The Internal Labor Markets of Business Groups∗

Cristobal Huneeus† Federico Huneeus‡ Borja Larrain§

Mauricio Larrain¶ Mounu Prem‖

September 28, 2018

Abstract

Labor reallocation in response to international shocks is stronger between pairsof firms in business groups than between pairs of unaffiliated firms. Reallocation isconcentrated among top-occupation employees. The effects are stronger when theemployee’s origin firm controls the destination firm, and in more complex ownershipstructures. Wages of top employees increase as they move within the group. Weprovide suggestive evidence that reallocation increases the destination firm’s prof-itability and is beneficial for the group as a whole. Our results are consistent withthe idea that common ownership facilitates the acquisition and transfer of intangibleinputs (e.g., management practices).

Keywords: Internal labor markets; business groups; ownership; intangible inputs.JEL Codes: G32.

∗We thank comments and suggestions from Will Dobbie, David Lee, Alex Mas, Adrien Matray, AtifMian, Gabriel Natividad, Esteban Rossi-Hansberg, David Schoenherr, Geoffrey Tate (discussant), andseminar participants at Princeton University and FinanceUC 14th International Conference. The analysiswas implemented when Cristobal Huneeus worked in the Ministerio del Trabajo de Chile. We thank theMinisterio del Trabajo de Chile for coordinating the access to an anonymized version of the unemploy-ment insurance dataset. Borja Larrain acknowledges funding from Proyecto Fondecyt Regular 1180593.Federico Huneeus acknowledges funding and support from the International Economic Section (IES), theSimpson Center for the Study of Macroeconomics, the Program in Latin American Studies (PLAS) andthe Fellowship of Woodrow Wilson Scholars, all from Princeton University. Their support is gratefullyacknowledged.†Unholster.‡Economics Department, Princeton University, e-mail: fhuneeus@princeton.edu§Escuela de Administracion and FinanceUC, Pontificia Universidad Catolica de Chile, e-mail:

borja.larrain@uc.cl¶Escuela de Administracion, Pontificia Universidad Catolica de Chile, e-mail:

mauriciolarraine@gmail.com‖Economics Department, Universidad del Rosario, e-mail: francisco.munoz@urosario.edu.co

1

1 Introduction

Business groups are common corporate structures around the world. More than half oflisted firms in emerging markets, and a sizeable fraction in many developed countries,are affiliated to business groups (Khanna and Yafeh 2007; Morck, Wolfenzon, and Yeung2005). Business groups consist of legally independent firms controlled by the same ulti-mate owner. For example, Figure 1 shows the corporate organization of Antarchile, oneof largest business group in Chile, and controlled by the Angelini family. There are sev-eral reasons that can explain the existence of these complex ownership structures. Oneset of explanations argues that, unlike stand-alone firms, business groups create factormarkets between affiliated firms to alleviate market frictions and improve profitability.Along these lines, there is growing evidence that business groups transfer capital betweenfirms to overcome financial constraints (Gopalan, Nanda, and Seru 2007; Boutin, Cestone,Fumagalli, Pica, and Serrano-Velarde 2013; Buchuk, Larrain, Munoz, and Urzua 2014;Almeida, Kim, and Kim 2015).

However, we still know very little about other factor markets in business groups, inparticular internal labor markets. Labor markets are affected by several frictions, fromtransaction costs in hiring and firing, to imperfect insurance, and asymmetric information.The use of internal labor markets could allow business groups to overcome these frictions,and give group firms an advantage over stand-alone firms. In this paper we provideevidence that business groups actively use internal labor markets. We explore potentialdrivers behind the use of internal markets. Our evidence is most consistent with anexplanation where common ownership facilitates the acquisition and transfer of intangibleinputs between firms (Atalay, Hortacsu, and Syverson 2014). These intangible inputs (e.g.,management practices, knowledge, know-how, organizational culture, etc.) are typicallyembedded in employees (Eisfeldt and Papanikolaou 2013). We also provide suggestiveevidence that internal labor mobility increases profitability and benefits the business groupas a whole. Overall, we identify a novel beneficial aspect of business groups.

We use a detailed matched employer-employee dataset to study labor market mobilitybetween Chilean firms. Our data allows us to follow the same employee switching jobs fromone firm to another. The data covers the universe of employment movements between pairsof Chilean firms between 2004 and 2015. We merge our employment data with business-group affiliation data. We start the analysis with evidence of unconditional (i.e., unrelatedto shocks) labor market mobility within business groups. We show that when a groupfirm releases employees, the flow towards the average firm in the same group is 5 timeslarger than towards the average unaffiliated firm (see Figure 2). Internal labor mobility is

2

monotonically stronger as we move from bottom-occupation to top-occupation employees,which we define in terms of the within-firm wage distribution. These effects control forfirm size (total employment) and wage differentials between origin and destination firms.

Stronger labor mobility between group firms could be the result of group firms facingmore shocks to the marginal product of labor than stand-alone firms. Therefore, weneed to account for profitability shocks faced by firms. We use firm-level internationaltrade shocks to proxy for changing business conditions (Autor, Dorn, and Hanson 2013;Hummels, Jørgensen, Munch, and Xiang 2014). We merge our employment data withadministrative customs data, which contains detailed information on firms’ exports andimports. We define trade shocks using a Bartik-type strategy: for each firm, we computethe weighted average of the world prices of the goods the firm exports (or imports), wherethe weights correspond to the initial share of the different goods exported (imported) bythe firm. These shocks are relevant profitability shocks for the firms in our sample sinceChile is fully open to international trade. Moreover, because Chile is a small economy thatdoes not affect world prices, we believe that these shocks are plausibly exogenous. The useof exogenous shocks gives us an advantage with respect to the literature, which typicallyuses more endogenous shocks such as firm closures and mass layoffs when studying internallabor markets.

Our main tests compare labor mobility between group firms and between unaffiliatedfirms as a function of relative international shocks between firms. For instance, imagine afirm receives a negative export shock relative to other firms. Neoclassical theory suggeststhat this firm should release employees to other firms in the market, regardless of whetherthey are affiliated to the same business group or not. Instead, we find that when a groupfirm faces a one-standard-deviation negative export shock, the average firm in the samegroup receives a larger flow of top employees than the average unaffiliated firm. Thedifferential flow is equivalent to 20% of the standard deviation of top employment flows.

Although we believe that international shocks are plausibly exogenous to the firm,our regression still compares group firms with stand-alone firms. Because the decisionto be part of a business group is a choice variable, this comparison could be subject toendogeneity. To deal in part with this concern, in our next test we restrict the sample tobusiness group firms. We show that, only among destination firms affiliated to the group,employees still move towards those firms facing the most favorable profitability conditionswithin the group.

Next, we explore the rationale behind the internal labor markets of business groups.We study different frictions that can make internal labor markets more convenient thanexternal labor markets. At face value, our evidence is not consistent with explanations

3

based on transaction costs, asymmetric information, diversification, and imperfect insur-ance. For example, internal mobility is uncorrelated with tenure, which can be a proxyfor firing costs and the extent of asymmetric information between the employee and thefirm. Sectoral diversification is also unrelated to internal mobility. A key piece of evidenceagainst the insurance hypothesis is that the wages of top employees increase as they movewithin the group. Under the insurance hypothesis, there should be no pass-through ofshocks to wages, or wages should decrease in exchange for the job security provided bythe group (Blundell, Pistaferri, and Preston 2008; Cestone, Fumagalli, Kramarz, and Pica2017).

Our results are most consistent with a model where common ownership facilitatesthe acquisition and transfer of intangible inputs as in Atalay, Hortacsu, and Syverson(2014). Intangibles are hard to transfer since they are often embedded in the humancapital of employees. This can explain, for example, persistent differences in managementpractices across firms (Bloom and Van Reenen 2007). An explanation along these lines canaccount for the concentration of the effects among top employees, for whom intangiblesare more relevant. We explore different complementary explanations for the impact ofbusiness groups on the production and distribution of intangible inputs, most notablycontractual incompleteness (Grossman and Hart 1986) and hierarchies (Hart and Moore2005; Garicano and Rossi-Hansberg 2015). For example, our results are stronger whenthe firm of origin directly controls the firm of destination of the employee, and when theownership structure is more complex, which showcases the importance of the hierarchicalstructure in the transfer of top employees. Although no particular model can account forall empirical findings, an explanation based on intangible inputs seems to be the mostsatisfying among the candidate explanations.

In the final part of the paper we study whether the impact of internal labor mobilityaggregates up to firm-level outcomes. The evidence in this part is only suggestive, becausewe cannot exclude that other margins of adjustment are simultaneously working in busi-ness groups. First, we show that international shocks lead to higher wage growth and loweremployment growth in group firms than stand-alone firms. This reduced-form evidenceis consistent with Faccio and O’Brien (2017). Second, we show that when a group firmreceives a top employee from within the group, the average wage in the firm (our proxy forproductivity) increases. This is relative to other group firms that are experiencing similarshocks and that receive top employees from unaffiliated firms. This is consistent with theidea that the employee transferred within the group is more productive and that she isequipped with more intangibles. Simply put, she is a better manager. Third, the averagefirm-level wage remains unchanged when a group firm releases a top employee. Because

4

the overall wage of the business group is a weighted average of the wages of the differentgroup firms, this suggests that internal reallocation increases group-level productivity andis therefore beneficial for the group as a whole.

Previous literature on internal labor markets is usually focused on conglomerates withfully-owned subsidiaries or plants. Although similar mechanisms apply in conglomeratesand business groups, a key difference is that firms in business groups remain independentcorporations. In conglomerates, different units share a common balance sheet, they typ-ically have integrated finance, marketing and human resources departments, and otherjoint processes (e.g., distribution channels), which blur or erase firm boundaries. The onlylink between firms in a business group is often the controlling shareholder. Common own-ership provides control rights without erasing firm boundaries as in conglomerates. Theadvantages of business groups vis-a-vis conglomerates with respect to labor markets canbe more subtle. In particular, our results suggest that business groups alleviate frictionsmostly among top employees, while previous literature on conglomerates or multi-plantfirms does not find major differences between employee types (see Giroud and Mueller2015; Tate and Yang 2015; and Silva 2017). From the point of view of ownership struc-tures and control rights (Hart 2001), our results suggest that exercising control oftenmeans transferring top employees.

