainty about China MFN treatment should have disproportionally increased the competition in sectors where the difference between the non-mfn tariffs and MFN tariffs were larger. However, the entry of China in the WTO should not have directly impacted the migration flows to the US. 7 7 Migration flows from China to the US are quite small and Chinese migrants represents roughly 1% of the US population. 15
5 Results 5.1 Main results Table 3 shows the impact of the Lerner index on different types of migration flows. Columns (1) to (3) reports the OLS results considering respectively all workers, low educated and highly educated workers. Columns (4) to (6) reports the IV results. In the first stage of IV regressions, the interaction variable between years and tariff gaps is always positive and significant. This suggests that a larger tariff gap is associated to less competitive industries, i.e. to a larger Lerner index. Moreover, the magnitude of the first stage F-statistic suggests that the instrument is sufficiently correlated with the endogenous explanatory variable. Results are quite consistent across the three dependent variables of interest, with IV coefficients that are always larger in magnitude than OLS coefficients. The upper part of the table reports the results for the share of foreign-born workers that were abroad the year before out of the total number of foreign-born workers. The coefficients are always negative, suggesting that less competitive industries are negatively related to migration inflows of foreign workers living abroad. 8 This is true for both low and highly educated workers. The coefficient in Column (4) for low and highly educated workers pooled together is 0.075, suggesting that a one strandard deviation increase in the Lerner index corresponds to a decrease of 45% in the mean of the dependent variable. Notice that two thirds of the coefficient are due to low educated foreign-born workers, while one third of the coefficient is due to highly educated workers. The central part of Table 3 reports the results for the share of foreign-born workers that migrated within the state, across states or from abroad the year before out of the total number of foreign-born workers. Now, the magnitude of the effect is smaller than before, with a standard deviation increase in the Lerner index inducing a decrease of 13% in the mean of the dependent variable. This smaller effect could be due to the fact that internal foreign-born migrants may be more assimilated into the US society with respect to international migrants. Indeed, if the lower reservation wages are the driving force inducing firms with smaller markups to hire more migrant workers, internal migrants may be less responsive than international migrants, because they may have higher reservation wages. In addition, notice that the coefficients for highly educated workers become positive. This suggests that less competitive industries are more likely to induce migration of highly educated workers, probably because of labor shortages of specific skills. Finally, the bottom part of Table 3 reports the results considering both native and foreignborn internal and international migrants. The impact of the Lerner index is still negative for low 8 Notice that an increase in the Lerner index corresponds to a decrease in the competitiveness of the industries. 16
educated migrants and positive and significant for highly educated migrants. The coefficient for low and highly educated workers pooled together is now 5% of the mean of the dependent variable, suggesting that more competitive industries attract less native workers. Nevertheless, industries with larger price cost margins seem to trigger more internal migration of highly educated workers. As suggested before, these results are in line with the theoretical model of Section 2. Indeed, while more competitive industries may be constrained by lower threshold prices to stay on the market and attract more low educated workers, less competitive industries may not be constrained by threshold prices but may be labor constrained in terms of more specialized labor, i.e. they need more highly educated workers to increase production. 