Policy Research Working Paper                  10371




Offshoring Response to High-Skilled Immigration
                        A Firm-Level Analysis

                              Devaki Ghose
                              Zhiling Wang




 Development Economics
 Development Research Group
 March 2023
Policy Research Working Paper 10371


  Abstract
  Using a policy change in the Netherlands in 2012 that                             non-European Union workers, small firms hire and fire
  made it easier and less costly for firms to employ high-                          more non-European Union workers in a given year. Many
  skilled short-stay non-European Union workers and a                               of these workers return to their home countries, establishing
  matched employer-employee data, this paper shows that                             direct connections that boost offshoring to firms in the
  firms in high-skill industries respond by both employing                          Netherlands. By contrast, large firms absorb some of the
  a higher share of non-European Union immigrants and                               workers leaving the small firms. These workers also estab-
  increasing the total amount of offshoring to non-European                         lish connections between their host and origin countries,
  Union countries. With reduced costs of hiring short-stay                          boosting offshoring.




 This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the
 World Bank to provide open access to its research and make a contribution to development policy discussions around the
 world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may
 be contacted at dghose@worldbank.org.




         The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
         issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
         names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
         of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
         its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                       Produced by the Research Support Team
   Offshoring Response to High-Skilled Immigration: A
                                       Firm-Level Analysis

                      Devaki Ghose*                                Zhiling Wang †

                       World Bank                ‡
                                                  Erasmus University Rotterdam




       JEL Codes: F22, F16, J61, F66, F68.
       Keywords: Immigration, Offshoring, Globalization.




   * Development Economics Research Group, The World Bank, dghose@worldbank.org.
   †
     Erasmus School of Economics, Erasmus University Rotterdam; Tinbergen Institute. z.wang@ese.eur.nl.
   ‡
     We thank Bob Rijkers, Andrew Bernard, Gianluca Orefice, seminar participants at EIIT, 12th Annual Conference
on “Immigration in OECD Countries”, Mid-West Trade, Southern Economic Association, Erasmus University, Rot-
terdam and World Bank for constructive comments. We further thank Statistics Netherlands for providing data. The
findings, interpretations, and conclusions expressed in this paper are entirely those of the author and do not necessar-
ily represent the views of the World Bank and its affiliated organizations or those of the Executive Directors or the
countries they represent. All errors are the author’s responsibility.
1       Introduction
Immigration of high-skilled workers and the relocation of production abroad by multinational firms
(commonly known as offshoring) are two important engines of globalization. Despite being at the
heart of intense public debate, our understanding of the link between immigration and offshoring
at the firm level is limited, especially given the various types and duration of immigration policies.
Recent empirical work has shown that increasing immigration can increase offshoring to migrants’
home countries through ethnic networks while reducing offshoring to other countries (Olney and
Pozzoli, 2019; Moriconi et al., 2020).
    In this paper, we argue that to assess the impact of high-skill immigration on offshoring, it
is crucial to understand the nature of the immigration policy and account for the effects of such
policies on firm level employment decisions. A policy that mandates high-skilled workers to work
for at least a year, such as the H1-B policy in the US, is likely to have a different impact than a
policy that allows migrant workers to work for as little as two months, such as the short-stay high-
skilled migrants scheme introduced in the Netherlands in 2012. Since the costs of hiring short-stay
workers are much lower than the costs of hiring long-term workers, a short-stay migrant policy
could have differential effects on firms depending on their abilities to sponsor and hire high-skilled
migrants, for example, across small versus large firms.1
    Combining detailed employer-employee-matched data with firm level data in the Netherlands,
we study how firms respond to changes in the costs of hiring short-stay, high-skilled immigrants
by changing their offshoring decisions. Does easier access to short-stay, high-skilled immigrant
workers lead firms to produce more inputs in-house? Or does the short-stay nature of the scheme
help firms acquire more inputs from abroad that they could have previously produced in-house as
these immigrants return to their home countries and establish connections with Dutch firms? We
exploit a change in the Dutch immigration policy in 2012 that made it easier for firms to recruit
high-skilled workers from non-EU countries and interact it with the firms’ historical share of non-
EU immigrant employees to causally identify the impact of immigration on offshoring.
    We find that the reductions in the costs of hiring short-stay high-skilled migrants introduced by
the policy led to an increase in firm level offshoring to non-EU countries from which the immi-
grant workers came. The mechanism through which this operates is very different for small versus
large firms. Small firms, typically constrained in paying the high costs of hiring long-term non-EU
workers in the past, immediately responded to the scheme by hiring high-skilled non-EU workers.
These firms hired non-EU workers and fired them within a year, suggesting an unmet need in the
labor market for high-skilled short-stay workers. In conversations with law firms that specialize in
recruiting non-EU immigrant workers, we learned that many firms in high-skill intensive industries
    1
    The fact that large firms are more able to hire costly immigrant workers has been documented in the literature
using German employer-employee matched data (Brinatti and Morales, 2021).


                                                       2
need mobile high-skilled workers working on limited-duration projects that require short on-site
stays in the Netherlands before they return to their home countries or move to different countries.
Using a return migration survey, we confirm that many of these non-EU workers returned to their
home countries. These return migrants fostered connections between firms in the Netherlands and
their home countries, leading to offshoring. We call this mechanism the “direct connection”, im-
plying that connections were established between firms in both countries due to workers physically
returning to their home countries.
     Large firms, on the other hand, do not immediately hire non-EU workers and fire them; that
is, they are less interested in hiring short-stay workers. However, we show that many non-EU
workers who leave small firms join large firms. This movement of workers to large firms increases
non-EU employment at large firms, but with some delay compared to the increase in small firms.
These workers also foster connections between firms in the Netherlands and their home countries,
increasing offshoring. We call this mechanism “indirect connection”, implying that connections
are established between firms in both countries without workers physically moving back.
     While there is a solid body of older work that analyzes the positive role of migrants on trade
(Gould, 1994; Head and Ries, 1998; Rauch, 1999; Rauch and Trindade, 2002; Felbermayr and
Toubal, 2012), these papers struggled to causally estimate the relationship between migration and
trade due to either very aggregate data or endogeneity issues (Felbermayr et al., 2015). Recent
work has addressed some of these endogeneity issues by using either credible instruments (Bur-
chardi et al., 2019; Cohen et al., 2017; Bastos and Silva, 2012; Orefice et al., 2021) or natural
experiments (Ariu, 2022; Parsons and V´    ezina, 2018; Steingress, 2018; Bahar et al., 2019). There is
a growing literature that documents the importance of the information value of immigrants in facil-
itating trade and offshoring using more dis-aggregated firm level matched employer-employee data
(Cardoso and Ramanarayanan, 2022; Erbahar and Gencosmanoglu, 2021; Andrews et al., 2017;
Hatzigeorgiou and Lodefalk, 2016; Hiller, 2013; Marchal and Nedoncelle, 2019). While the major-
ity of this literature looks at export and in some cases import promoting effects of trade, there are
only a handful of studies that look at the effects of immigration on offshoring (Fan et al., 2022; Ol-
ney and Pozzoli, 2019; Ottaviano et al., 2018; Glennon, 2020). As discussed in the literature review
by Hummels et al. (2018), the concept of offshoring is very different from pure imports or exports,
as offshoring embodies the concept of inputs that could have been produced domestically but are
imported from abroad. Differently from these papers, we focus on a policy that only changes the
costs of hiring short-stay, high-skilled workers who tend to return to their home countries faster
than more permanent high-skilled workers. We utilize this new policy in an instrumental variable
strategy to isolate the impact of changes in the costs of hiring short-stay, high-skilled immigrant
workers on firm offshoring.2 Does access to short-stay non-EU high-skilled immigrant workers
   2
       A closely related work, Glennon (2020) finds that restrictions on H-1B immigration in the US cause firm level


                                                         3
lead firms to offshore some products they previously produced or acquired domestically? Impor-
tantly, does this access benefit certain types of firms more than others, for example, smaller firms
that are generally more constrained in sponsoring and hiring longer-term immigrant workers?
    As pointed out by Ariu (2022), we still know little about the mechanisms through which foreign
workers foster trade and offshoring. The literature so far has identified the importance of networks
in reducing information costs by showing that immigrants increase trade for more differentiated
goods (Peri and Requena-Silvente, 2010) and origin countries with weaker institutions (V´       ezina,
2012; Moriconi et al., 2020), increase offshoring to their own countries but reduce offshoring to
other countries (Olney and Pozzoli, 2019), create more stable buyer-supplier relationships (Egger
et al., 2019), and increase the quality of intermediate inputs sourced from origin countries (Ariu,
2022).3 Several papers have documented that highly-skilled migrants have more trade-fostering ef-
fects than other types of migrants, mainly by looking at the stock of existing highly-skilled migrants
using an instrumental variable strategy (Bahar and Rapoport, 2018; Cardoso and Ramanarayanan,
2022). In contrast, we use a policy directly aimed at reducing the costs of only high-skilled, short-
stay migrants and isolate the effects of such a policy on firm level offshoring.
    Our most important contribution is to document, using detailed employer-employee matched
data, a new mechanism through which changes in very short-stay migration policies affect off-
shoring. Short-stay workers have a higher tendency to return home, thereby increasing direct con-
nections, which we are able to test using unique data that tracks return migration. This benefits
smaller firms more, which were historically less able to hire non-EU workers on costly perma-
nent contracts.4 However, bigger firms also see a rise in offshoring because they tend to absorb
the newly recruited non-EU workers who separate from small firms. The particular nature of the
high-skilled migrants’ policy that changed the costs of hiring only short-stay non-EU workers and
the detailed data that can track individual worker movements across firms help us to establish this
mechanism empirically, which, to our knowledge, is a unique contribution to the literature on the
nexus between immigration and offshoring.
    The paper is organized as follows. In section 2, we discuss the change in Dutch high-skilled
immigration policy in 2012 and the employer-employee matched data of the universe of Dutch
firms that we use to study the effect of this policy on firm level immigration and offshoring. In
foreign affiliate employment to increase in some countries. While changes in the foreign affiliate employment as a
proportion of multinational employment can directly capture any relocation of firm activities within the boundaries of
a multinational corporation (MNC), it misses any arms-length transactions, that is, foreign sourcing from unrelated
suppliers (Hummels et al., 2018; Grossman et al., 2006), which we can capture.
    3
      Using Danish employer-employee matched data, Olney and Pozzoli (2019) show that an influx of immigrants into
a municipality reduces firm level offshoring. Differently from Olney and Pozzoli (2019), we directly use firm level
immigrant employment and focus on a policy that changed the availability of only short-stay high-skilled immigrants.
    4
      This result relates to the findings of Mitaritonna et al. (2017) who show that a supply-driven increase in the share
of foreign-born workers in French regions increased the total factor productivity of firms, especially smaller firms, in
those regions.


                                                           4
section 3, we discuss our empirical strategy, including identification. We provide evidence that
firms hire more non-EU immigrant workers in response to a reduction in the cost of hiring non-
EU immigrants and an increase in offshoring to non-EU countries. Section 4 sheds light on the
different mechanisms consistent with our findings in section 3. Section 5 discusses the robustness
checks, including restricting the analysis to various sub-samples and the threats to identification
from concurrent policy changes. Section 6 concludes.