The internal labor markets of business groups are less studied than internal capitalmarkets, although there are a few recent exceptions such as Belenzon and Tsolmon (2016),Faccio and O’Brien (2017), and Cestone, Fumagalli, Kramarz, and Pica (2017). The firsttwo use firm-level data in a cross-country analysis. We are closer to Cestone, Fumagalli,Kramarz, and Pica (2017), who also use employer-employee data. We provide three mainnovelties compared to these previous papers. First, our identification strategy is based onthe response of business groups to arguably more exogenous shocks by taking advantage ofthe small and open nature of the Chilean economy. Second, we map the inner structure ofbusiness groups in a more detailed way in order to test for mechanisms. For instance, wetest for the importance of direct control links and the hierarchical structure of the groupon employment reallocations. Finally, our finding of increasing wages for employees thatmove between group firms (see Section 4) speaks against the insurance hypothesis, whichhas been the focus of much of the previous literature. We propose, instead, a novelexplanation for the use of internal labor markets, based on an advantage of groups infacilitating the acquisition and transfer of intangible inputs.

The remainder of the paper proceeds as follows. Section 2 describes the data. Section3 presents the main evidence in support of the existence of internal labor markets. Section4 tests different rationales for using internal labor markets. Section 5 discusses the firm-

5

level effects of internal labor markets. Section 6 concludes. The Appendix providessupplementary data description and results.

2 Data Sources

We combine three data sources to study the internal labor markets of business groups.First, we use a matched employer-employee dataset to identify employment flows betweenfirms. Second, we use a business-group dataset to link the firms in our sample accordingto their ownership structure. We classify firms into business group firms and stand-alone firms and we identify firms that belong to the same business group. Third, we useinternational trade data to construct export and import shocks, which we use to obtainexogenous firm-level profitability shocks.1

2.1 Matched Employer-Employee Dataset

For the purposes of providing unemployment insurance, Chilean firms are required by lawto pay a fraction of employees’ monthly wages into an individual savings account and acommon fund that can be used by all employees in case of unemployment. The unem-ployment insurance system is managed by a private entity, which keeps an administrativedataset, the Unemployment Insurance (UI) dataset.2 This dataset has the wage, at amonthly frequency, for each employer-employee relationship. Besides wages, firms needto report their main industry and the gender and birth date of the respective worker. Thedataset does not report information of the specific professional category or occupationof each employee. Past research on worker-firm matching has used similar datasets fromother countries, such as Germany and Portugal (Card, Heining, and Kline, 2013).

This dataset has three features that are relevant for our study. First, it covers theentire private (formal) labor market in Chile. Second, since Chile’s administrative datasetshave unique tax IDs for both workers and firms, we can keep track of both across timeand merge them to other datasets. In particular, this dataset includes listed firms, whichwe use to merge to the business group dataset. Finally, given that we have the employer-employee relationships, we have the entire wage distribution both within and across firms.

1The appendix describes in detail the limitations of these three datasets and the cleaning and mergingprocedures that we use.

2The U.S. equivalent to this dataset is the US Longitudinal Employer-Household Dynamics (LEHD).

6

2.2 Business Group Dataset

Chilean listed firms are required by law to report financial statements and ownershipstructures regularly to the local stock market regulator (Superintendencia de Valores ySeguros, SVS). From the universe of listed firms, we define a business group as a set oftwo or more listed firms with a common controlling shareholder (Buchuk, Larrain, Munoz,and Urzua 2014). Financial statements typically report links between corporations, butnot the names of individual shareholders. We identify the controlling shareholder bychecking the composition of boards, annual reports, and the financial press. Controllingshareholders are families, foreign multinationals, or small groups of large investors whoact in a coordinated way. The state is not a relevant controlling shareholder of listed firmsin Chile. The ownership stakes of controlling shareholders are stable across long periodsof time in the Chilean market (Donelli, Larrain, and Urzua 2013).

Using the information reported by the listed firms we can track the private firms thatare related to the listed firms, and hence that also belong to each group. Ownership linkswith private firms are reported in two ways. First, there is a list of firms that consolidatewith each listed firm. Accounting consolidation means that the firm exerts a “controllinginfluence” over the other firm. In practice, consolidation typically implies an ownershipstake above 50%. Second, there is a list of related investments by each listed firm. Thislist has information on firms where the listed firm has a large and permanent investment,although the type of influence does not imply accounting consolidation. Related invest-ments typically involve ownership stakes between 10% and 50%. Since ownership stakesare significant we consider that the firms in related investments also belong to a group iftheir parent has been identified as a group firm.3

Using all the previous information we define the set of firms –public and private–that conform each business group. During the sample period for which we have thisinformation, 2004-2015, we identify 29 groups comprising approximately 93 listed firmsand 743 private firms. Figure 1 provides an example of a business group in our data. Thegroup is controlled by the Angelini family and has 5 listed firms: the holding company–Antarchile– at the top of the pyramid plus 4 firms in the second layer of the ownershipstruture.

2.3 International Trade Dataset

In order to have exogenous variation to firms’ profitability, we build a dataset of in-ternational price shocks to firms that either export or import. We use two sources for

3If two or more groups have stakes in a firm we assign the firm to the group with the largest stake.

7

international trade data. The first provides Chilean firms’ international trade activity.Chilean law requires firms to report detailed import and export activity to the customsauthority. Each firm reports the quantity and value of every international transactionwith information at the product-country level combination. Product definition is at thesix-digit level from the Harmonized System (HS).

With this dataset we can compute unit values of both exports and imports at the firmlevel. Although Chilean firms are small relative to global markets, some of the variationof these unit values can be due to Chilean firms’ specific characteristics. In order toavoid endogeneity concerns we build international versions of these unit values for everyproduct-country combination. In the spirit of the trade and labor literature (Autor, Dorn,and Hanson 2013; Hummels, Jørgensen, Munch, and Xiang 2014), when building theseprices we ignore trade flows with Chile. We use international trade data at the product-country level coming from a repository of official international trade information collectedby the UN Statistical Division, COMTRADE.

Using both the firm-level and global datasets, we build international price shocks forboth exports and imports at the firm-level as follows:

∆pkit =

N∑j=1

skij0∆pG,k

jt

where k={export, import}, i index firms, t time, and j are product-country combinations,pG,k

jt is the global unit value of a product-country pair j being exported (imported), andsk

ij0 is the share of product-country j relative to all the corresponding activities k offirm i in year 0 (2004). Overall, this strategy is similar to how Bartik-type shocks areconstructed.4

Relevance of Exports and Imports. Since most trade flows are concentrated infirms that simultaneously export and import, we assess the relative importance of eachactivity. We compare exports and imports to firms’ wage expenditures. We find that, onaverage, firms in our sample spend 1.3 in imports and get 2 in exports for every dollarspent in wages. This suggests that export shocks are more relevant than import shocksin our sample.

Industry Distribution. We first compute the industry composition of the cleanedUI dataset and compare it with the ones merged with the business group dataset andwith both the business group and international trade datasets. The details are providedin Figure A.2 in the appendix. We see that, relative to the cleaned UI dataset, the other

4In Figure A.1 we show correlations between the initial shares of exposure (skij0) and future price

changes. We do not see any relationship between them.

8

two datasets are less intensive in services and more intensive in manufacturing. In otherwords, our sample is more intensive in tradable goods, which can be expected since westudy firms that are exposed to international shocks.

3 Internal Labor Markets at Work

For each pair of firms, we compute the number of employees moving from the origin firmto the destination firm every year. We look at employment flows from the point of view ofthe origin firm (employment outflow) and from the point of view of the destination firm(employment inflow). From the point of view of the origin firm, we define the employmentoutflow as the number of employees moving from the origin to the destination firm, relativeto the total number of employees released by the origin firm that year. From the point ofview of the destination firm, we define the employment inflow as the number of employeesmoving from the origin to the destination firm, relative to the total number of employeesabsorbed by the destination firm that year. A firm in the sample can either belong to abusiness group or can be a stand-alone firm. For every pair of firms, we say that the firmsare affiliated if they belong to the same group and unaffiliated if not.

3.1 Preliminary Evidence

Figure 2 plots the average employment outflows between pairs of affiliated and unaffiliatedfirms in our sample. We restrict the sample to pairs of firms in which the origin firmbelongs to a business group. We find that when a group firm releases employees, the flowtowards the average firm in the same group is more than 5 times larger than towardsthe average unaffiliated firm (0.115 vs. 0.021). When a group firm hires employees, theflow from the average firm in the same group is more than 4 times larger than from theaverage unaffiliated firm (0.098 vs. 0.021).5 Appendix Table A.1 reports the results intable format, documenting the high level of the statistical significance of the differences.The table also reports the results when restricting the sample to exporting and importingfirms, which is the sample we use in the next section, and the results are analogous.

Figure 2 also reports the averages flows for top-occupation employees alone, which wedefine as the employees in the top quartile of the firm’s wage distribution. Employmentflows between firms affiliated to the same business group are more concentrated for top-

5Average employment outflows and inflows between affiliated firms and unaffiliated firms do not sumup to one because these numbers are defined at the firm-pair level. In order for them to sum up to one,they need to be scaled up by how many relationships a given firm has on each margin, i.e., with affiliatedand unaffiliated firms, which is heterogeneous across firms.

9

occupation employees than for the average employee of the sample. For example, theoutflow towards the average firm in the same group is almost 3 times more concentratedfor top-employees than for all employees (0.294 vs. 0.115). Similarly to the flow of allemployees, we also see differences between affiliated and unaffiliated pairs. When a groupfirm releases top-occupation employees, the flow towards the average firm in the samegroup is much larger than towards the average unaffiliated firm (0.294 vs. 0.112), andaccounts for almost 1/3 of the employees released by the origin firm. It is important tonote that we do not find similar results when defining top-occupation workers from theoverall wage distribution in the economy. Hence, our findings are about top positionswithin the firm rather than about high nominal wages.

In Table 1 we study labor flows in a regression format. We regress the employmentoutflows between origin firm o and destination firm d on a dummy variable that takes thevalue of one if the firms belong to the same business group, and zero otherwise. Since weinclude only pairs of firms with positive flows between them, we are studying the intensivemargin of labor flows and not the extensive margin (i.e., the likelihood of observing anemployment flow between any two firms).