5.2 Robustness checks The robustness checks in this section have been anticipated in Sections 3 and 4 and aim to show the sensitivity of the results to different empirical specifications and computational choices. Table B.1 in Appendix shows the results on migration flows controlling for industry skill composition. Results are qualitatively similar to the main estimates, even though the coefficients are slightly smaller in magnitude for low educated workers and slightly larger for highly educated workers. Also, accounting for the skill composition of industries, the coefficients for highly educated workers are negative. Another robustness check on ACS and 1990 Census data consists in using the M igrate5 variable for 1990 without dividing it by 5. Results are very similar to the main estimates and are reported in Table B.2. In Table B.3, I completely exclude the year 1990. The results are still quite similar to the main estimates. In Table B.4, I replicate the analysis merging the ACS and the NBER-CES Manufacturing Industry Database in a different way. Rather than adjusting the ACS NAICS codes to 4-digit NAICS, I adjust the NBER-CES Manufacturing Industry Database NAICS codes. Particularly, I aggregate the Lerner index and the other relevant variables at shorter NAICS levels: 5-digit NAICS, 4-digit NAICS, etc. Then, I match these new categories with the NAICS codes in the ACS. Once again, the results are very similar to the main estimates. Finally, the last robustness check is on the definition of the Lerner index. Since for 2011 it was not possible to subtract the depreciation of capital from the value added, Table B.5 presents the results of the estimates without 2011. Again, estimates are very similar in magnitude to the main results. 6 Conclusion This paper investigates whether the competitive structure of an industry influences either international or internal migration flows. To this purpose, I develop a theoretical model and an instrumen- 17
tal variable approach. Theoretically, I model this problem combining a monopolistic competitive market for the final good with a labor market with search frictions and workers that are heterogeneous in their reservation wages. Particularly, there is a threshold maximum price for the final good above which the firm is forced to exit the market. For this reason, firms attempt to decrease the marginal cost and to hire lower reservation wage workers. If migrant workers have lower reservation wages with respect to non-migrant workers, more competitive firms hire more migrant workers. However, if there are labor shortages, all the workers are hired and firms need more workers to increase production. In this case, migration is needed to let the market grow. The empirical analysis confirms these results. More competitive industries seem to attract more international migrants, either low or highly educated. However, internal migration patterns are less affected by the structure of competition and the sign of the effect differs according to the education of workers. More competitive markets attract more low educated workers, while less competitive markets tend to attract more highly educated workers. These results are in line with the theoretical model, since more competitive firms may attempt to hire more international migrants or low educated internal migrants to decrease marginal costs, while less competitive firms may trigger internal migration of highly educated migrants to expand production. These results have strong policy implications. This analysis shows that migratory behaviors are intrinsecally related to competitive markets. As long as firm markups are small, firms always attempt to decrease their marginal costs hiring cheap labor. In this case, policies aiming at limiting migration setting stricter border rules may not be effective. Indeed, as long as international migrants are aware of the possibility of finding a job in the destination country, stricter border enforcement would only foster illigal immigration. Rather, an effective migration policy should also take into account the characteristics of the labor demand of national firms. For instance, the inflow of low educated workers may be better hampered in the long run through better investments in high markup industries. If anything, these industries trigger highly educated migration flows. This analysis is limited by the quality of the data used. Particularly, matched employee-employer datasets would be more effective in investigating this research question. Moreover, it would be interesting to complement this analysis with measures of industry offshoring and outsourcing. Further research will follow these directions. 18
References Aghion, P., Bloom, N., Blundell, R., Griffith, R., Howitt, P., 2005. Competition and innovation: An inverted-u relationship. The Quarterly Journal of Economics 120 (2), 701 728. Albrecht, J. W., Axell, B., 1984. An equilibrium model of search unemployment. Journal of political Economy 92 (5), 824 840. Antras, P., Helpman, E., 2004. Global sourcing. Journal of Political Economy 112 (3), 552 580. Autor, D. H., Dorn, D., Hanson, G. H., Song, J., 2014. Trade adjustment: Worker-level evidence. The Quarterly Journal of Economics 129 (4), 1799 1860. Becker, R., Gray, W., Marvakov, J., 2013. Nber-ces manufacturing industry database: Technical notes. NBER Working Paper 5809. Borjas, G. J., 2003. The labor demand curve is downward sloping: Reexamining the impact of immigration on the labor market. Quarterly Journal of Economics 118 (4), 1335 1374. Borjas, G. J., Katz, L. F., 2007. The evolution of the