2      Background, data, and summary statistics
Our empirical analysis uses a change in Dutch high-skilled immigration policy and employer-
employee-matched data from Statistics Netherlands. In this section, we provide background on
the Dutch high-skilled immigration policy and then describe our data.


2.1     Expansion of knowledge migrant scheme to short-stay workers:
The institutional setting for migration to the Netherlands differs fundamentally between European
Union member states (EU) and non-EU nationals. Every EU national has the right to settle in
the Netherlands and seek employment in the Dutch labor market. Non-EU migrants need to ap-
ply for residence and work permits in the Netherlands. There are different schemes for different
purposes of immigration. Most highly skilled migrants gain access to the Netherlands through the
“Highly Skilled Migrants Scheme” (Kennismigrantenregeling), introduced in 2004, which guar-
antees quick processing and high acceptance rates for migrants whose wages are above a certain
threshold (Berkhout et al., 2015).
    However, high-skilled workers seeking short-stay employment on an initial contract lasting less
than two months are not eligible to work using this scheme. A new scheme was started in January
2012 for highly skilled workers who want to come to the Netherlands to work for less than two
months. According to European Migration Network (2013), in order to gain admission on the
grounds of this pilot project, the following requirements were set:

      • The employer has been admitted to the Highly Skilled Migrants Scheme.

      • The salary must be at least proportionally equivalent to the salary as demanded for highly
        skilled migrants of 30 years and older.

      • It must be apparent that the job relates to one that can be deemed to be that of a highly
        qualified worker.

    It should be noted that all these provisions existed even before 2012. The only change this pol-
icy introduced was to include short-stay workers in this scheme: workers who came to work in the

                                                 5
Netherlands for less than two months. Generally, work permit applications are subject to a labor
market test, which entails that the employer submits information to the Dutch immigration author-
ities about the company, the job contract, and the recruitment process to show the position could
not be filled by an EU/EEA or Swiss national.The extension of the knowledge migrant scheme to
short-stay workers relaxed these restrictions as long as they met the above salary and education
requirements. This scheme made it easier for firms to employ high-skilled migrants from non-EU
countries on a short-stay basis post-2012 relative to low-skilled migrants.
    A flexible and fast short-stay feature of job contracts is important for the Dutch labor market to
hire non-EU workers, as the use of workers with temporary contracts by firms is a typical feature of
the Dutch labor market. Most industries have about 20% to 50% temporary contract holders among
foreign workers and slightly less among native workers (10% to 40%).5 Workers typically go
through several fixed-term temporary contracts to achieve a permanent contract. Before 2012, the
option of a less than 2 months contract was only possible for intra-corporate cross-border transfer,
the total processing time of which was about 6 to 8 weeks. After the 2012 short-stay project was
implemented, the total processing time of the intra-corporate transfer was shortened by about half,
while a more substantial change was the possibility of hiring high-skilled non-EU workers from
any company abroad. This change in policy allowed firms access to a flexible pool of high-skilled
workers whom they could hire much faster than before the change and without any restriction on
the contract duration. In conversations with the Ministry of Justice and Security and law firms
specialized in global recruitment, we found out that there was a long-standing need for such global
workers in industries that recruit highly-skilled workers on a project basis. For example, an IT
worker working on a specific project for a year may have to work in the Netherlands for 2 months
and then move to a different country. High-skilled firms thus greatly appreciated the relaxation in
the “Highly Skilled Migrants Scheme”.
    Importantly for our question analyzing the nexus between immigration and offshoring, short-
stay non-EU workers have higher mobility on average than workers with other migration motiva-
tions, for example, workers coming for a longer-term fixed contract or for family reasons. Workers
with short-stay contracts are more likely to leave the Netherlands. They may go to other countries
to do short-stay tasks or go back to their home countries, which potentially brings new international
networks to the Dutch firms they have previously worked at. In this case, firms may develop better
global connections that might lead them to change trade or offshoring activities. Figure 1 shows
the overall mobility of different visa holders by plotting the total number of non-EU immigrants
who leave the Netherlands within 5 years. As work and family reasons are two major motivations
for migration and labor supply in the Netherlands, we distinguish four types of migration moti-
vation: Highly skilled migrants scheme, labor migration, family reunion, and family formation.
  5
      In service-oriented sectors, the percentage of temporary contracts can be as high as 70%.


                                                           6
As the “Highly Skilled Migrants Scheme” started in 2004, we see a steady increase from 2004 to
2008, and a stable level up until a sharp rise again from 2012. The post-2012 increasing trend
was not detected for labor migration and family formation categories. Though we do see a slightly
increasing trend in family reunions very similar to the “Highly Skilled Migrants Scheme” this is
not surprising. Often a couple (both high-skilled) does not arrive in the Netherlands at the same
time. It is normal that one comes first under the “Highly Skilled Migrants Scheme” and his or her
partner comes under the family reunion visa.6
    According to the Ministry of Justice and Security and law firms specialized in global recruit-
ment, the new scheme for short-stay high-skill workers started as a pilot project in 2012 and became
regular in 2014. This paper uses this scheme to study how firms respond to changes in the relative
cost of hiring high-skilled non-EU immigrants.


2.2     Data sources
Our datasets are integrated from various sources provided by Statistics Netherlands.7 First, our firm
data cover the universe of firms with at least 10 employees at the General Business Register over
the period 1999-2016.8 The data include detailed information on size, industry, location, number
of establishments, productivity, and investment.
    Second, employee data cover the population of all employees over the period 1999-2016. Every
employee’s entire working history can be tracked as detailed as per day. Besides the linked firm
ID, there is information on the length of tenure at the firm, work experience, and other individual
characteristics such as age, gender, education, and country of origin. In addition, we have two
supplementary migration-related datasets. The migration motives dataset records the first-time mi-
gration motives since 1998 of all immigrants in the Netherlands. The foreign migration movement
dataset contains the dates of individual-level immigration and emigration identified by municipality
registration records. Two-thirds of emigrants inform the municipality about their emigration and
deregister from their place of residence in the Netherlands. In particular, the dataset also includes
which country emigrants go to after leaving the Netherlands for emigrants who have deregistered
at the municipality.9
   6
      Family reunification is the reunification of an existing family. So, for example, if a person is already in the
Netherlands and his/ her family comes to the Netherlands, the migration motive of the family will be regarded as
family reunification. Family formation is about the formation of a family. A person therefore specifically comes to the
Netherlands to marry or enter into a partnership with someone in the Netherlands.
    7
      Results are based on calculations by Erasmus University Rotterdam using non-public microdata from Statistics
Netherlands. Under certain conditions, this microdata is accessible for statistical and scientific research. For further
information: microdata@cbs.nl.”
    8
      The selection of firm size is mainly to exclude self-employment.
    9
      See appendix A.1 for detailed information about the process of registration and sample sizes in the foreign migra-
tion movement data.



                                                           7
    Third, trade data cover all firms in the Netherlands that have a VAT number and trade in goods
with foreign countries over the period 2010-2016. The data include country of import, country
of export, type of goods (8-digit Combined Nomenclature), the value of import, and the value
of export at the firm level. Using this information, we construct a trade dataset at the firm-by-
year-country level, spanning 2010-2016. We will use this data to empirically investigate whether
offshoring to non-EU countries is affected by the share of immigrants from this country at the firm
level. The first year for the analysis is 2010, due to the year coverage in the trade data. In addition,
we also create an ancillary annual dataset of firms from 1999 to 2016, which does not contain the
trade data. This allows us to investigate the changes in firm-level employment between 1999 and
2016 in all sectors.
    To sum up, data between 2010 and 2016 are used for the main analysis of offshoring outcomes,
and earlier data between 1999-2009 are used to analyze the pre-trend of firm-level immigrant hiring.
For an exhaustive list of data sources in Statistics Netherlands, see appendix A.2.


2.3       Variables
2.3.1      Firm-level non-EU immigrant share and offshoring measures

Our main independent variable of interest is the firm-level immigrant share from a non-EU country
in total firm employment. In section 3.1, we provide more details on why we chose this variable
to understand how changes in the costs of hiring short-stay, high-skilled non-EU immigrants affect
offshoring. Non-EU immigration drives the majority of the increase in total immigration in the
Netherlands over the period 2009-2016, while EU immigration is low in level, and its growth is
almost linear. We first show the national trend of immigration in figure 2 for non-EU countries
and EU-28 countries.10 The biggest boost in the increase in EU immigrants started after 2004,
when the largest expansion of the EU took place. Ten countries (Cyprus, the Czech Republic,
Estonia, Hungary, Latvia, Lithuania, Malta, Poland, the Slovak Republic, and Slovenia) became
EU member states, thus providing an important source of EU immigrants to the Netherlands. The
trend of non-EU immigration, however, shows a distinctively different pattern. The percentage of
non-EU immigrants remains at a steady level before 2012. In only 3 years after 2012, the percentage
of non-EU immigration rises by half a point, which exceeds the total percentage increase for the
previous decade, 2001-2011.
    This coincides with the 2012 expansion of the knowledge migrant scheme, which should only
affect immigrants from non-EU countries. We thus further look into the fast-growing source coun-
tries of immigrants for the Netherlands before and after 2012. Taking 2009, we calculate the
growth rate in the share of immigrants for different countries and plot the top 10 countries in figure
  10
       UK is still counted as an EU member as our main analysis period precedes Brexit.


                                                           8
3. Besides the 3 new EU member states, Poland, Romania, and Bulgaria, that joined the EU after
2004, India ranked particularly high. This could be linked to the expansion of the high-skilled
immigrant policy because the majority of Indian workers in the Netherlands are employed in high-
skill-intensive sectors such as IT.11 We will provide rigorous evidence on how this scheme affected
high-skilled non-EU immigration in section 3.2 on identification.
    Our main dependent variables of interest are firm-level offshoring measures. Following the
well-established method of using import data to construct offshoring measures (Hummels et al.,
2014; Bernard et al., 2020), we construct a “narrow offshoring” measure based on whether firms
import products from a non-EU country to which they also export at the HS4 level, where exports
are proxy for the existence of the firm’s domestic production capacity in that particular HS-4.12
This firm-level measure of offshoring is generally regarded as the “gold standard for accuracy”
(Hummels et al., 2018).13 We focus on two dimensions of offshoring: whether firms decide to
offshore from a non-offshoring status and the total values of offshoring products conditional on the
firm offshoring. More specifically, on the extensive margin, if firms import the same HS4 category
products from a non-EU country as firm production, the variable is equal to 1 and 0 otherwise.
On the intensive margin, the variable is the total log value of goods off-shored, conditional on
participation in offshoring.