Outflowodt = β Same Groupodt + αo + αd + αt + εodt (3.1)

We include origin-firm fixed effects (αo) to control for all time-invariant sending-firmcharacteristics; destination-firm fixed effects (αd) to control for receiving-firm character-istics; and year fixed effects (αt) to account for annual common shocks to all firms. Allof our regressions are at the annual frquency. We cluster the standard errors of this re-gression, and all firm-pair regressions, at both the origin-firm and destination-firm levels.For conciseness we report results only with employment outflows as dependent variablethroughout the paper. The results with inflows are analogous.

The results in column 1 of Table 1 confirm the graphical evidence of Figure 2. Whena group firm releases employees, the flow towards the average firm in the same group islarger than towards the average unaffiliated firm (coefficient 0.024, t-stat 4). This is a largeeffect, accounting for more than 55% of the standard deviation of employment outflows(=0.024/0.043). In column 2 we control for the difference in average wage between theorigin and destination firms, which in principle should be a sufficient statistic to explainemployment flows between firms. We also control for time-varying level of employmentof the sending and receiving firms, of the business groups involved, and their respectivenumber of firms.6

6We control for the time-varying size of the origin and destination firms because employment dynamics(e.g., labor churning) vary with firm size. Likewise, it can be the case that flows between firms vary with

10

In columns 3 to 6 of Table 1 we re-estimate equation (3.1) focusing on employees indifferent segments of the within-firm wage distribution. We confirm that labor mobilityinside business groups is stronger for top-occupation employees. The effect is mono-tonically increasing in the segment of the wage distribution. For example, the effect ofbusiness group affiliation on employees at the top 25th percentile of the distribution is al-most 2.5 times larger than for employees at the bottom 25th percentile of the distribution(=0.047/0.019).

Employment flows between firms belonging to the same business group are strongerthan between unaffiliated firms. However, the evidence does not say much regarding howinternal labor markets work. In particular, larger employment flows between affiliatedfirms could be the result of group firms facing more profitability shocks than stand-alonefirms. To study internal labor markets at work, we need to account for profitability shocksfaced by firms as we do in the next section.

3.2 The Internal Labor Mobility Test

In a simple neoclassical setting, employment flows between two firms should be the resultof shocks to the marginal product of labor. Employment flows between group firmsshould not react differently than employment flows between unaffiliated firms. Instead,in a setting with labor market frictions, the owner of a business group can create value byactively reallocating scarce resources across affiliated firms. Following the logic of Stein(1997), in a business group that owns two firms i and j, and if the appeal of investingin firm j suddenly increases, employment in firm i should decline (even if it is a positivenet-present-value investment at the margin), as the business-group owner channels morescarce resources towards firm j.

We use firm-level trade shocks as proxy for profitability shocks. Because Chile is asmall and open economy, trade shocks are plausibly exogenous to the firms in our sample.This is an advantage relative to the existing literature (Cestone, Fumagalli, Kramarz, andPica, 2017), which uses more endogenous shocks such as firm closures and mass layoffs.An increase in export prices constitutes a positive shock for an exporting firm; an increasein import prices constitutes a negative cost shock for an importing firm. Because we usetrade shocks as part of our identification strategy, we restrict our sample to firms thatsimultaneously export and import. In Section 3.4 below, we use an alternative measure ofprofitability shocks (changes in industry-level Tobin’s Q), which allows to use a broadersample of firms.

the size of the origin and destination firms.

11

In econometric terms, we regress the employment outflow between a pair of firms onthe relative profitability shocks between the two firms, the same-group dummy variable,and the interaction between the relative shocks and the same-group dummy:

Outflowodt = β Same Groupodt + γ(∆pkdt −∆pk

ot) (3.2)

+ δ[Same Groupodt × (∆pkdt −∆pk

ot)]

+ αo + αd + αt + εodt

The γ coefficient is the baseline sensitivity of employment flows to relative shocksbetween firms. We expect γ > 0 for export shocks and γ < 0 for import shocks. Wenormalize the relative shocks by its standard deviation, such that γ represents the effectof a one-standard-deviation increase in the relative shock. The main coefficient of interestis δ, which captures the excess sensitivity of intra-group flows to shocks. Under theneoclassical theory δ = 0 because the fact that the two firms are related through acommon owner should not matter for labor flows, i.e., this is the labor-market version ofthe irrelevance proposition of Modigliani and Miller (1958). In such a world, the relativeshock to the marginal product of labor is a sufficient statistic for employment flows, whichis captured by γ alone. If, instead, internal labor markets do matter for reallocations,then we should find a statistically significant δ. Internal labor markets can, in principle,dampen (δ < 0) or enhance (δ > 0) the response of employment flows to shocks.

Equation (3.2) includes fixed effects for the origin firm, destination firm, and year. Inprinciple, instead of controlling for origin-firm and destination-firm fixed effects separately,we could control for origin-destination pair fixed effects. This would allow to exploitwithin-pair variation across time. However, we observe too few repeated flows betweenpairs of firms over time. This is particularly the case when we focus on top employees.In Section 3.4 below, we explore the possibility of adding firm-pair fixed effects using abroader sample of firms.

We report the results in Table 2. The interaction term between the same group dummyand the differential export shock is marginally significant at the 10% level for the sampleof all employees (column 2). In columns (3) to (5), we report the results for employees indifferent segments of the within-firm wage distribution. The effect increases monotonicallyas we move up in the wage distribution. In fact, the effect is only significant amongemployees at the top 25th percentile. When a group firm faces a one-standard-deviationnegative export shock, the average firm in the same group receives a flow of top employeesthat is larger than the flow towards the average unaffiliated firm (coefficient 0.032, t-stat

12

2.4). The effect is large, accounting for 20% of the variation of top employment outflows(=0.032/0.16). While the effect of export shocks is statistically significant, the effect ofimport shocks is not significant (although it has the expected sign). This is consistentwith exports being more important than imports for the average firm in our sample asreported before.

3.3 Within-Group Employment Flows

The interaction term in Equation 3.2 depends on the relative profitability shock andthe same-group dummy. We believe that our proxy for profitability shocks (internationaltrade shocks) is plausibly exogenous to the firm. However, because the decision to be partof a business group or a stand-alone firm is a choice variable (at least in the long run),the same-group dummy can be endogenous. To deal in part with this concern, we nextrestrict the sample to business group firms. The idea is that even within the same group,some firms face stronger profitability shocks than others. We study whether, among allaffiliated destination firms, employees move towards those firms facing the most favorableprofitability conditions. We regress:

OutflowSameGroupodt = β(∆pk

dt −∆pkot) + αo + αd + αt + εodt, (3.3)

where firms o and d are in the same business group. We report the results in Table 3.The results indicate that a business group firm sends more employees to the affiliated firmsfacing the most favorable profitability shocks. As before, the effect is increasing in thesegment of the within-firm wage distribution and is statistically significant at the 5% levelfor top-occupation employees. The magnitude of the effects and statistical significanceare comparable to those reported in Table 2.

3.4 Industry-level Shocks

In this section we show that our results are robust to using an alternative measure ofprofitability shocks, namely industry-level changes in Tobin’s Q. This type of shock hasbeen used previously in the literature (Tate and Yang 2015). Although this measure is lessgranular and less exogenous than international trade shocks, it has two advantages. First,it allows us to use a broader sample of firms, rather than restricting our sample to firmsthat engage in international trade. Second, because we have more observations, we canadd firm-pair fixed effects to our core regression. Using firm-pair fixed effects (αo,d) allowsus to control for more unobservables than using origin-firm (αo) and destination-firm fixed

13

effects (αd) separately.We re-estimate Equation (3.2) replacing the relative international trade shocks with

the relative percentage change in Tobin’s Q of the industry of the origin firm and theindustry of the destination firm. We calculate Tobin’s Q of a given industry as theaverage ratio between the market value and the book value of assets for all publiclytraded firms in that industry. The results in Table 4 are similar to our previous results,namely internal labor markets respond to profitability shocks (proxied now with Tobin’sQ), and particularly so for top employees as sen in columns 2 and 4.7

4 Why do Groups use Internal Labor Markets?

In the previous section we showed that business groups use actively internal labor markets,and specially so for top-occupation employees. In this section, we explore the rationalefor internal labor markets. In particular, we study five potential labor market frictionsthat can make reallocation through internal labor markets more convenient than throughexternal labor markets.8

4.1 Transaction Costs

A firm has to pay firing costs (e.g., severance payments) and hiring costs (e.g., head-hunting services, training) when releasing or hiring workers from the external labor mar-ket. A business group firm can save these transaction costs when hiring between firms ofthe same group.9 These costs are typically proportional to wages, so they should be morerelevant for top-occupation workers. However, some of them (e.g., severance payments)are top-coded, which reduces their impact on the firm’s bottom line. These costs can alsobe small relative to the productivity gains of reallocating labor.

Because severance payments are increasing in the tenure of the employee being fired,the transaction costs hypothesis predicts that internal labor mobility within businessgroups should be stronger for employees with longer tenure. To explore this hypothesis,we sort the sample of top-occupation employees according to their tenure in the originfirm. Table 5 reports the results. We re-estimate Equation (3.2) adding a triple interaction

7The broader sample allows us to multiply the number of observations by a factor of 3 (for example,compare column 5 in Table 2 with column 3 in Table 4). There are many more repeated pairs in time,which gives enough variation to identify the pair fixed effects and our variables of interest simultaneosuly.

8Given our previous findings we do not explore theories on the “dark side” of internal labor markets,i.e., cases where internal markets have frictions that external markets do not have.

9Country’s labor laws typically exempt within-group labor adjustments from firing costs (Belenzonand Tsolmon, 2016). This is the case for Chile.

14

term between the relative export shocks, the same-group dummy, and the the tenure ofthe employee. Columns 2 and 3 report the results measuring tenure inside the firm andinside the group, respectively. In both cases the triple interaction term is not statisticallysignificant: internal labor mobility is not more active among employees with long tenure.We also use geographical distance (geodesic) between the origin and destination firm asa proxy for hiring costs. Business groups might have an advantage in bridging betweendistant labor markets. Column 4 shows that internal labor mobility is not increasing ingeographical distance. Overall, this evidence suggests that transaction costs are not themain rationale behind internal labor markets.

4.2 Asymmetric Information

Business-group firms can also have an informational advantage with respect to stand-alone firms. Employees’ true productivity is typically revealed over time. If an employeemoves within the same business group, this information is preserved. If, on the otherhand, an employee moves to or from outside the group, this information is lost or cannotbe credibly conveyed. According to this hypothesis, internal labor mobility should bestronger for employees with longer tenure, for whom the informational advantage of thegroup compared to the market is stronger. Similarly, the asymmetric-information hy-pothesis predicts stronger internal mobility for older employees. Columns 2-3 and column5 in Table 5 report the results for tenure and age, respectively. Neither of these tripleinteraction terms are significant, suggesting that information is not a key driver of theresults.