Mexican-born workforce in the United States. In: Mexican immigration to the United States. University of Chicago Press, pp. 13 56. Cadena, B. C., Kovak, B. K., 2016. Immigrants equilibrate local labor markets: Evidence from the great recession. American Economic Journal: Applied Economics 8 (1), 257 290. Card, D., 2001. Immigrant inflow, native outflow, and the local labor market impact of higher immigration. Journal of Labor Economics 19 (1), 22 64. Dustmann, C., Schönberg, U., Stuhler, J., 2017. Labor supply shocks, native wages, and the adjustment of local employment. The Quarterly Journal of Economics 132 (1), 435 483. Eckstein, Z., Van den Berg, G. J., 2007. Empirical labor search: A survey. Journal of Econometrics 136 (2), 531 564. Elyakime, B., Laffont, J. J., Loisel, P., Vuong, Q., 1994. First-price sealed-bid auctions with secret reservation prices. Annales d Economie et de Statistique, 115 141. Feenstra, R. C., Romalis, J., Schott, P. K., 2002. Us imports, exports, and tariff data, 1989-2001. NBER Working Paper 9387. Grossman, G. M., Helpman, E., 2002. Integration versus outsourcing in industry equilibrium. The Quarterly Journal of Economics 117 (1), 85 120. 19
Grossman, G. M., Rossi-Hansberg, E., 2012. Task trade between similar countries. Econometrica 80 (2), 593 629. Lewis, E., 2011. Immigration, skill mix, and capital skill complementarity. The Quarterly Journal of Economics 126 (2), 1029 1069. Meagher, K. J., Wilson, H., 2004. Different firm size effects on wages for supervisors and workers. Economics Letters 84 (2), 225 230. Melitz, M. J., Ottaviano, G. I., 2008. Market size, trade, and productivity. The review of economic studies 75 (1), 295 316. Nickell, S. J., 1996. Competition and corporate performance. Journal of political economy 104 (4), 724 746. Oi, W. Y., Idson, T. L., 1999. Firm size and wages. Handbook of labor economics 3, 2165 2214. Ottaviano, G. I., Peri, G., 2012. Rethinking the effect of immigration on wages. Journal of the European Economic Association 10 (1), 152 197. Ottaviano, G. I., Peri, G., Wright, G. C., 2013. Immigration, offshoring, and american jobs. The American Economic Review 103 (5), 1925 1959. Pierce, J. R., Schott, P. K., 2012. A concordance between ten-digit us harmonized system codes and sic/naics product classes and industries. Journal of Economic and Social Measurement 37 (1, 2), 61 96. Pierce, J. R., Schott, P. K., 2016. The surprisingly swift decline of us manufacturing employment. The American Economic Review 106 (7), 1632 1662. Piore, M. J., 1979. Birds of passage: migrant labor and industrial societies. Cambridge England Cambridge Univ. Press 1979. Troske, K. R., 1999. Evidence on the employer size-wage premium from worker-establishment matched data. The review of Economics and Statistics 81 (1), 15 26. 20
Figure 1: Evolution of employment, value of shipments, value added and real capital stock over time Notes - Employment, value of shipments, value added and real capital stock are summed across industries. Value of shipments and value added are deflated according to the value of shipment price deflator. Sources: NBER-CES Manufacturing Industry Database - years 1990-2011. 21
Figure 2: Evolution over time of migration flows in absolute numbers Notes - The graphs consider the evolution over time of the total number of foreign-born workers that were abroad the year before, the total number of foreign-born workers that were abroad or moved across states or counties, the total number of both native and foreign-born workers that moved from abroad or across states or counties, and the total employment in the manufacturing sector. Moreover, each graph shows the contribution of high educated and low educated workers. High educated workers are defined as workers with at least one year of college or more, while low educated workers are defined as workers with high school diploma or less. Years between 1990 and 2000 are missing. Reported trends between 2001 and 2010 are smoothed replacing the actual values with 3-year moving averages. Moving averages for 2000 and 2011 are based on 2-year spans. The data for 1990, instead, are reported without smoothing. Sources: ACS and 1990 Census - years 1990, 2000-2011. 22
Figure 3: Lerner index distribution Notes - The graph depicts the distribution of the Lerner index across industries and years. The Lerner index is computed as value added minus depreciation of capital divided by value of shipments. The value added is computed as the difference between the value of shipments and the cost of materials, supplies, containers, fuel, purchased electricity and paid work. In addition, it is adjusted for inventories and value added from merchandising operations. All these values are deflated according to the relevant deflator in the NBER-CES Manufacturing Industry Database. Sources: NBER-CES Manufcaturing Industry Database - years 1990, 2000-2011. 23