2.3.2   Control variables

We describe below the definitions of the firm workforce variables used in this paper. Male: share
of male workers among all workers. Age: the average age of all workers. Bachelor: share of
native workers with a bachelor’s degree among all native workers.14 Tenure: average workers’
tenure at a firm. Work experience: average years of work experience.15 In addition, the firm
characteristics consist of the following variables. Productivity: the log of average daily income per
employee. Capital: the log of the total purchase value of the tangible fixed assets per worker. Size1:
a dichotomous variable of small-sized firms equal to 1 if a firm has at least 10 and fewer than 50
workers. Size2: a dichotomous variable of medium-sized firms equal to 1 if a firm has at least 50
  11
      https://www.cbs.nl/en-gb/news/2019/30/indian-knowledge-migration-has-doubled
  12
      The Harmonized System (HS) is an internationally standardized system of names and numbers to classify traded
products. This code has a direct correspondence with the code of goods (Combined Nomenclature) in our data.
   13
      According to Hummels et al. (2018), there are three limitations of using this measure that is worth noting here.
One, it is available for a relatively small subset of countries. Two, the data are often times confidential and not
accessible to all researchers. Three, while measures of merchandise imports are of high quality, services- imports
coverage remains relatively weak.
   14
      The information on education level is only available for native workers because the majority of them completed
bachelor’s degrees at Dutch universities with records. However, for foreign workers, this information is not precise.
The earliest year of graduation records that can be obtained is 2000. Therefore, the reported education level is left
censored at 2000.
   15
      The earliest employee data can only be traced back to 1999. Therefore the count of work experience is truncated
at 1999 for every worker.


                                                         9
and fewer than 100 workers. Size3: a dichotomous variable of large-sized firms equal to 1 if a firm
has at least 100 workers. Multi-establishment: a dichotomous variable equal to 1 if the firm has
multiple establishments.
    For regional fixed effects, we adopt the Nomenclature of Territorial Units for Statistics (NUTS)
code for regions and choose the most disaggregated level NUTS3 to identify areas in the Nether-
lands. In total, there are 40 NUTS3 regions in the Netherlands. Every NUTS3 region contains
a central municipality. Widely used in structural analyses of Dutch labor markets, NUTS3 is an
appropriate spatial scale to define labor market areas (Corvers et al., 2009). Territorial disparities
are substantial, and the division has been used intensively to plan spatial policies by specific ad-
ministrative bodies. Our research should control for regional fixed effects that capture the local
labor market environment. The use of NUTS3 regions is more precise for our analysis than the
use of municipalities at a geographically finer level, as cross-municipality commuting for work is
prevalent due to well-connected transportation networks in the Netherlands. We use the Dutch stan-
dard industrial classification in 2008 (“Standaard Bedrijfsindeling” or SBI in Dutch) for industry
categories. The most detailed category is coded in a 5-digit number. In our analysis, we control for
2-digit industry code fixed effects, which include 84 categories in total.
    As we are interested in exploiting the response of high-skill firms in response to the expansion
of the “Highly Skilled Migrants Scheme” in 2012, we need to define the skill intensity of each
firm. Here we use the share of native workers with a bachelor’s degree among all native workers in
2004 to define whether a firm is high-skill or not.16 More specifically, we use the top 25% as the
threshold value. Therefore, we create a variable HS firm: Equal to 1 if a firm is in the top quarter
of skill intensity among all firms and 0 otherwise. We will explore more heterogeneity regarding
firm size later in the mechanism section 4, as the policy might impact these two types of firms
differently. We define a firm as big if the employment size is above 50 and small otherwise, in
2009.17


2.4     Descriptive statistics
Panel A of table 1 presents descriptive statistics of offshoring, immigration, workforce, and firm
characteristics over the period 2009-2016 in manufacturing sectors. This is the dataset for our main
analysis. Three percent of firms engage in offshoring to one non-EU country based on the narrow
definition. The average annual value of firm-level offshoring to a particular destination country is
5,803,000 euros.
   The average immigrant share from a particular non-EU country within a firm is 0.16%, while
  16
     2004 is the starting year of the “Highly Skilled Migrants Scheme”. We pick this year to ensure that our high-skill
definition should be least affected by any accumulating impacts from the scheme.
  17
     2009 is the first year of the study period.


                                                         10
the share of immigrant workers from all non-EU countries at the firm level is about 5.7%. In terms
of workforce characteristics, male workers make up 79% of all employees in the manufacturing
sector and 64% of all employees overall. The average age of workers in manufacturing is 43, while
the average age of all workers is 39. Among native workers, the share of bachelor’s degree holders
is 16.3% on average. An average worker has 11.6 years of work experience from 1999 onward,
including 7.8 years at the current firm. Regarding firm characteristics, the majority of firms (64%)
are small-sized, and 38.3% of firms have more than 1 establishment.
    We also show descriptive statistics of our ancillary dataset in panel B of table 1. It includes
firm characteristics over the period 1999-2016 for all sectors but excludes the offshoring outcomes.
Overall, they show similar statistics for firm characteristics as in panel A.


3      Empirical strategy
In this section, we discuss our empirical framework. Our main goal in this section is to study how
firm-level offshoring responds to changes in the cost of hiring immigrant workers. A-priori, it is
unclear whether increasing access to a pool of high-skilled short-stay workers induces the firm
to reduce offshoring and acquire more inputs domestically or the global connections these highly
mobile short-stay workers bring about lead firms to offshore more.


3.1     Specification
We estimate the following equation:

    Of fijmt,d = βo + β1 Immijmt−1,d + Xijmt−1 γ1 + Wijmt−1 γ2 + γi + γj + γm + γt + γd +   ijmt,d

                                                                                                 (1)

where Of fijmt,d is offshoring by firm i in industry j in region m at time t to destination country
d. Our analysis focuses on narrow offshoring at both the extensive and intensive margins. Xijmt−1
includes a set of firm characteristics that can influence offshoring decisions. Specifically, we in-
clude variables Productivity, Capital, Size, and Multi-establishment. The vector Wijmt−1 includes
detailed workforce characteristics, including variables Male, Age, Bachelor, Tenure and Work Ex-
perience. We also include firm, industry, region, and time-fixed effects. Post=1 when years are
greater than or equal to 2012. Immijmt−1,d is the share of hiring from country d out of total hiring
by firm i in industry j in region m in year t − 1. Immigration and other independent variables are
lagged because it takes time for firms to adjust in response to changing economic conditions.
    Note that Immijmt−1,d includes all the non-EU immigrants from country d hired by the firm i
in t − 1, not just short-stay workers. There are two main reasons for this: First, we do not have

                                                11
information in the data on the exact contract duration of workers to distinguish short-stay workers
from longer-term contract workers. Second, even if this information was available, counting only
short-stay workers would severely underestimate the impact of the scheme because many workers
who come on initial short-term contracts using the scheme can transition into permanent contracts.
Moreover, many workers using this scheme bring their partners, who can also be high-skilled im-
migrants, but those workers can come to the Netherlands using partner visas. To understand the
overall impact of this short-stay migrants scheme on high-skilled non-EU employment, we instead
turn to an instrumental variable strategy described in section 3.2 where we estimate the change in
non-EU immigrant employment driven by this scheme. An instrumental strategy is also crucial be-
cause equation 1 is simply a relationship between equilibrium immigrant hiring and off-shoring at
the firm level. Since both immigration and offshoring are determined in equilibrium by the firm, the
same demand and supply shocks affect both. Therefore β1 in equation 1 is simply a correlation. For
example, firms that are growing over time or expanding their multinational activities abroad could
tend to hire more immigrants as well as offshore more. Over time, firms could also specialize or
introduce certain products in their portfolios that require skills that are more available in foreign
workers. These factors would induce a spurious positive correlation between immigrant employ-
ment and offshoring at the firm level. Therefore, we turn to instrument the immigrant workers hired
by firms using the 2012 policy change.


3.2   Identification
We exploit a 2012 policy change that made it easier for firms in the high-skill-intensive sector to
hire immigrant workers. In a two-stage least squares framework, we isolate the part of changes
in immigrant hiring driven by changes in government immigration policy in the first stage and
use this variation to see how firm-level offshoring responds in the second stage. In other words,
we analyze how firms adjusted their equilibrium offshoring decisions in response to changes in
the cost of hiring immigrants induced by the 2012 policy change. This gives us an estimate of
how policy-induced changes in the costs of hiring immigrants affect firms’ offshoring decisions
in equilibrium by changing their optimal immigrant hiring. As noted in section 3, the changes in
non-EU immigrant hiring induced by the short-stay high-skilled migrant policy not only include
changes in the hiring of short-term workers but could also include high-skilled spouses of non-EU
immigrants using this scheme or short-stay workers who are later offered a long-term contract.
    To understand the effects of changes in the costs of hiring immigrants on firm offshoring deci-
sions, we first need to study whether the 2012 policy had any effect on firm-level immigrant hiring.
Before showing how immigrant hiring changed in high-skill intensive industries compared to low-
skill intensive industries post-2012 in a difference-in-difference framework, we plot the trend in


                                                 12
EU and non-EU immigrants in the entire population. Figure 2 shows that while the share of EU
immigrants has been growing at a constant rate over time, the share of non-EU immigrants has been
stable until 2012 before spiking in 2013. To quantify the differential trend in non-EU immigrant
hiring in high-skill-intensive industries relative to low-skill-intensive industries, we estimate the
following:

       Immijmt−1 = βo + Xijmt−1 γ1 + Wijmt−1 γ2 + γi + γj + γm + γt + αt−1 ∗ γs +                        ijmt−1      (2)

Immijmt−1 is the share of non-EU immigrant hiring out of total hiring by firm i in industry j in
region m in year t − 1. γs is a dummy indicating whether the firm belongs to a high-skill or a
low-skill industry. αt−1 measures the differential trend in immigrant hiring in high-skill relative to
low-skill intensive industries in year t − 1. All firm and worker level characteristics as well as fixed
effects are the same as in equation 1.18
    Since the 2012 policy directly increased the ease of hiring non-EU high-skilled immigrants, we
expect αt to be positive and significant from 2012 only for non-EU immigrants. We cluster the
standard errors at the industry level since the 2012 scheme targeted firms in high-skill industries.
We show the event study graphs for the shares of non-EU employment in figure 4, and the shares of
EU and native employment in figure 5. Figure 4 shows that between 2004 to 2011, there was no in-
creasing trend in non-EU employment share in high-skill relative to low-skill industries. However,
after the introduction of the high-skilled immigration scheme in January 2012, the share of non-EU
immigrants in high-skill relative to low-skill industries started increasing every year.19 The average
annual increase in the share of non-EU immigrants was 4% more post-2012 in high-skill relative
to low-skill industries compared to their pre-period difference.20 The share of EU immigrant em-
ployment in high-skill relative to low-skill industries does not show any pattern before and after
the 2012 policy change. It shows an insignificant increase from 2012-2014 and a decrease from
2014 to 2016 (figure 5, top panel). Native employment share reduces after the 2012 policy change,
consistent with the fact that the employment shares should add up to 1 (figure 5, bottom panel).
The tables corresponding to the event study graphs are available in table 9 in the online appendix.21
  18
      An emerging literature documents several problems associated with using the traditional difference-in-difference
methods in the case of staggered treatment timings and proposes new estimators that are valid in the case of staggered
treatments (see, for example, a recent literature review by Callaway (2022)). The short-stay high-skilled migrants
scheme was adopted in 2012, and there was no staggered rollout of this scheme.
   19
      There is a slight increase in non-EU immigrant share from 2010 to 2011, but the pairwise t-test confirms that the
difference in magnitudes is statistically insignificant. The year 2009 is a year of a structural break in the data because
of the recovery from the financial crisis. Both the non-EU and EU immigrant unemployment rates increased greatly
during the recession (Cerveny and Van Ours, 2013). After the recovery, both the shares of EU and non-EU immigrant
employment rose, especially in high-skill industries primarily affected by the recession.
   20
      This calculation is obtained by dividing the yearly coefficient by the average firm-level hiring of non-EU immi-
grants, 5.7%
   21
      In robustness checks in section 5, we check for other concurrent policies that can affect our identification.