4.3 Diversification

If frictions in the capital market prevent investors from achieving full diversification,business groups could have an advantage over stand-alone firms by allowing their owners tohold more diversified portfolios. In turn, more diversification allows more risk-taking andfaster response to profitability shocks (Faccio, Marchica, and Mura 2011). If capital andemployment are complementary factors of production, our results of labor mobility insidegroups could be reflecting a friction in the capital market rather than in the labor market.In Table 5, we explore this possibility by using two proxies of corporate diversification:the number of different industrial sectors covered by the business group (column 6) anddispersion of employment between the firms in the business group (column 7). Thesecond proxy can capture the presence of startups, different lines of business within thesame industry or other types of innovative behavior. In both cases, the triple interaction

15

term is not significant, suggesting that our results on internal labor mobility are not areflection of poor diversification in capital markets.

4.4 Imperfect Insurance

If employees have limited access to financial markets, their employers can provide im-plicit unemployment insurance. In our context, business groups can provide workers withgreater job stability within the group, although in exchange for a lower wage. This is theargument used by Cestone, Fumagalli, Kramarz, and Pica (2017) to explain their findingsof labor mobility within French business groups.

We believe that imperfect insurance is not the main mechanism driving our results fortwo reasons. First, lack of insurance should be more relevant for poorer workers (Blun-dell, Pistaferri, and Preston, 2008) and our findings are concentrated on top-occupationemployees. Second, we find that wages of top-occupation employees actually increasewhen they move between firms of the same business group. We re-estimate Equation(3.2) using as dependent variable the log change in wages of the employee that movedwithin a group. We report the results in Table 6. When a group-affiliated firm faces aone-standard-deviation negative relative export shock, the wage of top-occupation em-ployees that move inside the group increases by 6.6 log points (column 5). The wagechange of the employee could reflect an increasing wage path in the receiving firm. Toaddress this concern, we subtract from the wage change of the top-occupation employeethat moves the wage changes of the same type of workers of the receiving firm. The resultremains unchanged. Overall, the evidence is not consistent with the imperfect insurancehypothesis, which predicts that unemployment insurance should come at the cost of lowerwages for employees.

4.5 Intangible Inputs

Intangible inputs such as management practices, knowledge, know-how, organizationalculture and others are key factors of production. For instance, Bloom, Eifert, Mahajan,McKenzie, and Roberts (2013) show that better managerial practices have large effectson productivity. Intangible inputs are hard to transfer between firms, precisely becausethey are intangible. It is hard to pin them down, and they are difficult to quantify.This can explain persistent differences in intangibles across firms (see Bloom and VanReenen (2007) on cross-firm differences in managerial practices). Intangible inputs areoften embedded in the human capital of employees (Eisfeldt and Papanikolaou 2013), sotransferring intangible inputs implies transferring people.

16

Common ownership, which characterizes firms in business groups, can provide anenvironment that facilitates the acquisition and transfer of intangible inputs (Atalay,Hortacsu, and Syverson 2014). We explore two complementary hypotheses —the owner-ship hypothesis and the hierarchy hypothesis– that emphasize different frictions involvedin the relationship with top employees and how business groups can alleviate them.

Along the lines of Grossman and Hart (1986), one can argue that common ownershipallows for the acquisition and transfer of intangible inputs in a simpler way than marketrelationships because of control rights over productive assets. Top employees make non-contractible relationship-specific investments to acquire intangible inputs such as man-agement practices. For instance, employees put effort in training to adjust to the neworganization or in earning the trust of the controlling shareholder. An employee who hasinvested in these intangibles is more valuable to the firm than an average employee fromthe market. Business-groups have an advantage over stand-alone firms because, by virtueof the control rights over multiple firms, they reduce the specificity of the investment.Simply put, intangibles can be used over many more assets. A business group controlsfirms in multiple sectors, firms in different stages of their life-cycle, firms in different mar-kets or countries, so it is reasonable to believe that opportunities for managers will indeedarise. By reducing specificity, business groups provide more incentives for employees toinvest in intangibles in the first place, which overall produces better managers.10

A key prediction of the ownership hypothesis is that the specificity of the investmentleads to the threat of expropriation and hold-up. The controlling shareholder favors theuse of internal employees when facing a profitability shock because they are more valuablethan outside employees. The internal employee knows this, and in renegotiating the wagefor the new position she can extract some of the rents that the shock produces. The factthat wages of top-occupation employees increase when they move between firms of thesame group provides support for this hypothesis (Table 6). The wage increase also givesincentives for acquiring intangibles. If the bargaining power of the controlling shareholderis absolute, then no employee would invest in acquiring intangibles.

An aspect that the ownership hypothesis does not capture is the hierarchical structureof business groups. In most groups, control is exercised in steps or layers. The headquarteror holding company at the top of the pyramidal structure decides major strategic moves(e.g., mergers and acquisitions) while the implementation of the overall strategy andsmaller decisions are taken in firms below in the hierarchy.

According to Garicano and Rossi-Hansberg (2015) the role of such a hierarchy is to10Non-contractibility implies that monetary incentives are often insufficient to insure that these invest-

ments are made (Burkart, Gromb, and Panunzi, 1997).

17

allow employees to better use their knowledge –an intangible input– by delegating andcoordinating with other employees.11 In short, the hierarchy allows employees to leveragetheir knowledge. By providing a large set of assets and production problems across whichto apply knowledge, the hierarchies that characterize business groups give incentives forthe acquisition of intangibles and ultimately produce better managers (Garicano andRossi-Hansberg 2006).12

In Table 7 we show evidence consistent with the view of business groups as knowledgehierarchies. First, in column 2 we find that internal labor mobility for top-occupationemployees is more pronounced when the firm that sends the employee controls the firmthat receives the employee; that is, when the top-occupation employee is being transferredfrom above in the hierarchical structure. This suggests that the employee is a bettermanager since there is positive sorting in hierarchies (Garicano and Rossi-Hansberg 2006).In column 3, we relate our results to the pyramidality of the control structure of the group.In a more pyramidal group there are more links between intermediate firms, instead offirms depending directly from the parent company.13 We find that internal labor mobilityof top employees is more pronounced in these more complex hierarchies (column 3).

The downside of the hierarchy hypothesis for our setup is that there is no explicit rolefor ownership, or rather, it is assumed that all production happens in an environmentof full control rights (i.e., within a firm). There is no difference between having a bighierarchy across firms in a business group and a similarly big hierarchy within a firm.The other problem is that it is not clear if the hierarchy hypothesis can account for thewage dynamics of movers in our setup. Under this hypothesis wage changes are explainedby changes in the structure of the hierarchy (e.g., adding layers to the hierarchy as inCaliendo, Monte, and Rossi-Hansberg 2015) and not by moving within the structure.

5 Firm-level Effects of Internal Labor Mobility

In this section we study the effect of internal labor market mobility on firm-level outcomes.The objective is to see whether the impact of internal labor markets aggregates up tovariables such as firm employment, wages, and productivity. We start with reduced-form evidence. Then we perform a tighter comparison of firms that have received a top-

11See also Hart and Moore (2005) for a related model of hierarchies.12See a model of business groups as knowledge-based hierarchies in Altomonte, Ottaviano, and Rungi

(2018)13We measure group pyramidality as the difference between the control rights and the cash-flow rights

of the controlling shareholder in the average firm of the group (Khanna and Yafeh, 2007). In a standalonefirm this measure is zero. This measure tends to one as the pyramid gets infinitely large.

18

employee from within the group to other firms. The evidence in this section is suggestiveand exploratory in nature. We cannot rule out other mechanisms working simultaneouslywith internal labor markets.

5.1 Reduced-Form Evidence

We study in a reduced-form fashion whether business-group firms respond to profitabilityshocks (i.e., export shocks) differently than stand-alone firms. We focus on wage growthand employment growth. Since most of the firms in our sample are private firms we donot have access to financial information to study other firm-level outcomes. We run firm-level regressions of wage and employment growth on lagged export shocks, the interactionbetween the shocks and the business-group indicator, and firm and year fixed effects. Wereport the results in Table 8. We find that export shocks lead to higher wage growthin group firms and this effect holds for both top- and bottom-occupation workers. Atthe same time, employment growth in group firms responds less to export shocks thanstand-alone firms (column 5). Higher wage growth and lower employment growth suggestincreasing productivity in group firms as a response to profitability shocks. In line with ourresults in column 5, recent literature has shown in a cross-country setting that employmentin business-group firms is less responsive to aggregate shocks than standalone firms (Faccioand O’Brien, 2017). Our findings suggest that internal labor markets can be one reasonfor the differential response of group firms to shocks.

5.2 Synthetic Control Comparisons

We now study whether the arrival of a new employee from within the group creates valuefor the destination firm. If the new employee is equipped with more intangibles, and ifshe is in a top position, she can potentially have an impact on the entire firm. As shownby Bloom, Eifert, Mahajan, McKenzie, and Roberts (2013), better managerial practicescan have a large effect on firm-level productivity.

We implement an event study strategy following the synthetic controls estimationproposed by Abadie and Gardeazabal (2003), and Abadie, Diamond, and Hainmueller(2010). Treated firms are group firms that receive a top-occupation employee from withinthe same group. The synthetic control units are group firms that receive a top employeefrom an unaffiliated firm. We contrast the productivity gains of the treated and controlfirms. We use average payroll per employee as a proxy for the firm’s productivity. Inorder to build our synthetic controls we use a window of three years before receivinga top-occupation worker. The weights in the synthetic controls are defined by a mini-

19

mum distance approach based on firm-level trade shocks, industry, employment level, anddummies for being an exporter-importer.

Panel A of Figure 3 shows that business-group firms receiving a top-occupation em-ployee from the same group experience an increase in wages, while firms receiving a top-occupation employee from outside the group experience a slight decrease. The differenceis roughly 10 percentage points in the period after the arrival of the top employee. Asshown in the previous section, the wage of the top-occupation employee increases after shemoves between firms of the same group. This could mechanically explain the result thatthe average wage of a group firm increases after receiving a top employee. To address thisissue, we re-calculate the average wage of the receiving firm excluding the top employeethat arrived (Panel B). The result remains unchanged. This indicates that the movementof the top employee has a positive spillover on the wages (i.e., productivity) of the otheremployees of the receiving firm.

