On the heterogeneous employment effects of offshoring: identifying productivity and downsizing channels.
Moser, Christoph ; Urban, Dieter ; Di Mauro, Beatrice Weder 等
I. INTRODUCTION
For the most part of the last two decades Germany suffered from a
hangover of the reunification boom, an overvalued exchange rate, high
unemployment, and low growth--so The Economist famously named it the
"Sick Man of Europe." At the same time, German companies were
relocating production, restructuring, and offshoring. The general public
associated such offshoring activities--not only in Germany--with plant
closures which made the headlines and confirmed the perception that
offshoring was a job killer.1 What usually does not make the news is
that such downsizing effects of offshoring may be counterbalanced by
productivity effects in the restructuring firm. Depending on their
relative size, employment changes are likely to be heterogeneous across
offshoring firms. Moreover, possible indirect employment effects on
local suppliers and competitors further complicate the picture.
The theoretical literature has distinguished two main channels
through which offshoring can impact domestic employment (Kohler and
Wrona 2011) (2): a positive productivity effect from cost savings and
gains in competitiveness which allows firms to increase market shares as
well as employment and a negative downsizing effect from the relocation
of production processes abroad. In addition to these direct effects
there may be indirect ones on those firms that do not offshore. In
particular, offshoring firms might substitute domestic suppliers for
foreign ones (supplier-substitution effect) and increase their market
share at the cost of less competitive, domestic competitors
(business-stealing effect). (3)
This article proposes a way to estimate these different channels
through which offshoring may affect employment, using a representative
sample of German establishments, covering the years 1998 to 2004 and 16
sectors. We identify heterogeneous employment changes by different
behaviors of offshoring firms, learn about different adjustment
scenarios for employment, and estimate an overall effect, which accounts
for indirect effects on nonoffshoring firms. Our measure of offshoring
is the increase in an establishment's (4) intermediate inputs from
abroad. This is the result of either (1) relocating parts of production
to foreign affiliates (vertical foreign direct investment [FDI]) or (2)
purchasing these inputs from foreign suppliers (international
outsourcing). (5) Offshoring is measured at the plant level through a
(qualitative) increase of a plant's share of foreign intermediate
inputs in total inputs.
The use of a plant-level measure of offshoring has advantages over
using industry-level measures. Even within narrowly defined sectors,
there is considerable heterogeneity in trade flows and the use of
imported intermediate inputs that cannot be captured by industry-level
measures. Thus, industry measures might suffer from an aggregation bias,
whereas firm-level information can serve to identify direct and indirect
effects of offshoring on employment.
We employ a difference-in-differences matching estimator, a
nonparametric estimator that differs from a standard
difference-in-differences estimator in the weighting function. Based on
the propensity scores, the matching estimator gives those observations
of the control group a bigger weight in the regression that are
according to pretreatment characteristics similarly likely to receive
treatment as a treated observation. There are several applications of
difference-in-differences matching estimators in international trade,
but to the best of our knowledge this is the first paper on offshoring
that discusses how interactions between the control and treatment group
can affect the estimated average treatment effects and violate the
stable unit treatment value assumption (SUTVA).
This article adds to the offshoring literature by looking at
different adjustment scenarios that can occur with offshoring decisions.
Thereby, we identify different channels of offshoring on employment by
varying the treatment and control groups. An innovation is the
construction of three additional types of treatment variables at the
plant level that allow us to identify downsizing and productivity
effects of offshoring on employment: (1) offshoring as defined above
that does not coincide with any simultaneous restructuring, that is, a
partial shut-down, sell-off, or spin-off
(offshoring-without-restructuring), (2) offshoring with such a
simultaneous restructuring event (offshoring-with-restructuring), and
(3) offshoring -without-
restructuring-and-without-supplier-substitution, which is equivalent to
offshoring-without-restructuring, but excludes those firms from the
treatment group with a simultaneous reduction in outsourcing to German
suppliers. While the difference between the first two measures allows us
to isolate the negative downsizing effect on employment, the third
treatment variable identifies the productivity channel, since there is
neither a direct employment effect from downsizing in a German plant,
nor are German suppliers negatively affected. Still, there may be a cost
advantage, rendering such a plant more competitive than its domestic
competitors.
To anticipate our main findings: Our baseline results for the
overall sample (including the restructuring and nonrestructuring
offshoring plants) show that the net employment growth of offshoring
plants is between 2 and 4.5 percentage points higher than nonoffshoring
plants. This positive employment effect indicates that the productivity
effect dominates the downsizing effect for a representative sample of
German plants. But we also find strong and negative employment effects
for those roughly 10% of offshoring plants that experience at the same
time a discrete organizational change, that is, that restructure
(downsizing effect). This simultaneous restructuring is consistent with
the view (but does not prove it) that these establishments relocate
parts of their production abroad. Hence, there is indeed evidence for
heterogeneous employment effects of offshoring in Germany. All our main
results are robust to a number of sensitivity analyses, among others, to
a careful investigation of whether self-selection into offshoring could
confound these treatment effects.
We also find indirect evidence for a productivity effect, since
offshoring firms enjoy increases in sales growth, the export share,
and--to a certain extent--average labor productivity. This is consistent
with, but does not prove the existence of the business-stealing effect.