                                                          13
    Since offshoring and immigrant hiring in both high and low-skill industries have been changing
over time due to globalization, it is not enough to just look at how offshoring changed post-2012
in high-skill relative to low-skill industries to understand the impact of the changes in the costs of
hiring immigrants on offshoring. To isolate the impact of changes in the cost of hiring immigrants
on firm-level offshoring, we use two sources of variation: First, we use the fact that the 2012 policy
primarily affected the high-skill-intensive industry. Second, we use historical variations in firm-
level immigrant employment as these variations can affect future immigrant employment. Network
effects, similar to the literature that use the historical tendency of immigrants to settle in already
immigrant-intensive areas, would imply that firms that were already immigrant-intensive would
hire more immigrants.22 On the other hand, if the short-stay migrants scheme primarily helped
small firms, which historically could not afford to hire long-term high-skilled non-EU workers due
to several costs associated with employing immigrants, the high-skilled migrants scheme would
have the opposite effect. In this situation, small firms in the high-skilled sector would benefit
more from the short-stay migrants scheme and hire more high-skilled non-EU immigrant workers
compared to both large firms in high-skilled sectors and small firms in low-skilled sectors.
    Our empirical strategy uses a two-stage least squares framework using these two variations.
First Stage:

  Immijmt−1,d = βo + Xijmt−1 γ1 + Wijmt−1 γ2 + γi + γj + γm + γt−1 + γd + α1 post ∗ γs
                            + α2 post ∗ Imgid + α3 γs ∗ Imgid + α4 post ∗ γs ∗ Imgid +             ijmt−1,d   (3)

Second stage:

  Of fijmt,d = βo + β1 Immijmt−1,d + Xijmt−1 γ1 + Wijmt−1 γ2 + γi + γj + γm + γt + γd + εijmt,d
                                                                                              (4)

where post=1 if year ≥ 2012. Imgid is the proportion of non-EU immigrants in firm i in year
1999 from origin country d out of total firm hiring. γs is a dummy indicating whether the firm is
high or low-skill-intensive as defined in section 2.3.2. This is a triple difference setup where: α1
measures the differential trend in immigrant hiring in high relative to low-skill-intensive industries
post-2012. α2 measures the differential trend in how the past network of immigrant employees in
firm i affects immigrant hiring post-2012. α3 measures the differential effect of the past network
of immigrant employees on future immigrant hiring in high relative to low-skill industries. α4 is
the triple difference term that measures how the effect of the past network of immigrant workers on
immigrant hiring by firms in high relative to firms in low-skill industries changed after 2012.
  22
   This theory was pioneered by Card (2001) and has been extensively used thereafter. See, for example, Cascio and
Lewis (2012), Boustan (2010) as examples in this large literature.


                                                       14
    Since the expansion of the knowledge worker scheme in 2012 affected firms in the high-skill-
intensive industry much more than firms in the low-skill-intensive industry, we expect α1 to be
positive and significant if the policy was effective. If firms with higher past immigrant employees
also attract more employees as is consistent with the network effect of immigration, we expect α2
to be positive. If the effect of past immigrant employees on future hiring by firms is stronger in
high-skill relative to low-skill industries, α3 will be positive. The a priori expected sign of α4 is
ambiguous as it depends on which type of high-skilled firms were able to hire more immigrant
workers post-2012: those with a higher or a lower network of past immigrant employees? Large
firms, which already have many immigrant employees, may have a lower need to hire high-skilled
migrants on a short-stay basis compared to small firms that were less able to hire non-EU immi-
grant workers in the past. In that case, we expect α4 to be negative. On the other hand, if firms
with already large immigrant networks were able to utilize the high-skilled migrants scheme better
compared to firms with lower immigrant networks, α4 will be positive.
    The 2SLS estimate of β1 thus measures the changes in the firm’s equilibrium offshoring in
response to changes in the cost of hiring immigrants. The identification assumption behind the
instrument is that if we see a differential change in firm-level offshoring in firms in high-skill
relative to low-skill industries after the 2012 policy change depending on the existing network of
immigrant employees, it must be because firms adjust their offshoring activities in response to a
reduction in the cost of hiring high-skilled immigrants.
    Table 2 reports the results from the first stage for firms in the manufacturing sector. First-stage
results for both the extensive (columns (1) and (2)) and the intensive (columns (3) and (4)) using
the mean firm-level non-EU immigrant share in 1999 and 2007 as instruments show that non-EU
immigration increased in high-skill industries post-2012, as confirmed already. However, from
rows (4) and (7), we find that non-EU immigrant hiring increased more in firms that had less non-
EU immigrant employment, to begin with, that is, mostly small firms. In section 4, we investigate
this further and analyze the effects of this policy on small and large firms separately.


3.3   Effect on firm level off-shoring
In this section, we study how changes in the costs of hiring high-skilled immigrants affect firm-level
offshoring for manufacturing firms.
    In table 3, we report the results of estimating equation 4 on offshoring to non-EU countries. The
dependent variable in panel A is a dummy indicating whether a firm engages in offshoring, that is,
the extensive margin of offshoring. The dependent variable in panel B is the log value of offshoring,
conditional on the firm participating in offshoring. Columns (1) and (2) report the results from OLS
estimation, while (3) and (4) report the results using firm-level immigrant employment share from


                                                 15
the origin country from 1999 and 2007 as IVs. The results from the IV specification are close:
We find that even after accounting for firm, region, destination, and year fixed effects and a battery
of firm and workforce controls, firms respond to the reductions in the costs of hiring immigrants
by hiring more immigrant workers from the same country and increasing offshoring. In panel A
(column 4), the immigration coefficient of 0.285 implies that a one percentage point increase in the
immigrant share is associated with a 0.0028 increase in the probability that a firm will offshore,
which represents a 7.50% increase relative to the mean. In panel B (column 4), we see that a one
percentage point increase in the immigrant share is associated with a 13.18% increase in offshoring.
Table 4 shows that firms do not increase the volume of off-shoring to EU countries. If anything,
there is evidence of a weak fall in offshoring to EU countries. Overall, the results in tables 3 and
4 provide compelling evidence that firms hire more non-EU immigrant workers in response to a
reduction in the cost of hiring non-EU immigrants and increase the amount of off-shoring to non-
EU countries.23 Note that, in contrast to the existing literature, recent work has shown that firms
do not necessarily reduce domestic production of the products they offshore (Bernard et al., 2020).
Our results, therefore, do not imply that firms that now have better access to high-skilled foreign
workers reduce or stop the production of the same good that they import.
    Since larger firms historically hired more long-term non-EU immigrants and are more able to
sponsor the costs of hiring non-EU workers, we expect the smaller firms to primarily benefit from
the current scheme that reduces the costs of hiring short-stay workers. We already found evidence
for this in section 3.2, where we showed that non-EU immigrant hiring increased more in firms with
historically less non-EU immigrant share, typically smaller firms. We will further explore how the
policy affected large versus small firms differently in section 4. In table 5, we repeat regression 4
separately for large and small firms. From columns (1) and (2), we see that offshoring increases at
the extensive margin for both small and large firms, and there are no significant differences. Bigger
firms see a larger increase in the intensive margins of offshoring than smaller firms. For smaller
firms, all the effects seem to be driven by the extensive margin: Small firms that were not offshoring
before begin offshoring as they get access to short-stay high-skilled immigrants.
    In light of the literature that shows that higher immigrant employees can increase trade and
offshoring to source countries (Olney and Pozzoli, 2019; Moriconi et al., 2020; Lodefalk, 2016;
Steingress, 2018; Parsons and V´   ezina, 2018), we hypothesize that these effects could be driven by
immigrants establishing connections to local firms back home that engage in offshoring with Dutch
firms. But what is the exact mechanism driving this result? Given that the high-skilled migrants
scheme is short-stay, we test whether these results are driven by high-skilled migrants returning
home after the termination of employment in the Netherlands, thereby increasing connections with
  23
     We have similar estimates when we put the interaction term post ∗ γs in the second stage to account for any
differential trend in offshoring across high and low-skill firms. See table 10 in online appendix A.3.



                                                      16
their former employers. We test this mechanism in the following section.


4        Mechanisms
From the first stage results in section 3.2 (table 2) we found that non-EU high-skilled immigrant
hiring increased more for firms that historically hired a lower share of non-EU immigrants, that is,
smaller firms. In 2009, the non-EU immigrant share is 6.6% for small firms and 10.1% for big firms
on average. EU immigrant share is quite similar for small and big firms, possibly because there are
no additional costs of hiring EU immigrants over hiring native workers. It is, therefore, possible
that the small firms were the primary beneficiaries of the short-stay high-skilled migrants scheme as
it allowed them to hire high-skilled workers on a short-stay basis without incurring any substantial
immigration-related hiring costs. If true, this policy would have significant distributional impli-
cations by helping ease the hiring constraints for smaller firms, which typically find it harder to
bear the fixed costs of sponsoring non-EU immigrants (Orrenius et al., 2020). In a standard Melitz
(2003) model with heterogeneous firms, reductions in the fixed costs of exporting help small and
less productive firms by enabling them to participate in the export market. Similarly, immigration,
by reducing the fixed cost of establishing trade relationships, can especially benefit small and less
productive firms to access additional markets (Mitaritonna et al., 2017).


4.1      Labor mobility across firms
We investigate the effect of this policy on the hiring of non-EU immigrants separately for small and
large firms by doing an event study specification, similar to equation 2.

          log (1 + Yijmt ) = βo + Xijmt γ1 + Wijmt γ2 + γi + γj + γm + γt + αt ∗ γs +             ijmt       (5)

Yijmt is the number of non-EU immigrants hired by firm i in industry j in region m in year t.24 γs is
a dummy indicating whether the firm belongs to a high-skill or a low-skill industry. αt measures the
differential trend in immigrant hiring in high-skill intensive industries relative to low-skill intensive
industries in year t. All firm and worker level characteristics as well as fixed effects, are the same
as in equation 1.
    In figure 6, we plot the results of this event study. We can see that right when the policy was
implemented in 2012, there was a discrete jump in the number of non-EU workers hired by small
firms, whereas, for big firms, we barely see any such change. However, two years later, from 2014,
    24
   We take a snapshot of employment at the end of each year, December 31. Changing the date to any other randomly
sampled data does not affect our results.