In Panel C of Figure 3, we report the effects on wages of the group firm that sendsthe top employee (compared to standalone firms that also experience the loss of a topemployee). We find that the average wage of the firm that releases the top employee doesnot change.

In sum, the average wage increases when a group firm receives a top employee fromwithin the group, and the average wage remains unchanged when a group firm releasesa top employee. Because the overall wage of the business group is a weighted average ofthe wages of the different group firms, this evidence suggests that internal reallocation isoverall beneficial for productivity and, hence, for the group as a whole.

6 Conclusions

In this paper we use a matched employer-employee dataset for Chile to study labor marketmobility inside business groups. We start the analysis by providing evidence of uncondi-tional (i.e., unrelated to shocks) internal labor mobility. We find that labor reallocationis stronger between pairs of firms in the same group that between pairs of unaffiliatedfirms. The effects are concentrated on top-occupation employees defined as those at thetop of the within-firm wage distribution.

We test whether the direction of the employment flows follows changing businessconditions. We measure these changes using firm-level international trade shocks. Weshow that in response to negative profitability shocks, group firms adjust by sendingmore workers to affiliated firms than to unaffiliated firms. The effects are stronger whenthe employee’s origin firm controls the destination firm, and in more complex ownership

20

structures. Wages of top employees increase as they move within the group. These resultsare consistent with the idea that common ownership facilitates the acquisition and transferof intangible inputs (e.g., management practices).

Finally, we looked at the effect on average wages, our proxy of productivity, on the firmreceiving the employee and on the firm sending the employee. To a first order, the wagedifference between the destination and the origin firms reflects the effects of reallocationat the business-group level. We find a positive effect on the destination-firm wage andno effect on the origin-firm wage, suggesting that reallocation is beneficial for the groupsince there is an overall increase in productivity. Overall, our findings are consistent withthe idea that internal labor markets add value by actively reallocating scarce resourcesacross affiliated firms.

21

ReferencesAbadie, Alberto, Alexis Diamond, and Jens Hainmueller, 2010, Synthetic control meth-

ods for comparative case studies: Estimating the effect of California’s tobacco controlprogram, Journal of the American Statistical Association 105, 493–505. 5.2

Abadie, Alberto, and Javier Gardeazabal, 2003, The economic costs of conflict: A casestudy of the Basque country, American Economic Review 93, 113–132. 5.2

Almeida, Heitor, Chang Soo Kim, and Hwanki Brian Kim, 2015, Internal capital marketsin business groups: Evidence from the Asian financial crisis, Journal of Finance 70,2539–2586. 1

Altomonte, Carlo, Gianmarco Ottaviano, and Armando Rungi, 2018, Business groups asknowledge-based hierarchies of firms, Working paper. 12

Atalay, Enghin, Ali Hortacsu, and Chad Syverson, 2014, Vertical integration and inputflows, American Economic Review 104, 1120–1148. 1, 4.5

Autor, David H., David Dorn, and Gordon H. Hanson, 2013, The China syndrome: Locallabor market effects of import competition in the united states, American EconomicReview 103, 2121–68. 1, 2.3

Belenzon, Sharon, and Ulya Tsolmon, 2016, Market frictions and the competitive ad-vantage of internal labor markets, Strategic Management Journal 37, 1280–1303. 1,9

Bloom, Nicholas, Ben Eifert, Aprahit Mahajan, David McKenzie, and John Roberts, 2013,Does management matter? Evidence from India, Quarterly Journal of Economics 128,1–51. 4.5, 5.2

Bloom, Nicholas, and John Van Reenen, 2007, Measuring and explaining managementpractices across firms and countries, Quarterly Journal of Economics 122, 1351–1408.1, 4.5

Blundell, Richard, Luigi Pistaferri, and Ian Preston, 2008, Consumption inequality andpartial insurance, American Economic Review 98, 1887–1921. 1, 4.4

Boutin, Xavier, Giacinta Cestone, Chiara Fumagalli, Giovanni Pica, and Nicolas Serrano-Velarde, 2013, The deep-pocket effect of internal capital markets, Journal of FinancialEconomics 109, 122–145. 1

Buchuk, David, Borja Larrain, Francisco Munoz, and Francisco Urzua, 2014, The inter-nal capital markets of business groups: Evidence from intra-group loans, Journal ofFinancial Economics 112, 190–212. 1, 2.2

Burkart, Mike, Denis Gromb, and Fausto Panunzi, 1997, Large shareholders, monitoring,and the value of the firm, Quarterly Journal of Economics 112, 693–728. 10

22

Caliendo, Lorenzo, Ferdinando Monte, and Esteban Rossi-Hansberg, 2015, The anatomyof French production hierarchies, Journal of Political Economy 123, 809–852. 4.5

Card, David, Jorg Heining, and Patrick Kline, 2013, Workplace heterogeneity and the riseof west German wage inequality, Quarterly Journal of Economics 128, 967–1015. 2.1,A.1

Cestone, Giacinta, Chiara Fumagalli, Francis Kramarz, and Giovanni Pica, 2017, Insur-ance between firms: The role of internal labor markets, Working paper. 1, 3.2, 4.4

Donelli, Marcelo, Borja Larrain, and Francisco Urzua, 2013, Ownership dynamics withlarge shareholders: An empirical analysis, Journal of Financial and QuantitativeAnalysis 48, 579–609. 2.2

Eisfeldt, Andrea, and Dimitris Papanikolaou, 2013, Organization capital and the cross-section of expected returns, Journal of Finance 68, 1365–1406. 1, 4.5

Faccio, Mara, Maria-Teresa Marchica, and Roberto Mura, 2011, Large shareholder diver-sification and corporate risk-taking, Review of Financial Studies 24, 3601–3641. 4.3

Faccio, Mara, and William O’Brien, 2017, Business groups and employment, Workingpaper. 1, 5.1

Garicano, Luis, and Esteban Rossi-Hansberg, 2006, Organization and inequality in aknowledge economy, Quarterly Journal of Economics 121, 1383–1435. 4.5

, 2015, Knowledge-based hierarchies: Using organizations to understand the econ-omy, Annual Review of Economics 7, 1–30. 1, 4.5

Giroud, Xavier, and Holger Mueller, 2015, Capital and labor reallocation within firms,Journal of Finance 70, 1767–1804. 1

Gopalan, Radha, Vikram Nanda, and Amit Seru, 2007, Affiliated firms and financialsupport: Evidence from Indian business groups, Journal of Financial Economics 86,759–795. 1

Grossman, Sanford J., and Oliver D. Hart, 1986, The costs and benefits of ownership: Atheory of vertical and lateral integration, Journal of Political Economy 94, 691–719. 1,4.5

Hart, Oliver D., 2001, Financial contracting, Journal of Economic Literature 39, 1079–1100. 1

, and John Moore, 2005, On the design of hierarchies: Coordination versus spe-cialization, Journal of Political Economy 113, 675–702. 1, 11

Hummels, David, Rasmus Jørgensen, Jakob Munch, and Chong Xiang, 2014, The wageeffects of offshoring: Evidence from Danish matched worker-firm data, AmericanEconomic Review 104, 1597–1629. 1, 2.3

23

Khanna, Tarun, and Yishay Yafeh, 2007, Business groups in emerging markets: Paragonsor parasites?, Journal of Economic Literature 45, 331–372. 1, 13

Modigliani, Franco, and Merton H Miller, 1958, The cost of capital, corporation financeand the theory of investment, American Economic Review 48, 261–297. 3.2

Morck, Randall, Daniel Wolfenzon, and Bernard Yeung, 2005, Corporate governance,economic entrenchment, and growth, Journal of Economic Literature 43, 655–720. 1

Silva, Rui, 2017, Internal labor markets, wage convergence and investment, Working Pa-per. 1

Song, Jae, David J. Price, Fatih Guvenen, Nicholas Bloom, and Till Von Wachter, 2015,Firming up inequality, Discussion paper, National Bureau of Economic Research. A.1

Stein, Jeremy C., 1997, Internal capital markets and the competition for corporate re-sources, Journal of Finance 52, 111–133. 3.2

Tate, Geoffrey, and Liu Yang, 2015, The bright side of corporate diversification: Evidencefrom internal labor markets, Review of Financial Studies 28, 2203–2249. 1, 3.4

24

Appendix

A.1 Data AppendixThis appendix describes in detail the limitations of the datasets used in the paper andthe cleaning procedures and merging procedures used.

Limitations of Unemployment Insurance Dataset. Wage records are subject tothree important limitations. First, they include only employees who have formal contractswith employers. According to standard estimates of the main employment survey in Chilecarried out by the Institute of National Statistics (INE, for its acronym in Spanish), around83% of dependent employment is formal in Chile, one of the highest in the region. Second,they include only workers who signed their contracts after 2002. Thus, workers with formalcontracts that either (i) started working after 2002, (ii) changed formal jobs after 2002and/or (iii) renegotiated any contract characteristic after 2002, should be in the dataset.Since 2002, the coverage of the dataset has been increasing. In fact, aggregate employmentthat appears in the UI dataset in 2015, i.e., around 4.6 million, is less than the populationthat contributes to pension funds, 5.1 million, which is not affected by the constraintsof the UI dataset. Nevertheless, employees contributing to pension funds include publicemployees, which are around 600,000. Thus, by 2015, aggregate employment of the UIdataset is approximately the same as the one reported by pension funds administrators.Third, wages are top-coded at a certain threshold. Around 4% of workers were top-coded in their main jobs in 2015, which represents around 20% of firms’ aggregate wageexpenditures.14

Cleaning Process of Unemployment Insurance Dataset. We implement a set ofcleaning filters to the dataset that are standard in the literature (Song, Price, Guvenen,Bloom, and Von Wachter 2015; Card, Heining, and Kline 2013). First, we drop firmsthat have less than 5 employees. These represent around 2/3 of the total number of firmsand represent around 8% of total employment in 2015. We drop these firms because theidentification of a firm in the UI dataset is done through tax IDs. Thus, there might beconcerns that tax IDs with few employees might not be firms but, e.g., shell corporationsfor tax purposes. This decision also helps to make more meaningful the statistics ofwithin-firm distribution of workers and it follows the restrictions applied in matchedemployer-employee datasets. Second, we focus on spells at the highest-paying jobs foreach individual in each year. This implies dropping around 17% of total job spells peryear. Regarding workers that have several jobs, the income of their main jobs representsaround 90% of their total labor income across jobs in 2015. Third, we exclude workersthat have a weak labor attachment. That is, we exclude workers that earn less than aquarter of the annual equivalent of the minimum wage. These workers represent around15% of employment and around 1% of firms’ aggregate wage expenditures in 2015. Giventhese restrictions, we end up with a dataset of around 100,000 employers and 4.6 millionemployees.