Furthermore, given the increase in the foreign share of intermediate
inputs, first evidence that offshoring firms' overall input share
(foreign and domestic intermediate inputs relative to sales) remains
constant leaves the door for an important supplier-substitution effect
open. There might be an unobservable negative effect of offshoring on
domestic suppliers' employment if they are substituted by foreign
ones. (6) In a final step, we show that the sector-specific average
treatment effects do not depend on the offshoring intensity in the same
sector. Based on this evidence, it is unlikely that spillover effects
between offshoring and nonoffshoring firms drive the overall results.
The article is organized as follows. Section II discusses the
related literature and measurement of offshoring. Section III presents
the empirical methodology, the dataset, and the identification strategy
for the different channels of offshoring. Section IV shows the main
results of the average treatment effect on the treated (ATT) of
offshoring and some important extensions, and Section V concludes.
II. RELATED LITERATURE AND MEASUREMENT OF OFFSHORING
There is a large and growing empirical literature on the labor
market effects of offshoring. The early literature on offshoring has
employed dependent and independent variables at the industry level.
Feenstra and Hanson (1996, 1999) in two seminal papers measure
offshoring as an increase of the share of imported intermediate inputs
in the total purchase of nonenergy materials of an industry. (7)
Some recent empirical studies by, for instance, Geishecker (2006,
2008), Geishecker and Gorg (2008), Munch and Skaksen (2009), and Senses
(2010) exploit plant- or worker-specific dependent variables, but the
offshoring variable remains industry-specific. (8) This approach has the
major advantage to control for microlevel heterogeneity and allows for
interesting further analysis. But we also know that there is
considerable variation in the usage of imported inputs even within
narrowly defined industries. Hence, a microlevel measure of offshoring
seems indispensable for our main objective to identify different
channels through which offshoring affects employment.
We propose a proxy for offshoring at the plant level by measuring
the qualitative increase in the share of imports in intermediate goods
of an establishment from any sector. Hence, our measure is closest in
spirit to the broad definition in Feenstra and Hanson (1996, 1999), but
more precise in practice by capturing firm heterogeneity. Thereby, we
add to a small but growing literature that accounts for the
heterogeneity in the use of imported intermediate inputs. Some recent
studies have even offered continuous microlevel measures of offshoring.
Biscourp and Kramarz (2007), Lo Turco and Maggioni (2012), and Defevre
and Toubal (2013) use measures in spirit of the broad and narrow
definition of international outsourcing, respectively. Hummels et al.
(2014) use the most detailed offshoring measure to date, but their study
does not focus on employment. (9) Hummels et al. (2014) analyze the
effects of offshoring and exports on wages for Danish manufacturing
firms with more than 50 employees. For a continuous firm-level
offshoring measure, they find that offshoring (but not exporting)
contributes to wage inequality between low- and high-skilled workers.
(10) Balsvik and Birkeland (2012) also rely on a continuous offshoring
measure at the firm-level to investigate wage effects.
Biscourp and Kramarz (2007) provide an important early account of
the impact of offshoring on employment in France (11) and Lo Turco and
Maggioni (2012) investigate these effects in Italy. Both studies are
most closely related to our article due to their firm-level measure of
offshoring similar to Feenstra and Flanson (1996, 1999) and their focus
on employment. Lo Turco and Maggioni (2012) estimate a dynamic panel
model using a System GMM estimator and find a negative employment effect
for traditional Italian manufacturing sectors, if imported intermediate
inputs come from low-income countries.
We are not the first to study the effects of globalization on
employment in Germany, but most studies have so far relied on
industry-level measures of offshoring (see, for instance, Geishecker
2006,2008) or have focused on the consequences of foreign direct
investment using firm-level data. Becker and Muendler (2008) find that
German multinationals expanding their workforce abroad experience fewer
worker separations at home and Becker and Muendler (2010) show that
international wage differentials play a role for domestic employment,
with domestic and foreign employment being substitutes within German
multinationals. Becker, Ekholnt, and Muendler (2013) shed light on
another dimension by documenting an important shift of multinational
enterprises' (MNEs) domestic employment toward nonroutine and
interactive tasks. Wagner (2011) employs an interesting treatment
variable, namely "relocation abroad," and investigates the
employment effects using a matching approach and data from the German
federal statistical office on German manufacturing firms. Wagner (2011)
finds small negative, but statistically not robust employment effects.
We complement these studies with a more direct focus on offshoring. Most
estimates could be driven by horizontal and/or vertical FDI and (except
for Wagner 2011) do not encompass international outsourcing.
Finally, our article is related to two studies that deal with
general equilibrium effects. Ferracci, Jolivet, and van den Berg
(forthcoming) provide an important contribution on how to test for a
potential violation of the SUTVA due to spillover effects between the
treatment and control group. Their study on training programs for
unemployed individuals in France finds evidence in favor of such
spillover effects. In particular, Ferracci, Jolivet, and van den Berg
(forthcoming) show that the average treatment effect depends on the
fraction of treated in the same labor market. Sethupathy (2013) models
and tests for one of the two general equilibrium effects of offshoring
considered in our study. He measures vertical FDI as the share of
intra-firm affiliate sales from Mexico to total sales of U.S. MNEs and
uses an exchange rate and legislative shock in Mexico for identification
of employment and firm performance measures. The results most relevant
to our paper: Sethupathy (2013) finds no evidence for the
business-stealing effect or any significant employment effects.