                                                       17
we see a small fall in the number of non-EU workers hired by small firms and a slight concurrent
increase in the number of non-EU workers employed by big firms.
    If this rise in non-EU employment is indeed driven by the high-skilled migrants scheme that
changes the cost of hiring short-stay non-EU workers, we would expect to see both a rise in the
number of separations for non-EU workers as well as a rise in new hiring of non-EU workers,
especially for small firms. The detailed nature of the matched employer-employee data that tracks
workers daily allows us to check whether workers separate from their employers and join new
firms. We run equation 5 again, where the outcomes Yijmt are the total number of new hires and
job separations. New hires are the total number of workers who started working in a firm in a
year. Job separations are the total number of workers who leave a firm in a year. We aggregate this
firm-level information from daily individual job data. From figure 7, we can see that post-2012,
the large number of new hires in a year is also accompanied by a large number of non-EU worker
separations for small firms. From figure 8, we again find a slightly delayed effect for larger firms:
the number of new non-EU hires, as well as separations, only start picking up after 2014.
    This provides evidence that the reductions in the costs of hiring short-stay high-skilled non-EU
workers disproportionately benefited the smaller firms that were likely less able to hire costly long-
term non-EU workers. Smaller firms immediately started hiring more short-stay non-EU workers,
as evidenced by the spike in the number of new hires and the number of separations. However,
this does not explain the rise in offshoring and employment of new hires post-2014 for large firms.
We start with the question, where do the workers who separate from the small firms go? Given
the delayed increase in non-EU employment in large firms, and given that many workers start their
careers at small firms and later move to bigger firms, we first check whether some of these newly
separated workers join bigger firms.25 We check this mechanism by running equation 5 below:

       log (1 + Yijmt ) = βo + Xijmt γ1 + Wijmt γ2 + γi + γj + γm + γt + α ∗ post ∗ γs +                 ijmt     (6)

Yijmt is the number of non-EU immigrants separated from firm i in industry j in region m in year
t who move to bigger firms. Everything else is identical as in equation 5, except that here we
use a post-2012 dummy instead of the yearly dummies interacted with dummies indicating high-
skill sectors.26 We report the results in table 6. From column (1), we see that there was indeed a
1.8% jump in the number of non-EU workers who moved to bigger firms post-2012. There are no
significant effects on the number of EU and native workers moving to bigger firms.
  25
      In our data, for firms with at least 1 job separation in a year, the share of non-EU workers moving from small to
large firms among all non-EU separated workers is 68%. The share of native workers moving to larger firms among all
native separated workers is 72%.
   26
      The firm datasets from Statistics Netherlands only provide the exact number of employment size from 2009 on-
wards, which leaves us a shorter panel dataset.



                                                         18
4.2    Networks of return migrants
We have established that the high-skilled short-stay migrants scheme especially helps small firms
that increase their hiring of short-stay high-skilled workers. Large firms also benefit from increased
access to high-skilled non-EU workers once they separate from the small firms. This increase
in hiring non-EU immigrant workers leads firms to offshore more to countries from which these
workers came. However, there could be a second, more direct channel that is much harder to
establish in the literature: If a large number of these workers go back to their own countries and
establish networks between firms in their home countries and the Netherlands, we would expect to
see a rise in offshoring to non-EU countries. The reason for the difficulty in directly testing this
mechanism in the literature lies in the scarcity of data to track return migration. We rely on the
foreign migration movement data, which tracks first-generation migrants in the Netherlands and
records whether they return to their home countries. To investigate whether the return of emigrants
from the Netherlands is a potential mechanism to explain the increase in offshoring, we merge the
foreign migration movement information with firms. We select all first-generation immigrants who
have reported emigration and have a duration of fewer than 3 years in the Netherlands. We do this
sample restriction to minimize the impact of non-EU immigrant workers who have settled in the
Netherlands for a long time. We merge them with the matched employer-employee dataset using
personal ID and the earliest year of employment in the Netherlands. We collapse the employer-
employee matched dataset to firm-year data. For the analysis, we only include firms that have at
least 1 new hire leaving the Netherlands within 3 years, which makes up about 13% of all firms.
The outcome variables we are interested in are the total number of short-stay emigrants who leave
the country and the total number of short-stay emigrants who return to their home countries. Note
that the total number of short-stay emigrants who leave the country includes workers who return to
their home countries or emigrate to other countries.
    We check whether return migration increased after the short-stay high-skilled migrants scheme
was implemented by running equation 6 using a simple difference-in-difference. Yijmt is then the
number of non-EU emigrants from firm i in industry j in the region m who leave the Netherlands
within 3 years of year t or the number of non-EU emigrants from firm i in industry j in the region
m who leave the Netherlands and return to their home countries within 3 years of year t.
    Despite the rare wealth of information on return migration in the data, there are a few limitations
of this data, which restrict the types of analysis we can do. The source of the return migration
information is the foreign migration movement dataset described in section 2.2. About 1/3 of
non-EU individuals in the data do not deregister when they leave the country and therefore do
not appear in our data. When we merge the return migration data to the employer-employee data
and then further to the firm-year data, we are left with a small sample of firms as reported in the
regression results in table 7.

                                                  19
    From table 7, we see that after the short-stay high-skilled migrants policy was implemented,
non-EU workers employed in the high-skilled sector returned to their home countries at a higher
rate than non-EU workers in the low-skilled sector (columns 1 and 2 of table 7). We do not find
similar patterns for EU return migrants (columns 3 and 4 of table 7).
    To summarize, we find evidence that smaller firms take advantage of the high-skilled short-stay
migrants scheme. They hire and fire a large number of non-EU workers, and at least some of these
workers go back to their home countries, thereby increasing offshoring from the Netherlands to
their home countries. This is consistent with our findings in section 3.3 that the main increase in
offshoring is driven by the extensive margin.


5     Robustness checks
In this section, we first test whether other concurrent policies implemented around the same time
as the high-skilled migrants scheme can threaten our identification. We then test the sensitivity of
our results to alternative definitions of offshoring. Finally, we address different concerns regarding
our study sample, such as the attrition of firms and the differential hiring opportunities of firms in
the border cities.


5.1   Concurrent policies
We check two potential threats to our identification in table 11 in appendix A.3, by considering two
other immigrant-related policies taking place during the analysis period 2010-2016.
     First, we take into account one concurrent policy on the Dutch expatriate tax regime (known as
the 30% ruling). Foreign workers with special skills or expertise which are scarce in the Dutch labor
market are entitled to tax subsidies. Some amendments were issued in 2012, among which EU im-
migrant workers living close to the Dutch border are directly affected. The amendment stated that
foreign workers living in the border area of 150 kilometers from the Dutch borders no longer qual-
ify for the 30% ruling. This distance range mainly applies to Belgium, Germany, and Luxembourg.
If high-skilled immigrant workers from the three countries respond strongly and stop working at
Dutch companies, firms’ hiring strategy for foreign workers might differ under a high-skilled EU
worker shortage. Here we explicitly test for the change in firm-level immigrant share from Bel-
gium, Germany, and Luxembourg. Column (2) shows that there are no significant differences in
employment shares from these three countries before and after the 2012 policy change.
     Second, we consider another concurrent policy taking place in 2014. The migration restriction
lift was issued for Romanians and Bulgarians working in EU countries in 2014. Although Roma-
nians and Bulgarians have had the right to settle freely in the EU since the countries’ accession


                                                 20
in 2007, they still needed a permit to work in EU member states. This condition was added to
prevent mass migration out of these countries into other EU countries and remained in force until
2014. Since the average percentage of high-skilled workers from Romania and Bulgaria is higher
than the national average for the foreign population in the Netherlands, it is important to check if
this policy led to changes in firm-level hiring of Romanians and Bulgarians differently across high
and low-skill industries. If we see an increasing pattern after 2014, this might also directly affect
the firm-level hiring of non-EU immigrant workers. However, the result does not show significant
changes, alleviating the concern about the impact of this migration restriction lift on firm-level
hiring outcomes.


5.2   Alternative definitions of offshoring
Following Bernard et al. (2020) who show that firms do not necessarily only offshore inputs but
can also offshore final goods, in our main analysis, we do not restrict our offshoring measures to
only intermediate inputs. We test the robustness of our results to some variations in the offshoring
definition. First, according to Hummels et al. (2018), offshoring is about intermediate inputs (or
tasks) used for production, not final goods used for consumption. We, therefore, repeat our main
specification where we define offshoring as the imports of only intermediate goods, as classified by
the Broad Economic Categories (BEC), in the same HS-4 digit category of firm exports. Second,
firms may import intermediate products belonging to a different HS-4 digit code. Although this
does not constitute offshoring, which involves a notion of the existence of domestic production
capability for the product in question, this gives us an idea whether the imports of intermediates,
in general, increased following an influx of high-skilled short-term migrants. We directly calculate
the total imports of intermediates of firms from a country according to BEC classifications. The
results barely change, as shown in table 12 in appendix A.3.


5.3   Alternative samples
We address four possible concerns regarding sample selection. First, the attrition rate of firms might
bias our results if entering and exiting firms have particular characteristics related to offshoring
measures. To alleviate the influences from the inflow and outflow of firms, we pick a sub-sample
of firms registered continuously in the data between 2010 and 2016. Second, our results might
also be affected by firms clustered in several big cities. Potentially, these cities are more active
in the economy and possess more international networks via immigrant workers and multinational
firms. To reduce the impact of these cities, we exclude the four biggest cities (“Big 4”) in the
Netherlands, i.e., Amsterdam, Rotterdam, Den Haag, and Utrecht. Third, firms’ relocation across
cities might be a strategic response to the inflow of immigrants if firms have already planned to

                                                 21
adjust offshoring decisions. Firms with multiple plants are also likely to adjust offshoring decisions
more easily. In order to minimize the impact of these channels, we restrict our sample to firms that
never changed their location or firms that have only one plant throughout the period 2009-2016.
Lastly, the Netherlands is a small country, and cross-border commuting trips for jobs are common.
Firms located in border cities face a larger labor market, including EU-immigrant workers from
countries nearby, especially Belgium, Germany, and Luxembourg. These firms in border cities
may be systematically different from firms in non-border cities in hiring strategies of the high-
skilled non-EU immigrants. For example, since these firms have better access to EU workers, they
may care less about the non-EU high-skilled migrants scheme. On the other hand, if the EU and
non-EU workers are complements, they may like to hire more non-EU workers compared to firms
in non-border areas. To alleviate this concern, we remove firms in border cities. Both the extensive
and the intensive results are robust across various choices of sub-samples, as shown in table 13 in
appendix A.3.
    In addition, we expand the sample to the universe of firms in all sectors and check if our results
are robust. The Dutch industry code is defined by the main activity of a firm, and therefore firms
that appear in the data as non-manufacturing firms could still be engaged in offshoring activities.
Previous studies also allow for such possibilities (Olney and Pozzoli, 2019). Results of table 14 in
appendix A.3 show that the estimates are statistically significant and have very similar magnitudes
as those in the main analysis with the manufacturing sector only.


6    Conclusion
This paper examines how firms respond to increased access to short-stay high-skilled immigrant
workers made available through a national change in visa policy by changing their offshoring de-
cisions. Using matched employer-employee data covering the universe of Dutch firms from 2010-
2016, we find that an influx of high-skilled immigrants from non-EU countries increased offshoring
to their origin countries. We establish a new mechanism via which smaller firms directly benefit
from the short-stay migrants scheme: They hire more short-stay high-skilled workers, a fraction of
whom return to their home countries, thereby establishing direct offshoring links. Another section
of the short-stay non-EU workers switches to working in large firms after separating from smaller
ones. These workers also foster connections between firms in the Netherlands and firms in their
home countries, leading to an increase in offshoring. Our research shows that it is important to
account for the exact nature of the change in immigration policy in order to understand its effect
on offshoring and the labor market. The nature and duration of immigration policies also matter in
understanding which types of firms are more likely to be directly affected by immigration policies,
which in the case of the short-stay high-skilled migrants scheme in the Netherlands are the small


                                                 22
firms. However, given the movements of workers in the labor market, large firms also benefit from
the scheme, underscoring the need for policy-makers to consider the direct and indirect effects of a
policy.