14This number represents a lower bound due to the fact that wages are censored for top-coded workers.

25

Business Groups Dataset Limitations. Two caveats about the ownership data are inorder. First, our business group data excludes financial firms (commercial banks, mutualfund companies, pension fund administrators, etc.). This is not a big omission for ourpurposes because ties between business groups and banks have been limited by regulationsince the aftermath of the debt crisis in the 1980s. Second, although we are able to identifythe private firms that belong to each group, we do not know the direct ownership stakebetween pairs of private firms in the group. For example, if X is a listed firm, and Y andZ are private firms that consolidate with X, then we know that X, Y, and Z belong tothe same group. We also know the stake between X and Y. However, we do not know thestake between Y and Z.

Merge of UI Dataset with Business Group Dataset. As with the employer-employee dataset, firms in the business groups data are identified with unique tax IDs.We use these unique identifiers to merge both datasets. From the business group dataset,around 45% of firms are on average across years in the employer-employee dataset. Fromthe aforementioned cleaned employer-employee dataset, around 0.3% of firms across yearsare in the business group dataset. These firms represent on average about 3% and 5%of aggregate employment and wage bill, respectively. At the same time, these firms rep-resent close to 80% of the stock market capitalization of the Chilean market, and theirtotal revenue over GDP is almost 60%.

COMTRADE dataset. This dataset is organized and cleaned by the Centre d’EtudesProspectives et d’Informations Internationales (CEPII, for its acronym in French). CEPIIorganizes COMTRADE’s database into BACI, which is a cleaned version of COMTRADE’sdatabase. This dataset, identifies the value and quantity for each 6-digit HS product andcountry that is traded globally. This dataset is merged to the firm-level internationaltrade dataset using unique HS codes and country IDs.

Merge of Trade with UI Dataset. Once the international price shocks are createdat the firm level, the international trade dataset is merged at the firm level with the UIand BG data using firms’ unique tax IDs.15 From the trade dataset, around 45% of firmsthat either export or import are merged with the UI dataset on average across years.These firms that are merged account for around 98% and 95% of export and importflows, respectively. From the cleaned UI dataset, around 4% and 15% of firms export andimport, respectively. Among firms that are merged, importers account for around 38%and 50% of employment and aggregate wage bill, respectively, whereas exporters accountfor 15% and 25%, respectively. Among firms that appear in both the UI and trade dataset,73% are only importers, 6% are only exporters and 21% do both. Firms that do bothactivities account for 97% and 72% of aggregate exports and imports, respectively.

15It is important to note that, due to confidentiality restrictions, this merge is implemented by theSecretary of Labor, which manages the UI dataset, and has an agreement with Customs to use theInternational Trade data. The authors in this paper only have access to the merged dataset once all taxIDs are erased and replaced with fake IDs. This guarantees anonymity and the confidentiality of thedataset.

26

Figure 1: Example of Business Group Ownership Structure: Antarchile

Notes: This figure presents the ownership structure of Antarchile, one of the largest businessgroups in Chile, controlled by the Angelini family.

27

Figure 2: Employment Flows Between Pairs of Affiliated and Unaffiliated Firms

0

.1

.2

.3

All Top

Affiliated Non-Affiliated

Notes: This figure plots the average employment flow between pairs of affiliated and unaffiliatedfirms. Two firms are affiliated if they belong to the same business group. All refers to anyemployee, while Top refers to top-occupation employees defined as the top quartile of the within-firm distribution. Employment outflows are defined as the number of employees moving from a firmof origin to a firm of destination, relative to the total number of employees released by the originfirm that year. We restrict the sample to employment outflows in which the firm of origin of theemployee belongs to a business group. Bands represent 5% confidence intervals around averages.

28

Figure 3: Impact of Receiving and Sending a Top-Occupation Employee

7.5

7.55

7.6

7.65

7.7

7.75

-2 -1 0 1 2 3Years since treatment

Treated Firms Synthetic Control Firms

A. Receiver: Payroll per emp

7.5

7.6

7.7

7.8

7.9

-2 -1 0 1 2 3Years since treatment

Treated Firms Synthetic Control Firms

B. Receiver’s Stayers: Payroll per emp

8

8.025

8.05

8.075

8.1

-2 -1 0 1 2 3Years since treatment

Treated Firms Synthetic Control Firms

C. Sender: Payroll per emp

Notes: This figure shows the logarithm of average payroll per employee for firms in businessgroups and for synthetic controls. Treated firms correspond to group firms that receive a top-occupation employee from the same business group (Panels A and B), or group firms that send atop-occupation employee to other firms (Panel C). In Panels A and C, average payroll per employeeis computed using all the employees of the firm, while in Panel B it is computed using employeesthat were at the firm before the treatment (stayers). The choice of synthetic controls is explainedin the main text.

29

Table 1: Labor Markets Flows: Affiliated vs Unaffiliated Firm-PairsAll Workers All Workers Bot. 25 Mid. 50 Top 25

(1) (2) (3) (4) (5)Same BG 0.024∗∗∗ 0.022∗∗∗ 0.019∗∗∗ 0.023∗∗ 0.047∗∗∗

(0.006) (0.006) (0.007) (0.010) (0.015)∆ Log Avg. Wage 0.000 -0.002 0.003∗ -0.017∗∗

(0.001) (0.001) (0.002) (0.008)Log Employment Origin -0.046∗∗∗ -0.085∗∗∗ -0.074∗∗∗ -0.159∗∗∗

(0.009) (0.017) (0.012) (0.030)Log Employment Destination -0.002∗ -0.004∗∗ -0.006∗ -0.023∗∗

(0.001) (0.002) (0.003) (0.011)Log Emp. BG Origin -0.103 -0.297∗ -0.206 0.235

(0.082) (0.178) (0.241) (0.348)Log Emp. BG Dest. -0.118∗∗ -0.295∗∗∗ -0.122 -0.286

(0.048) (0.101) (0.143) (0.201)Log N. Firms BG Origin 0.061∗∗ 0.087∗ 0.084∗∗ 0.285∗

(0.030) (0.045) (0.032) (0.154)Log N. Firms BG Dest. 0.011 0.017 0.023 -0.036

(0.016) (0.021) (0.032) (0.161)R2 0.522 0.594 0.680 0.638 0.694Origin FE X X X X XDestination FE X X X X XYear FE X X X X XMean DV .02 .02 .037 .047 .12SD DV .043 .042 .066 .084 .158N 21269 20424 9267 7612 2493

Notes: This table reports the results from the regression presented in equation (3.1). The depen-dent variable is the employment flows for all workers (column 1 and 2), for the bottom quartileworkers in the within-firm wage distribution of the origin firm (column 3), for the middle twoquartile of workers in the within-firm wage distribution (column 4), for the top quartile workers inthe within-firm wage distribution (column 5). Same BG is a dummy that takes the value of onefor those pairs where the origin and the destination are part of the same business group. ∆ LogAvg. Wage is the difference in the logarithm of average wage between destination and origin. LogEmployment Origin (Destination) is the logarithm of the total number of employees in the firmof origin (destination). Log Emp. BG Origin (Destination) is the logarithm of the total numberof employees in the business group of the firm of origin (destination). Log N. Firms BG Origin(Destination) is the logarithm of the total number of firms that are part of the business group ofthe firm of origin (destination). Robust standard errors in parentheses are double clustered at thelevel of the firm of origin and destination of the flow. *** p<0.01, ** p<0.05, * p<0.1.

30

Table 2: Internal Labor Mobility TestAll Workers Bot. 25 Mid. 50 Top 25

(1) (2) (3) (4) (5)∆ Exports x Same BG 0.004 0.004∗ -0.000 0.008 0.032∗∗

(0.002) (0.002) (0.002) (0.005) (0.013)∆ Imports x Same BG -0.004 -0.004 -0.009 -0.015∗ -0.004

(0.004) (0.004) (0.007) (0.009) (0.012)∆ Exports 0.000 0.001 0.001 0.002 0.006

(0.001) (0.001) (0.001) (0.002) (0.004)∆ Imports -0.001 -0.001 0.001 -0.001 -0.002

(0.001) (0.001) (0.001) (0.002) (0.002)Same BG 0.024∗∗∗ 0.022∗∗∗ 0.018∗∗∗ 0.024∗∗ 0.044∗∗∗

(0.006) (0.006) (0.007) (0.010) (0.016)Log Avg. Wage Origin 0.003 0.016 -0.012 0.046

(0.018) (0.018) (0.017) (0.080)Log Avg. Wage Destination 0.000 -0.001 -0.002 -0.057

(0.003) (0.005) (0.008) (0.038)Log Employment Destination -0.002∗ -0.004∗∗ -0.006∗ -0.022∗∗

(0.001) (0.002) (0.003) (0.010)Log Employment Origin -0.046∗∗∗ -0.086∗∗∗ -0.074∗∗∗ -0.158∗∗∗

(0.009) (0.017) (0.012) (0.030)Log Emp. BG Origin -0.083 -0.281 -0.149 0.301

(0.079) (0.175) (0.248) (0.326)Log Emp. BG Dest. -0.119∗∗ -0.287∗∗∗ -0.137 -0.209

(0.049) (0.099) (0.143) (0.215)Log N. Firms BG Origin 0.068∗ 0.098∗ 0.081∗∗ 0.390∗∗

(0.037) (0.056) (0.039) (0.168)Log N. Firms BG Dest. 0.013 0.018 0.030 -0.029

(0.016) (0.021) (0.032) (0.171)R2 0.522 0.594 0.680 0.639 0.697Origin FE X X X X XDestination FE X X X X XYear FE X X X X XMean DV .02 .02 .037 .047 .122SD DV .043 .042 .066 .084 .16Mean Den. DV 1.2 1.2 1.2 1.3 1.1N 21009 20188 9184 7523 2457

Notes: This table reports the results from the regression presented in equation (3.2). ∆ Exports(∆ Imports) is defined as the first difference in firm-level export (imports) prices between twoconsecutive years and between the destination and the origin as described in section 2.3. See table1 for more details about variable definitions. Robust standard errors in parentheses are doubleclustered at the level of the firm of origin and destination of the flow. *** p<0.01, ** p<0.05, *p<0.1.