                                                23
References
Andrews, M., T. Schank, and R. Upward (2017). Do foreign workers reduce trade barriers? Mi-
  croeconomic evidence. The World Economy 40(9), 1750–1774.

Ariu, A. (2022). Foreign workers, product quality, and trade: Evidence from a natural experiment.
  Journal of International Economics 139, 103686.

                               ¨ uzel, and H. Rapoport (2019). Migration and knowledge dif-
Bahar, D., A. Hauptmann, C. Ozg¨
  fusion: The effect of returning refugees on export performance in the former Yugoslavia. The
  Review of Economics and Statistics, 1–50.

Bahar, D. and H. Rapoport (2018). Migration, Knowledge Diffusion and the Comparative Advan-
  tage of Nations. The Economic Journal 128(612), F273–F305.

Bastos, P. and J. Silva (2012).   Networks, firms, and trade.      Journal of International Eco-
  nomics 87 (2), 352–364.

Berkhout, E., A. Heyma, M. Volkerink, S. van der Werff, et al. (2015). Attracting and retaining
  highly skilled migrants in the Netherlands. SEO-report 88.

Bernard, A. B., T. C. Fort, V. Smeets, and F. Warzynski (2020). Heterogeneous Globalization:
  Offshoring and Reorganization. Technical report, National Bureau of Economic Research.

Boustan, L. P. (2010). Was Postwar Suburbanization “White Flight”? Evidence from the Black
  Migration. The Quarterly Journal of Economics 125(1), 417–443.

Brinatti, A. and N. Morales (2021). Firm Heterogeneity and the Impact of Immigration: Evidence
  from German Establishments.

Burchardi, K. B., T. Chaney, and T. A. Hassan (2019). Migrants, Ancestors, and Foreign Invest-
  ments. The Review of Economic Studies 86(4), 1448–1486.

Callaway, B. (2022).      Difference-in-Differences for Policy Evaluation.       arXiv preprint
  arXiv:2203.15646.

Card, D. (2001). Immigrant Inflows, Native Outflows, and the Local Labor Market Impacts of
  Higher Immigration. Journal of Labor Economics 19(1), 22–64.

Cardoso, M. and A. Ramanarayanan (2022). Immigrants and exports: Firm-level evidence from
                                                           economique 55(3), 1250–1293.
  Canada. Canadian Journal of Economics/Revue canadienne d’´



                                               24
Cascio, E. U. and E. G. Lewis (2012). Cracks in the Melting Pot: Immigration, School Choice, and
  Segregation. American Economic Journal: Economic Policy 4(3), 91–117.

Cerveny, J. and J. C. Van Ours (2013). Unemployment of Non-Western Immigrants in the Great
  Recession. De Economist 161(4), 463–480.

Cohen, L., U. G. Gurun, and C. Malloy (2017). Resident Networks and Corporate Connections:
  Evidence from World War II Internment Camps. The Journal of Finance 72(1), 207–248.

Corvers, F., M. Hensen, and D. Bongaerts (2009). Delimitation and Coherence of Functional and
  Administrative Regions. Region Studies 43(1), 19–31.

Egger, P. H., K. Erhardt, and A. Lassmann (2019). Immigration and firms’ integration in interna-
  tional production networks. European Economic Review 111, 1–34.

Erbahar, A. and O. T. Gencosmanoglu (2021). Migrants and Global Value Chains: Evidence from
  Dutch Firms.

European Migration Network (2013). Attracting Highly Qualified and Qualified Third-Country
  Nationals. European Commission.

Fan, J., E. Lee, and V. Smeets (2022). High-Skill Immigration, Offshore R&D, and Firm Dynamics.

Felbermayr, G., V. Grossmann, and W. Kohler (2015). Migration, International Trade, and Cap-
  ital Formation: Cause or Effect? In Handbook of the Economics of International Migration,
  Volume 1, pp. 913–1025. Elsevier.

Felbermayr, G. J. and F. Toubal (2012). Revisiting the Trade-Migration Nexus: Evidence from
  New OECD Data. World Development 40(5), 928–937.

Glennon, B. (2020). How Do Restrictions on High-Skilled Immigration Affect Offshoring? Evi-
  dence from the H-1B Program. SSRN Electronic Journal, 1–47.

Gould, D. M. (1994). Immigrant Links to the Home Country: Empirical Implications for U.S.
  Bilateral Trade Flows. The Review of Economics and Statistics, 302–316.

Grossman, G. M., E. Rossi-Hansberg, et al. (2006). The Rise of Offshoring: It’s Not Wine for
  Cloth Anymore. The New Economic Geography: Effects and Policy Implications 2006.

Hatzigeorgiou, A. and M. Lodefalk (2016). Migrants’ Influence on Firm-level Exports. Journal of
  industry, competition and trade 16(4), 477–497.



                                              25
Head, K. and J. Ries (1998). Immigration and Trade Creation: Econometric Evidence from Canada.
  Canadian Journal of Economics, 47–62.

Hiller, S. (2013). Does immigrant employment matter for export sales? Evidence from Denmark.
  Review of World Economics 149(2), 369–394.

Hummels, D., R. Jogersen, J. Munch, and C. Xiang (2014). The Wage Effects of Offshoring:
  Evidence from Danish Matched Worker-Firm Data. American Economic Review 104(6), 1597–
  1629.

Hummels, D., J. R. Munch, and C. Xiang (2018). Offshoring and Labor Markets. Journal of
  Economic Literature 56(3), 981–1028.

Lodefalk, M. (2016). Temporary expats for exports: micro-level evidence. Review of World Eco-
  nomics 152(4), 733–772.

Marchal, L. and C. Nedoncelle (2019). Immigrants, occupations and firm export performance.
 Review of International Economics 27 (5), 1480–1509.

Melitz, M. J. (2003). The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry
 Productivity. Econometrica 71(6), 1695–1725.

Mitaritonna, C., G. Orefice, and G. Peri (2017). Immigrants and firms’ Outcomes: Evidence from
 France. European Economic Review 96, 62–82.

Moriconi, S., G. Peri, and D. Pozzoli (2020). The role of institutions and immigrant net-
 works in firms’ offshoring decisions. Canadian Journal of Economics/Revue canadienne
   economique 53(4), 1745–1792.
 d’´

Olney, W. W. and D. Pozzoli (2019). The Impact of Immigration on Firm-Level Offshoring. Review
  of Economics and Statistics 24(3), 272–284.

Orefice, G., H. Rapoport, and G. Santoni (2021). How Do Immigrants Promote Exports? Networks,
  Knowledge, Diversity.

Orrenius, P. M., M. Zavodny, and A. T. Abraham (2020). The Effect of Immigration on Business
  Dynamics and Employment.

Ottaviano, G. I., G. Peri, and G. C. Wright (2018). Immigration, trade and productivity in services:
  Evidence from U.K. firms. Journal of International Economics 112, 88–108.

                       ezina (2018). Migrant Networks and Trade: The Vietnamese Boat People
Parsons, C. and P.-L. V´
  as a Natural Experiment. The Economic Journal 128(612), F210–F234.

                                                26
Peri, G. and F. Requena-Silvente (2010). The Trade Creation Effect of Immigrants: Evidence
  from the Remarkable Case of Spain. Canadian Journal of Economics/Revue canadienne
  d’´economique 43(4), 1433–1459.

Rauch, J. E. (1999). Networks versus markets in international trade. Journal of international
  Economics 48(1), 7–35.

Rauch, J. E. and V. Trindade (2002). Ethnic Chinese Networks in International Trade. Review of
  Economics and Statistics 84(1), 116–130.

Steingress, W. (2018). The causal impact of migration on US trade: Evidence from political
                                                             economique 51(4), 1312–1338.
  refugees. Canadian Journal of Economics/Revue canadienne d’´

 ezina, P.-L. (2012). How Migrant Networks Facilitate Trade: Evidence from Swiss Exports.
V´
  Swiss Journal of Economics and Statistics 148(3), 449–476.




                                             27
Tables

                                   Table 1: Descriptive statistics

                                                                            (1)         (2)
          VARIABLES                                                        mean         sd
               A-Firm-year-country data 2009-2016 manufacturing sector
          Whether firms offshored                                          0.0316 0.175
          Offshored value (in 1000 euros) for offshoring firms              5,803 78,003
          Immigrant share from a specific non-EU country                   0.00162 0.0111
          Male                                                             0.792   0.180
          Age                                                              43.02   4.483
          Percentage of bachelor degree among natives                      0.163   0.166
          Tenure (in years)                                                7.786   2.344
          Work experience (in years)                                       11.64   2.228
          Log of average daily income per employee                         4.583   0.301
          Log of capital stock per employee                                7.365  2.646
          Firm size between 50 and 100                                     0.170   0.376
          Firm size larger than 100                                        0.186   0.389
          Multi-establishment                                              0.383  0.486
          High-skill                                                       0.127  0.334

                            B-Firm-year data 1999-2016 all sectors
          Non-EU immigrant share                                          0.0565      0.0971
          EU immigrant share                                              0.0215      0.0516
          Firm size smaller than 50                                        0.716      0.451
          Firm size between 50 and 100                                     0.136       0.343
          Firm size larger than 100                                        0.148       0.355
          Age                                                              38.94       6.274
          Male                                                             0.640       0.282
          Percentage of bachelor degree among natives                      0.235       0.264
          Log of average daily income per employee                         4.319       0.464
          Log of capital stock per employee                                5.359      3.873
          High-skill                                                       0.210      0.407

          Notes: In panel A, offshoring variables are only available from 2010 onwards and off-
         shored value is zero for non-offshoring firms. There are 1,452,239 observations covering
         5,912 firms and 43 non-EU countries. In panel B, there are 571,732 observations covering
         52,358 firms.