31

Table 3: Within-Group Employment FlowsAll Workers Bot. 25 Mid. 50 Top 25

(1) (2) (3) (4) (5)∆ Exports 0.004∗ 0.004∗ 0.001 0.013∗ 0.035∗∗

(0.002) (0.002) (0.001) (0.007) (0.017)∆ Imports -0.005 -0.005 -0.005 -0.006 -0.011

(0.004) (0.003) (0.004) (0.010) (0.016)Log Avg. Wage Origin 0.043 0.140∗∗ 0.088 0.293

(0.040) (0.062) (0.072) (0.207)Log Avg. Wage Destination -0.139∗ -0.064 -0.427∗∗∗ -0.470∗∗

(0.073) (0.039) (0.147) (0.185)Log Employment Origin -0.052∗∗∗ -0.165∗∗∗ -0.055∗ -0.127∗∗

(0.016) (0.039) (0.029) (0.048)Log Employment Destination -0.024 -0.031 -0.079∗∗ -0.078∗∗∗

(0.017) (0.021) (0.038) (0.021)Log Emp. BG -0.379 -0.149 -0.199 1.890

(0.370) (0.620) (0.855) (1.374)Log N. Firms BG 0.065 0.131 -0.232 0.175

(0.204) (0.240) (0.440) (0.494)R2 0.505 0.592 0.797 0.651 0.756Origin FE X X X X XDestination FE X X X X XYear FE X X X X XMean DV .05 .05 .08 .096 .187SD DV .084 .084 .115 .134 .213N 576 576 255 234 218

Notes: This table reports the results from the regression presented in equation (3.2) restrictingthe sample to only those flows that are within the same business group. ∆ Exports (∆ Imports) isdefined as the first difference in firm-level export (imports) prices between two consecutive yearsand between the destination and the origin as described in section 2.3. See table 1 for more detailsabout variable definitions. Robust standard errors in parentheses are double clustered at the levelof the firm of origin and destination of the flow. *** p<0.01, ** p<0.05, * p<0.1.

32

Table 4: Internal Labor Mobility Test Using Industry-level ShocksAll Workers Top 25

(1) (2) (3) (4)∆Q-Tobin x Same BG 0.021 0.024 0.063∗∗ 0.057∗∗

(0.015) (0.017) (0.028) (0.024)∆Q-Tobin 0.000 -0.000 0.007 0.005

(0.001) (0.001) (0.006) (0.006)Same BG 0.039∗∗∗ 0.059∗∗∗

(0.005) (0.015)Log Avg. Wage Origin 0.053∗∗∗ 0.037∗∗ 0.093∗∗ 0.116

(0.014) (0.014) (0.037) (0.067)Log Avg. Wage Destination -0.001 -0.003 -0.033∗ -0.064∗∗

(0.003) (0.005) (0.018) (0.026)Log Employment Destination -0.002∗∗∗ -0.004∗∗ -0.015∗∗∗ -0.023∗∗

(0.001) (0.001) (0.005) (0.008)Log Employment Origin -0.052∗∗∗ -0.035∗∗∗ -0.139∗∗∗ -0.115∗∗∗

(0.006) (0.004) (0.014) (0.016)Log Emp. BG Origin 0.030 0.041 0.353∗∗ 0.661∗∗∗

(0.042) (0.022) (0.173) (0.172)Log Emp. BG Dest. -0.170∗∗∗ -0.168∗∗ -0.339 -0.305

(0.055) (0.058) (0.221) (0.315)Log N. Firms BG Origin 0.028 0.017 0.003 -0.055

(0.027) (0.011) (0.096) (0.077)Log N. Firms BG Dest. 0.027 0.063 -0.134 0.087

(0.027) (0.038) (0.200) (0.229)R2 0.666 0.743 0.755 0.815Pair FE X XOrigin FE X XDestination FE X XYear FE X X X XMean DV .028 .025 .136 .124SD DV .075 .07 .205 .199Mean Den. DV 1.3 2 1.3 1.6N 50591 24041 7427 3023

Notes: This table reports the results from the regression presented in equation (3.2). ∆ Q-Tobinis defined as the percentage difference in the measure of Q-Tobin between two consecutive yearsand between the destination and origin industries. Q-Tobin is defined as the ratio between themarket and book value of assets at the industry level of listed firms. See table 1 for more detailsabout variable definitions. Robust standard errors in parentheses are double clustered at the levelof the firm of origin and destination of the flow. *** p<0.01, ** p<0.05, * p<0.1.

33

Table 5: Why do Groups use Internal Labor Markets?Baseline Tenure BG Tenure Distance Age BG N. Sectors BG Emp. Dispersion

(1) (2) (3) (4) (5) (6) (7)∆ Exports x Same BG x Z -0.008 -0.010 0.013 -0.096∗ -0.011 -0.049

(0.015) (0.014) (0.085) (0.052) (0.018) (0.033)∆ Exports x Same BG 0.032∗∗ 0.041∗∗ 0.043∗∗ 0.029∗ 0.382∗∗ 0.048 0.106∗∗

(0.013) (0.020) (0.020) (0.017) (0.191) (0.030) (0.053)∆ Exports 0.006 0.012 0.012 0.008 0.004 0.005 0.006

(0.004) (0.008) (0.008) (0.005) (0.055) (0.005) (0.004)Same BG 0.044∗∗∗ 0.022 0.029 0.044∗∗ 0.104 0.000 0.001

(0.016) (0.027) (0.031) (0.019) (0.227) (0.061) (0.084)Log Employment Destination -0.022∗∗ -0.022∗∗ -0.023∗∗ -0.022∗∗ -0.022∗∗ -0.022∗∗ -0.022∗∗

(0.010) (0.010) (0.010) (0.011) (0.010) (0.010) (0.010)Log Employment Origin -0.158∗∗∗ -0.157∗∗∗ -0.157∗∗∗ -0.158∗∗∗ -0.156∗∗∗ -0.159∗∗∗ -0.157∗∗∗

(0.030) (0.031) (0.031) (0.030) (0.030) (0.031) (0.030)Log Emp. BG Origin 0.301 0.323 0.324 0.307 0.289 0.307 0.287

(0.326) (0.332) (0.332) (0.326) (0.320) (0.327) (0.324)Log Emp. BG Dest. -0.209 -0.227 -0.188 -0.187 -0.190 -0.257 -0.163

(0.215) (0.203) (0.199) (0.211) (0.207) (0.225) (0.216)Log N. Firms BG Origin 0.390∗∗ 0.385∗∗ 0.385∗∗ 0.391∗∗ 0.399∗∗ 0.385∗∗ 0.394∗∗

(0.168) (0.172) (0.171) (0.168) (0.166) (0.166) (0.169)Log N. Firms BG Dest. -0.029 -0.007 -0.011 -0.042 -0.033 -0.100 -0.059

(0.171) (0.175) (0.174) (0.164) (0.174) (0.152) (0.151)R2 0.697 0.700 0.700 0.699 0.699 0.700 0.699Origin FE X X X X X X XDestination FE X X X X X X XYear FE X X X X X X XMean DV .122 .122 .122 .122 .122 .122 .122SD DV .16 .16 .16 .16 .16 .16 .16Mean Den. DV 1.1 1.1 1.1 1.1 1.1 1.1 1.1N 2457 2457 2457 2457 2457 2457 2457

Notes: This table reports the heterogeneous effects for the regression presented in equation (3.2).The dependent variable is employment flows for the top quartile workers in the within-firm wagedistribution. The variables used for the triple interactions (Z) are Tenure (tenure of the workerin the origin firm), BG Tenure (tenure of the worker in the business group of the origin firm),Distance (geodesic distance between the origin and destination), Age (age of the worker), BG N.Sectors (number of different sectors that covers the business group of the firm of origin), and BGEmp. Dispersion (employment dispersion between the firms in the Business group of the originfirm). ∆ Exports (∆ Imports) is defined as the first difference in firm-level export (imports) pricesbetween two consecutive years and between the destination and the origin as described in section2.3. See table 1 for more details about variable definitions. Robust standard errors in parenthesesare double clustered at the level of the firm of origin and destination of the flow. *** p<0.01, **p<0.05, * p<0.1.

34

Table 6: Job-to-Job Wage Changes for MoversAll Workers Bot. 25 Mid. 50 Top 25 Top 25 vs Stayers Dest.

(1) (2) (3) (4) (5) (6)∆ Exports x Same BG 0.002 0.003 -0.013 0.007 0.066∗∗∗ 0.048∗∗

(0.008) (0.008) (0.012) (0.013) (0.021) (0.023)∆ Exports -0.001 -0.002 0.005 -0.002 -0.009 -0.002

(0.003) (0.003) (0.005) (0.005) (0.008) (0.009)Same BG -0.014 -0.018 -0.026 0.012 0.085∗ 0.075

(0.024) (0.024) (0.038) (0.041) (0.046) (0.052)Log Avg. Wage Origin -0.446∗∗∗ -0.611∗∗∗ -0.538∗∗∗ -0.449∗∗∗ -0.441∗∗∗

(0.031) (0.044) (0.047) (0.056) (0.054)Log Avg. Wage Destination 0.107∗∗ 0.042 0.060 0.073 0.318∗∗∗

(0.048) (0.035) (0.043) (0.083) (0.090)Log Employment Destination -0.005 -0.050∗∗ -0.015 -0.007 -0.049

(0.015) (0.022) (0.017) (0.027) (0.030)Log Employment Origin -0.059∗∗∗ -0.036∗∗ -0.044∗∗∗ -0.055∗∗ -0.050∗

(0.008) (0.013) (0.011) (0.026) (0.026)Log Emp. BG Origin 0.363 0.053 0.449 -0.664 -0.475

(0.326) (0.384) (0.442) (0.630) (0.754)Log Emp. BG Dest. 0.350 0.374 0.274 -0.628 0.667

(0.320) (0.388) (0.420) (0.500) (0.466)Log N. Firms BG Origin -0.004 -0.051 -0.136 -0.070 0.237

(0.179) (0.148) (0.184) (0.358) (0.353)Log N. Firms BG Dest. 0.084 0.163 0.264 0.372 0.188

(0.238) (0.206) (0.216) (0.430) (0.430)R2 0.449 0.457 0.527 0.566 0.523 0.547Origin FE X X X X X XDestination FE X X X X X XYear FE X X X X X XMean DV .2 .2 .243 .176 .134 .087SD DV .408 .407 .428 .404 .355 .373N 23372 22343 7384 9273 2604 2512

Notes: This table reports the results from the regression presented in equation (3.2). The de-pendent variable is the log difference of wages for the employees that move (movers) between thedestination and the origin firm. Column 6 has the same dependent variable as column 5, with theexception that it subtracts the log difference of wages for stayers in the destination firm. This isdone to address the concern of mechanical wage gains for movers due to wage gains that the des-tination firm is experiencing. ∆ Exports (∆ Imports) is defined as the first difference in firm-levelexport (imports) prices between two consecutive years and between the destination and the originas described in section 2.3. See table 1 for more details about variable definitions. Robust standarderrors in parentheses are double clustered at the level of the firm of origin and destination of theflow. *** p<0.01, ** p<0.05, * p<0.1.

35

Table 7: Internal Labor Markets and Ownership StructureBaseline Origin Controls Dest. BG Pyramidality

(1) (2) (3)∆ Exports x Same BG x Z 0.424∗∗∗ 0.510∗∗∗

(0.107) (0.156)∆ Exports x Same BG 0.032∗∗ 0.038∗∗ 0.019

(0.013) (0.015) (0.012)∆ Exports 0.006 0.007 0.006

(0.004) (0.004) (0.004)Same BG 0.044∗∗∗ 0.022 0.035∗∗

(0.016) (0.021) (0.017)Log Employment Destination -0.022∗∗ -0.017 -0.023∗∗

(0.010) (0.010) (0.011)Log Employment Origin -0.158∗∗∗ -0.162∗∗∗ -0.158∗∗∗

(0.030) (0.031) (0.030)Log Emp. BG Origin 0.301 0.324 0.308

(0.326) (0.332) (0.327)Log Emp. BG Dest. -0.209 -0.112 -0.152

(0.215) (0.213) (0.188)Log N. Firms BG Origin 0.390∗∗ 0.390∗∗ 0.384∗∗

(0.168) (0.169) (0.172)Log N. Firms BG Dest. -0.029 -0.051 -0.101

(0.171) (0.187) (0.182)R2 0.697 0.713 0.703Origin FE X X XDestination FE X X XYear FE X X XMean DV .122 .122 .122SD DV .16 .16 .16Mean Den. DV 1.1 1.1 1.1N 2457 2457 2457

Notes: This table reports the heterogeneous effects for the regression presented in equation (3.2).The dependent variable is employment flows for the top quartile workers in the within-firm wagedistribution. The variables used for the triple interactions (Z) are Origin Controls Dest. (adummy that takes the value one if the origin firms has an ownership stake greater than 50% of thedestination firm) and BG Pyramidality (a dummy for business groups with high pyramidality). ∆Exports (∆ Imports) is defined as the first difference in firm-level export (imports) prices betweentwo consecutive years and between the destination and the origin as described in section 2.3. SeeTable 1 for more details about variable definitions. Robust standard errors in parentheses aredouble clustered at the level of the firm of origin and destination of the flow. *** p<0.01, **p<0.05, * p<0.1.

36

Table 8: The Effect of Export Shocks on Firm-Level Wages and Employment∆ Wages(t) ∆ Employment(t)

(1) (2) (3) (4) (5)All Bot. 25 Mid. 50 Top 25

∆Export(t) x BG 0.004 0.002 0.004 0.002 -0.016(0.004) (0.009) (0.005) (0.006) (0.017)

∆Export(t-1) x BG 0.017∗∗∗ 0.016∗ 0.019∗∗∗ 0.015∗∗∗ -0.040∗

(0.005) (0.010) (0.006) (0.005) (0.021)∆Export(t-2) x BG 0.011∗∗ 0.005 0.015∗∗ 0.009 -0.043∗

(0.006) (0.011) (0.006) (0.006) (0.023)∆Export(t-3) x BG 0.010∗ 0.011 0.008 0.011∗∗ -0.041∗∗

(0.006) (0.010) (0.008) (0.004) (0.021)∆Export(t) -0.000 0.003 -0.001 0.000 0.003

(0.002) (0.004) (0.002) (0.003) (0.005)∆Export(t-1) -0.004∗ -0.002 -0.005∗ -0.003 0.006

(0.002) (0.004) (0.003) (0.003) (0.007)∆Export(t-2) -0.006∗∗∗ -0.006 -0.007∗∗ -0.005∗ 0.009

(0.002) (0.004) (0.003) (0.003) (0.006)∆Export(t-3) -0.005∗∗∗ -0.005 -0.005∗∗∗ -0.004∗ 0.005

(0.002) (0.004) (0.002) (0.003) (0.005)R2 0.173 0.103 0.147 0.176 0.256Year FE X X X X XFirm FE X X X X XMean DV .035 .026 .036 .039 .068SD DV .117 .219 .149 .129 .304N 6359 6359 6359 6359 6359

Notes: This table reports firm-level panel regressions for annual (log) wage growth of employeesin different segments of the wage distribution (All, Bottom 25, Middle 50, Top 25) and totalemployment. BG is a dummy for those firms that are part of a business group. Robust standarderrors in parentheses are clustered at the firm level. *** p<0.01, ** p<0.05, * p<0.1.

37

Online Appendix

Figure A.1: Correlation Between Initial Market Shares and Future Price Changes

-.2-.1

0.1

.2

0 .2 .4 .6 .8 1Global Market Exposure (imports)

A. Imports

-.20

.2.4

0 .2 .4 .6 .8 1Global Market Exposure (exports)

B. Exports

Notes: This figure shows the scatter plot between the initial share exposure (skij0) measured in

2004 and future price changes between 2005 and 2015. See section 2.3 for more details.

38

Figure A.2: Industry Distribution for Different Samples

0 10 20 30 40%

Utilities

Transports and Telecomms

Services

Mining

Manufacturing

Finance

Construction

Commerce

Agriculture

UI UI+BG UI+BG+Trade

Notes: This figure presents the industry distribution for the full sample of UI, the matchedsampled between UI and business group firms (UI-BG), and the matched sample between UI,business group firms, and trade data (UI-BG-Trade).

39

Table A.1: Employment flows between pairs of affiliated and unaffiliated firmsAll Employees

All Firms Exp. & Imp.

Affiliated Unaffiliated DifferenceP-value Affiliated Unaffiliated Difference

P-value(1) (2) (3) (4) (5) (6)

Outflows 0.115 0.021 0.094*** 0.053 0.019 0.034***(0.004) (0.000) [0.000] (0.004) (0.000) [0.000]

Inflows 0.098 0.021 0.077*** 0.043 0.018 0.026***(0.004) (0.000) [0.000] (0.003) (0.000) [0.000]

Top EmployeesAll Firms Exp. & Imp.

Affiliated Unaffiliated DifferenceP-value Affiliated Unaffiliated Difference

P-value(1) (2) (3) (4) (5) (6)

Outflows 0.294 0.112 0.182*** 0.198 0.115 0.082***(0.010) (0.001) [0.000] (0.014) (0.002) [0.000]

Inflows 0.197 0.090 0.107*** 0.125 0.075 0.050***(0.008) (0.001) [0.000] (0.010) (0.001) [0.000]

Notes: This table shows the average employment outflows and inflows between pairs of affiliatedand unaffiliated firms. The upper panel presents the average flows for the full sample of employees,while the bottom panel shows the average flows only for top-occupation employees (top quartile ofwithin-firm wage distribution). We restrict the sample of employment outflows (inflows) to pairsof firms in which the sending (receiving) firm belongs to a business group. Exp. and Imp. arefirms that simultaneously export and import. Standard errors in parenthesis, p-values in brackets.

40

Table A.2: Reactions to International Price Shocks: Worker Groups Across Entire IncomeDistribution

All Workers Bot. 25 Mid. 50 Top 25(1) (2) (3) (4) (5)

∆ Exports x Same BG 0.004 0.003∗ 0.005∗ 0.001 0.006(0.002) (0.002) (0.003) (0.003) (0.006)

∆ Imports x Same BG -0.003 -0.004 -0.018 -0.024∗∗∗ -0.008(0.004) (0.004) (0.012) (0.009) (0.008)

∆ Exports 0.000 0.001 -0.003∗ 0.002 0.005∗∗∗

(0.001) (0.001) (0.002) (0.001) (0.002)∆ Imports -0.001 -0.001 0.002 0.001 -0.000

(0.001) (0.001) (0.002) (0.001) (0.001)Same BG 0.024∗∗∗ 0.022∗∗∗ 0.016∗ 0.033∗∗ 0.044∗∗∗

(0.006) (0.006) (0.009) (0.013) (0.011)Log Avg. Wage Origin 0.003 0.037∗ 0.028 -0.089

(0.018) (0.019) (0.027) (0.060)Log Avg. Wage Destination 0.000 -0.000 -0.009 0.001

(0.003) (0.006) (0.006) (0.020)Log Employment Destination -0.002∗ -0.003 -0.004 -0.010∗

(0.001) (0.002) (0.003) (0.005)Log Employment Origin -0.046∗∗∗ -0.065∗∗∗ -0.068∗∗∗ -0.080∗∗∗

(0.009) (0.015) (0.014) (0.020)Log Emp. BG Origin -0.095 -0.033 -0.150∗∗ 0.112

(0.082) (0.120) (0.060) (0.139)Log Emp. BG Dest. -0.123∗∗ -0.130 -0.178∗ -0.115

(0.049) (0.111) (0.106) (0.175)Log N. Firms BG Origin 0.061∗∗ 0.031 0.073∗ 0.282∗∗

(0.030) (0.025) (0.037) (0.129)Log N. Firms BG Dest. 0.010 0.036 0.028 0.080

(0.015) (0.036) (0.030) (0.090)R2 0.522 0.594 0.801 0.664 0.685Origin FE X X X X XDestination FE X X X X XYear FE X X X X XMean DV .02 .02 .039 .042 .067SD DV .043 .042 .085 .076 .111N 21084 20253 6259 7532 5083

Notes: Robust standard errors in parentheses are double clustered at the level of the firm thatsends and receives the flow. *** p<0.01, ** p<0.05, * p<0.1

41