                                                   28
                                          Table 2: First stage results

                                                     (1)               (2)              (3)              (4)
 VARIABLES                                           ex                ex                in               in

 HS*Post                                        0.00122***       0.00175***        0.00244***       0.00233***
                                                (0.000105)       (0.000155)        (0.000307)       (0.000631)
 Immigrant Share 2007*HS                                           0.818***                           0.606***
                                                                   (0.0295)                            (0.114)
 Immigrant Share 2007*Post                                         0.708***                           0.669***
                                                                   (0.0155)                           (0.0648)
 Immigrant Share 2007*HS*Post                                     -0.792***                          -0.543***
                                                                   (0.0284)                            (0.117)
 Immigrant Share 1999*HS                          0.709***                           0.699***
                                                  (0.0656)                            (0.103)
 Immigrant Share 1999*Post                        0.509***                           0.590***
                                                  (0.0171)                           (0.0553)
 Immigrant Share 1999*HS*Post                    -0.582***                          -0.779***
                                                  (0.0333)                           (0.0924)

 Observations                                     666,242           919,383           25,731           35,072
 Industry/ Region/ Year Fixed Effects               yes               yes              yes              yes
 Firm Fixed Effects                                 yes               yes              yes              yes
 Firm and Workforce Characteristics                 yes               yes              yes              yes
 Destination Fixed Effects                          yes               yes              yes              yes
 Mean Immigrant Share 1999                        0.00200                            0.00260
 Mean Immigrant Share 2007                                          0.00220                           0.00300
 Notes: The table reports the results of estimating equation 3 for firms in the manufacturing sector. Columns (1) and
(2) show results at the extensive margin while columns (3) and (4) show results for the intensive margin. Robust
standard errors in parentheses. Standard errors are clustered at 2-digit SBI industry code. *** p<0.001, ** p<0.01,
* p<0.05




                                                        29
                       Table 3: Offshoring results: Manufacturing only

                                                     (1)             (2)           (3)           (4)
                                                        A-Extensive margin of offshoring
 VARIABLES                                          OLS             OLS            IV             IV

 Lagged immigrant share within firm               0.308***        0.245***       0.276**      0.285***
                                                 (0.0678)        (0.0648)       (0.0954)     (0.0717)

 Observations                                    1,192,476      1,182,414       666,242       919,383
 Industry/ Region/ Year Fixed Effects                yes            yes            yes           yes
 Firm Fixed Effects                                  no             yes            yes           yes
 Firm and Workforce Characteristics                  no             yes            yes           yes
 Destination Fixed Effects                           yes            yes            yes           yes
 Mean Y                                            0.0320         0.0320         0.0390        0.0380
 Year of Instruments                                                              1999          2007
 Second Stage: F-stat p value                                                   0.00613       0.0128
 First stage: KP F-stat                                                           257.1         684.3
                                                        B-Intensive margin of offshoring
 Lagged immigrant share within firm                12.54**        12.32**         16.52*        13.18*
                                                  (3.773)        (4.137)         (6.325)       (5.698)

 Observations                                      38,113         37,529         25,731        35,072
 Industry/ Region/ Year Fixed Effects                yes            yes            yes           yes
 Firm Fixed Effects                                  no             yes            yes           yes
 Firm and Workforce Characteristics                  no             yes            yes           yes
 Destination Fixed Effects                           yes            yes            yes           yes
 Mean Y                                             10.11          10.11          10.13         10.14
 Year of Instruments                                                              1999          2007
 Second Stage: F-stat p value                                                   1.44e-08      1.31e-07
 First stage: KP F-stat                                                           61.83         47.97
 Notes: The table reports the results of estimating equation 4 for firms in the manufacturing sector. Panels
A and B show the extensive and intensive margins respectively. Robust standard errors in parentheses.
Standard errors are clustered at 2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05




                                                   30
                 Table 4: Offshoring to EU countries: Manufacturing only

                                                             (1)           (2)          (3)
         VARIABLES                                           ex             in         ex+in

         Lagged immigrant share within firm                 -0.0147       -0.242*        -0.253
                                                           (0.0263)      (0.113)       (0.355)

         Observations                          919,383                   361,931 919,383
         Industry/ Region/ Year Fixed Effects    yes                       yes      yes
         Firm Fixed Effects                      yes                       yes      yes
         Firm and Workforce Characteristics      yes                       yes      yes
         Destination Fixed Effects               yes                       yes      yes
         Mean Y                                 0.390                     15.38    6.003
         Second Stage: F-stat p value         0.000882                   0.00161 1.74e-05
         First stage: KP F-stat                 684.3                     1868     684.3
        Notes: The table reports the results of estimating equation 4 for firms in the manu-
       facturing sector offshoring to EU countries. Columns (1),(2), and (3) show the results
       for the extensive, intensive, and combined extensive and intensive offshoring margin.
       For column (3), the dependent variable is log(1+offshoring). Robust standard errors in
       parentheses. Standard errors are clustered at 2-digit SBI industry code. *** p<0.001,
       ** p<0.01, * p<0.05


                      Table 5: Offshoring results: Small verus big firms

                                                     (1)           (2)           (3)           (4)
 VARIABLES                                           ex            ex             in            in

 Lagged immigrant share within firm                0.239*        0.309*        3.842         22.82**
                                                 (0.0853)       (0.114)      (4.251)        (6.558)

 Observations                                     478,160      421,959        7,391         26,961
 Firm size                                         Small          Big         Small           Big
 No. of firms                                        1961         1612          757           1191
 Industry/ Region/ Year Fixed Effects                yes          yes          yes            yes
 Firm Fixed Effects                                  yes          yes          yes            yes
 Firm and Workforce Characteristics                  yes          yes          yes            yes
 Destination Fixed Effects                           yes          yes          yes            yes
 Mean Y                                            0.0160       0.0640        9.631          10.29
 Second Stage: F-stat p value                     0.00524      0.00565      0.000617       0.000782
 First stage: KP F-stat                            325.6        536.7         53.23          232.6
 Notes: The table reports the results of estimating equation 4 for small and big firms in the manufacturing
sector separately. Columns (1) and (2) report the results for the extensive margin for small and big firms
respectively. Columns (3) and (4) report the results for the intensive margin for small and big firms
respectively. Robust standard errors in parentheses. Standard errors are clustered at 2-digit SBI industry
code. *** p<0.001, ** p<0.01, * p<0.05


                                                   31
        Table 6: Job separations of non-EU workers from small firms that move to
        bigger firms next year

                                                            (1)             (2)           (3)
          VARIABLES                                       Non-EU            EU           Native

          HS*Post                                         0.0180*        0.00852       0.00928
                                                         (0.00701)      (0.00489)      (0.0110)

          Observations                                    121,468        121,468        121,468
          R-squared                                        0.662          0.575          0.717
          Industry/ Region/ Year Fixed Effects              yes            yes            yes
          Firm Fixed Effects                                yes            yes            yes
          Firm and Workforce Characteristics                yes            yes            yes
         Notes: The table reports the results of estimating equation 6 for the sample of firms with
        at-least 1 job separation in a year. The dependent variable is the log (1+job separations to
        bigger firms in the next year). Columns (1), (2), and (3) report the results for Non-EU, EU,
        and native workers respectively. Robust standard errors in parentheses. Standard errors
        are clustered at 2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05




                                      Table 7: Return migration

                                                  (1)              (2)           (3)              (4)
 VARIABLES                                      Emigrants       Returnees      Emigrants       Returnees

 HS*Post                                          0.132*          0.108*          0.00911       0.00807
                                                 (0.0575)        (0.0451)         (0.141)       (0.143)

 Observations                                     6,180           6,180            4,197          4,197
 R-squared                                        0.767           0.583            0.685          0.586
 Worker type                                     Non-EU          Non-EU             EU             EU
 Industry/ Region/ Year Fixed Effects              yes             yes              yes            yes
 Firm Fixed Effects                                yes             yes              yes            yes
 Firm and Workforce Characteristics                yes             yes              yes            yes
 Notes: We report the results from estimating equation 6 with dependent variables the log 1+number of
emigrants (columns (1) and (2)) and the log number of returnees. Firms that have at least 1 new hire leaving
the Netherlands within 3 years are included in the sample. Robust standard errors in parentheses. Standard
errors are clustered at 2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05




                                                    32
Figures

                Figure 1: Total number immigrants who emigrate within 5 years




Notes: The Y-axis shows the total number of immigrants in the Netherlands who emigrate within
5 years. We use this restriction (5 years) to eliminate non-EU workers who have settled in the
Netherlands for a long time. This information is obtained from the migration motives dataset.




                                             33
                         Figure 2: Percentage of immigrants over time




Notes: The top panel shows the total number of non-EU immigrants out of the total Dutch popula-
tion over time. The bottom panel shows the total number of EU immigrants out of the total Dutch
population over time.




                                              34
                           Figure 3: Fast-growing countries of origin




Notes: The dots show the growth rates in immigrant share relative to 2009 for 10 countries with
the highest growth rates of immigrants in the Netherlands.




                                              35
                  Figure 4: Event study of firm share of non-EU employment




Notes: These figures plot the event study coefficients from estimating equation 2 for the share of
Non-EU workers within a firm.




                                              36
               Figure 5: Event study of firm share of EU and native employment




Notes: These figures plot the event study coefficients from estimating equation 2 for the share of
EU (top) and native (bottom) workers within a firm.




                                              37
               Figure 6: Event study of the size of Non-EU employment in firms




Note: These figures plot the event study coefficients from estimating equation 5 for small and big
firms, respectively.




                                              38
    Figure 7: Event study of the number of Non-EU new hires and separations in small firms

                                           New hires




                                        Job separations




Note: These figures plot the event study coefficients from estimating equation 5 for the log number
of new hires and separations, respectively.

                                               39
     Figure 8: Event study of the number of Non-EU new hires and separations in big firms

                                           New hires




                                        Job separations




Note: These figures plot the event study coefficients from estimating equation 5 for the log number
of new hires and separations, respectively.

                                               40
A     Online appendix

A.1    Detailed information about foreign migration movement data
We rely on municipal records, which track the dynamics of first-generation migrants in the Nether-
lands and track whether they return to their home countries. A person is identified in an occurrence
of immigration/emigration if he or she newly registers/ deregisters the place of residence at the
municipality office. Emigrants who do not de-register from the municipality might face certain
consequences, such as tax bills and inconveniences in applying for visas next time. Therefore,
more than half of the emigrants (about 2/3) do inform the municipality about their emigration and
fill in the destination country after leaving the Netherlands. The percentage of returnees among
emigrants is about 50% for non-EU workers and about 70% for EU workers.
    Note that the municipality updates data from time to time. For migrants who did not inform
the municipality, the administrative removal can be done by the municipality of residence. If the
municipality detects that an immigrant cannot be reached for a long period and data from other
sources also confirm the absence of this person, it can classify this person as an emigrant and labels
the status of the person as “administrative removal”. In our mechanism analysis, we exclude these
emigrants identified by the municipality to reduce measurement errors.




                                                 41
     A.2 CBS data sources

                                 Table 8: Full list of data sources in Statistics Netherlands

      Main data   Dutch names                      English translation
      Firm        ABR                              General business register
                  BAANPRSJAARBEDRAGTAB             Annual wages of employees
                  INVESTERINGEN                    Total investments in tangible and intangible fixed assets
      Employee    BAANKENMERKENBUS                 Characteristics of employees’ jobs
                  GBAPERSOONTAB                    Individual characteristics of all persons registered in the municipal records
                  HOOGSTEOPLTAB                    Highest attained and highest followed educational level in the Netherlands
                  GBAADRESOBJECTBUS                Address characteristics of persons ever registered in the municipal records
                  VSLGWBTAB                        Municipal, district and neighborhood codes of a residential object
                  GBAMIGRATIEGEBEURTENISBUS        Migration characteristics of persons ever registered in the municipal records
                  MIGMOTIEFBUS                     Migration motives of immigrants with a foreign nationality
      Trade       IHG                              International trade in goods
42




                  PRODCOM                          Industrial products by product group
A.3       Supplementary tables

            Table 9: Trends in immigration and the 2012 policy change

                                                    (1)              (2)            (3)
 VARIABLES                                        Non-EU             EU            Native

 HS*Year 1999                                    0.000413        0.000381        -0.000794
                                                 (0.00197)       (0.00126)        (0.00248)
 HS*Year 2000                                   -0.000653        0.000665         -1.24e-05
                                                 (0.00171)       (0.00116)        (0.00210)
 HS*Year 2001                                    -0.00106        0.000778         0.000284
                                                 (0.00145)       (0.00112)        (0.00170)
 HS*Year 2002                                   -0.000797        0.000560         0.000237
                                                 (0.00138)       (0.00102)        (0.00161)
 HS*Year 2003                                   -0.000730        0.000594         0.000135
                                                 (0.00151)      (0.000977)        (0.00186)
 HS*Year 2004                                    -0.00183        0.000461          0.00136
                                                 (0.00156)       (0.00109)        (0.00202)
 HS*Year 2005                                    -0.00114        0.000721         0.000418
                                                 (0.00139)       (0.00107)        (0.00203)
 HS*Year 2006                                    -0.00200        0.000679          0.00132
                                                 (0.00108)       (0.00107)        (0.00165)
 HS*Year 2007                                   -0.00175*        0.000123          0.00163
                                                (0.000683)      (0.000761)        (0.00108)
 HS*Year 2008                                   -0.000705        0.000514         0.000191
                                                (0.000803)      (0.000871)        (0.00111)
 HS*Year 2009                                   -0.000933        0.000289         0.000644
                                                (0.000654)      (0.000557)       (0.000819)
 HS*Year 2010                                   -0.000750        0.000158         0.000592
                                                (0.000431)      (0.000363)       (0.000540)
 HS*Year 2012                                    0.000350        -8.75e-05       -0.000262
                                                (0.000562)      (0.000398)       (0.000738)
 HS*Year 2013                                     0.00116        0.000559         -0.00172
                                                (0.000653)      (0.000642)       (0.000963)
 HS*Year 2014                                    0.00192*         0.00102        -0.00294*
                                                (0.000729)       (0.00101)        (0.00139)
 HS*Year 2015                                   0.00246**        0.000895        -0.00336*
                                                (0.000776)       (0.00119)        (0.00155)
 HS*Year 2016                                    0.00269*        0.000285         -0.00298
                                                 (0.00123)       (0.00138)        (0.00203)

 Observations                                     561,905         561,905         561,905
 Industry/ Region/ Year Fixed Effects               yes             yes             yes
 Firm Fixed Effects                                 yes             yes             yes
 Firm and Workforce Characteristics                 yes             yes             yes
 Notes: The table reports the results of the event study of the share of non-EU workers for firms
in all sectors. The omitted category is HS*Year 2011. Robust standard errors in parentheses.
Standard errors are clustered at 2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05




                                                                43
                            Table 10: Offshoring results: Another identification

                                                     (1)            (2)            (3)           (4)
 VARIABLES                                           ex             ex              in            in

 Lagged immigrant share within firm                0.276**       0.285***        16.44*        13.13*
                                                  (0.0954)       (0.0717)       (6.315)       (5.669)
 HS*Post                                          0.00176       0.000922         0.116         0.133
                                                 (0.00193)      (0.00133)      (0.0784)      (0.0642)

 Observations                                     666,242        919,383        25,731        35,072
 Industry/ Region/ Year Fixed Effects                yes            yes           yes           yes
 Firm Fixed Effects                                  yes            yes           yes           yes
 Firm and Workforce Characteristics                  yes            yes           yes           yes
 Destination Fixed Effects                           yes            yes           yes           yes
 Year of Instruments                                1999           2007          1999          2007
 Mean Y                                            0.0390         0.0380         10.13         10.14
 Second Stage: F-stat p value                     0.00188        0.00582           0         7.87e-08
 First stage: KP F-stat                              330           912.2         41.78         62.52
 Notes: The table reports the results of estimating equation 4 for firms in the manufacturing
sector but with post ∗ γs added to the second stage. Ex. and in. stand for extensive and intensive
margins, respectively. Robust standard errors in parentheses. Standard errors are clustered at
2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05




                                                           44
                        Table 11: Consider other concurrent policies

                                                       (1)                            (2)
 VARIABLES                                 Belgium/Germany/Luxembourg           Romania/Bulgaria

 HS*Year 1999                                         -0.000196                       -0.000120
                                                      (0.000523)                     (0.000102)
 HS*Year 2000                                         -0.000352                       -0.000180
                                                      (0.000479)                      (9.33e-05)
 HS*Year 2001                                         -0.000278                       -0.000122
                                                      (0.000554)                     (0.000114)
 HS*Year 2002                                         -0.000274                       -0.000175
                                                      (0.000597)                     (0.000106)
 HS*Year 2003                                         -0.000167                     -0.000264**
                                                      (0.000505)                      (9.79e-05)
 HS*Year 2004                                          9.31e-06                       -0.000214
                                                      (0.000523)                     (0.000108)
 HS*Year 2005                                          0.000221                    -0.000284***
                                                      (0.000518)                      (8.23e-05)
 HS*Year 2006                                          -9.62e-05                     -0.000204*
                                                      (0.000469)                      (8.95e-05)
 HS*Year 2007                                          -3.61e-05                    -0.000190**
                                                      (0.000358)                      (6.53e-05)
 HS*Year 2008                                          4.60e-05                       -0.000109
                                                      (0.000366)                      (8.78e-05)
 HS*Year 2009                                          0.000232                       -7.85e-05
                                                      (0.000405)                      (7.57e-05)
 HS*Year 2010                                          5.82e-05                       -0.000118
                                                      (0.000256)                      (6.13e-05)
 HS*Year 2012                                          -9.13e-06                       4.49e-05
                                                      (0.000185)                      (7.43e-05)
 HS*Year 2013                                          -3.39e-05                     0.000247*
                                                      (0.000178)                     (0.000109)
 HS*Year 2014                                          0.000397                        0.000180
                                                      (0.000257)                     (0.000170)
 HS*Year 2015                                          0.000534                        0.000139
                                                      (0.000313)                     (0.000183)
 HS*Year 2016                                          0.000409                       -9.38e-05
                                                      (0.000396)                     (0.000219)

 Observations                                           561,905                       561,905
 Industry/ Region/ Year Fixed Effects                     yes                           yes
 Firm Fixed Effects                                       yes                           yes
 Firm and Workforce Characteristics                       yes                           yes
 Notes: The table reports the results of the event study of the share of workers from Bel-
gium/Germany/Luxembourg in column (1) and Romania/Bulgaria in column (2) for firms in all sectors.
The omitted category is HS*Year 2011. Robust standard errors in parentheses. Standard errors are clus-
tered at 2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05


                                                        45
       Table 12: Alternative definitions of offshoring using BEC classifications

                                                     (1)          (2)            (1)           (2)
 VARIABLES                                           ex            in            ex             in

 Lagged immigrant share within firm               0.278**        17.15*       0.387***       12.30**
                                                 (0.0747)       (7.584)      (0.0820)       (4.179)

 Observations                         919,383                    21,777      919,383        37,673
 Within the same HS4 of export           yes                       yes          no             no
 Industry/ Region/ Year Fixed Effects    yes                       yes          yes           yes
 Firm Fixed Effects                      yes                       yes          yes           yes
 Firm and Workforce Characteristics      yes                       yes          yes           yes
 Destination Fixed Effects               yes                       yes          yes           yes
 Mean Y                                0.0550                     9.993       0.0610         9.048
 Second Stage: F-stat p value         5.85e-05                  1.98e-06     5.10e-05      8.49e-05
 First stage: KP F-stat                 684.3                     106.8        684.3         25.15
 Notes: The table reports the results of estimating equation 4 with different dependent variables defined
by BEC. In columns (1) and (2), offshoring is defined as the imports of only intermediate goods defined
by BEC in the same HS-4 digit category of firm exports. In columns (3) and (4), we directly construct
the dependent variables by imports of intermediates defined by BEC from a country. Robust standard
errors in parentheses. Standard errors are clustered at 2-digit SBI industry code. *** p<0.001, **
p<0.01, * p<0.05




                                                           46
                            Table 13: Sub-sample analysis with 2007 instruments

                                                  (1)            (2)              (3)              (4)
 VARIABLES                                     Balanced        No Big4    1 location/ 1 plant No border cities
                                                               A-Extensive margin of offshoring
 Lagged immigrant share within firm             0.298*** 0.284***               0.278***               0.232***
                                               (0.0779) (0.0731)               (0.0713)               (0.0530)

 Observations                                  785,180         888,509          895,518               589,014
 Industry/ Region/ Year Fixed Effects             yes             yes              yes                   yes
 Firm Fixed Effects                               yes             yes              yes                   yes
 Firm and Workforce Characteristics               yes             yes              yes                   yes
 Destination Fixed Effects                        yes             yes              yes                   yes
 Mean Y                                         0.0380          0.0380           0.0380                0.0380
 Second Stage: F-stat p value                  0.00704          0.0194           0.0133               0.000429
 First stage: KP F-stat                          564.6           1048              705                  519.5
                                                               B-Intensive margin of offshoring
 Lagged immigrant share within firm              12.31*         13.60*            12.61*                13.13*
                                                (5.880)        (5.954)           (5.765)               (4.977)

 Observations                                   32,214          33,712           33,830                 21,961
 Industry/ Region/ Year Fixed Effects             yes             yes              yes                    yes
 Firm Fixed Effects                               yes             yes              yes                    yes
 Firm and Workforce Characteristics               yes             yes              yes                    yes
 Destination Fixed Effects                        yes             yes              yes                    yes
 Mean Y                                          10.14           10.14            10.14                  10.14
 Second Stage: F-stat p value                  1.21e-08        3.79e-08         5.08e-06               1.12e-07
 First stage: KP F-stat                          46.74           367.8            49.96                  49.58
 Notes: The table reports the results of estimating equation 4 for firms in the manufacturing sector with different sub-
samples. Panels A and B show the extensive and intensive margin of offshoring, respectively. Robust standard errors in
parentheses. Standard errors are clustered at 2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05




                                                          47
                             Table 14: Offshoring results: All sectors

                                                     (1)             (2)            (3)            (4)
                                                 A-Extensive margin of offshoring combined
                                                    OLS            OLS              IV             IV

 Lagged immigrant share within firm                0.240**        0.188**         0.284*         0.280**
                                                  (0.0861)       (0.0693)        (0.120)        (0.105)

 Observations                         6,467,802                 6,453,139      2,759,611       4,733,225
 Industry/ Region/ Year Fixed Effects     yes                       yes            yes             yes
 Firm Fixed Effects                       no                        yes            yes             yes
 Firm and Workforce Characteristics       no                        yes            yes             yes
 Destination Fixed Effects                yes                       yes            yes             yes
 Mean Y                                 0.0150                    0.0150         0.0190          0.0190
 Year of Instruments                                                              1999            2007
 Second Stage: F-stat p value                                                   0.00286        0.000404
 First stage: KP F-stat                                                          91.09            391.5
                                                  B-Intensive margin of offshoring combined
 Lagged immigrant share within firm               13.88***        17.46***       22.42***       22.44***
                                                  (2.244)         (3.589)        (5.001)        (5.713)

 Observations                                      98,704         97,793         51,593          88,350
 Industry/ Region/ Year Fixed Effects                yes            yes            yes             yes
 Firm Fixed Effects                                  no             yes            yes             yes
 Firm and Workforce Characteristics                  no             yes            yes             yes
 Destination Fixed Effects                           yes            yes            yes             yes
 Mean Y                                             10.49          10.49          10.52           10.50
 Year of Instruments                                                              1999            2007
 Second Stage: F-stat p value                                                       0               0
 First stage: KP F-stat                                                           46.43           576.7
 Notes: The table reports the results of estimating equation 4 for firms in all sectors. Panels A and B show
the extensive and intensive margins, respectively. Robust standard errors in parentheses. Standard errors are
clustered at 2-digit SBI industry code. *** p<0.001, ** p<0.01, * p<0.05




                                                           48