A comprehensive literature review of the effects of offshoring and
FDI on wages, employment, and firm performance is beyond the scope of
this article.12 Crino (2009), Feenstra (2010), Gorg (2011), Olsen
(2006), and Pfluger et al. (2013) and provide excellent surveys.
III. EMPIRICAL STRATEGY
A. Theoretical Channels of Offshoring on Employment
Offshoring can affect employment of offshoring and nonoffshoring
firms through different channels. We illustrate these channels in Figure
1, with the treatment (control) group below (above) the dotted line. The
decision to offshore is motivated by expected cost advantages. Recent
economic theory suggests that productivity gains due to offshoring can
lead to higher profits and/or higher wages, but do not necessarily lead
to job creation (Grossman and Rossi-Hansberg 2008; Kohler and Wrona
2011). If productivity gains indeed result in positive employment
effects, we will call this a direct productivity effect (A). (13)
An offshoring firm may substitute either in-house production or
domestic intermediate input demand with foreign one. (14) The downsizing
effect of offshoring (B) results in a loss of domestic employment at the
offshoring plant, if own production is replaced by foreign intermediate
inputs. We do not expect any direct, negative employment effects for
offshoring firms, if its own production is not relocated abroad. It is a
priori unclear whether the productivity or downsizing effect dominates
for offshoring firms.
[FIGURE 1 OMITTED]
If offshoring firms substitute domestic for foreign suppliers
(without increasing their usage of overall intermediate inputs), we do
not expect a direct employment effect at the offshoring firm. But
offshoring firms are still expected to save costs analogously to a
productivity gain from offshoring. More importantly, this type of
relocation abroad can lead to an employment loss among domestic
suppliers, which are part of the control group of plants that do not
offshore. We will call this potential spillover effect the
supplier-substitution effect (C). Similarly, offshoring firms might
profit from increased competitiveness, allowing them to gain market
share at home and/or abroad. If these gains in market share are at the
cost of domestic competitors that do not offshore, their employment can
be adversely affected. Sethupathy (2013) models a similar mechanism
calling it business stealing effect (D). It is important to note that
the last two channels imply general equilibrium effects of offshoring on
employment of firms that do not offshore. Imbens and Wooldridge (2009)
write that such general equilibrium effects may, or may not, be a
serious problem in practice. We discuss this issue in the following
section.
B. Empirical Methodology: Matching Estimator
To identify the different channels through which offshoring has an
influence on the employment of the offshoring plant, we employ a
difference-in-differences matching technique. The basic idea of a
matching estimator is to compare outcomes of establishments that
offshore with those nonoffshoring establishments, whose pretreatment
characteristics xi0 indicate a very similar likelihood to offshore (but
which has not realized it in the end). For a given plant, it is exactly
those similar plants that receive a bigger weight in the regressions in
order to mitigate any potential selection bias. In the case of
difference-in-differences matching, we compare before (f') and
after (t) treatment values of the outcome variable [Y.sub.t'] -
[Y.sub.t] = [DELTA]Y between the treatment and control groups and
thereby control for any time-invariant factors.
Matching estimators depend on three crucial assumptions: the
conditional independence assumption (CIA), the common support
assumption, and the stable unit treatment value assumption (SUTVA). We
discuss each of these assumptions in turn.
The CIA states that conditional on observable pretreatment
characteristics assignment to treatment is random and the selection bias
disappears. This is an important, but standard assumption in matching.
We follow Ferracci, Jolivet, and van den Berg (forthcoming) and modify
this assumption in the following way. We condition not only on
plant-characteristics, but also on a vector of aggregate measures
[M.sub.st] that captures the offshoring intensity by computing the
proportion of treated observations in sector s at the time t.
It is important to note that the CIA is not testable, but we
provide some corroborating proof that there is no evidence of violation
of this important assumption. Beyond the usual balancing tests
(discussed in Appendix B and reported in Tables B3-B5), we provide a
placebo test as proposed by Heckman and Hotz (1989). We condition on the
same set of variables as in the baseline specification and estimate the
average treatment effect for the main outcome variable prior to the true
treatment, when by definition no treatment effect can have materialized
yet. Hence, any significant differences before treatment would likely
signal a violation of the CIA due to self-selection effects (but Table
B6 does not indicate so).
The second important assumption, the common support assumption,
basically insures that there exist both treated and control observations
with a positive probability of being treated. This assumption can be
easily enforced in the matching estimates.
Finally, the SUTVA precludes that the outcome of observation i is
affected by the treatment status of other observations j. This
assumption is at odds with general equilibrium effects such as the
supplier-substitution and business-stealing effect, since employment of
firms that do not offshore might be affected by offshoring firms. We
follow Ferracci, Jolivet, and van den Berg (forthcoming) by allowing
that the outcome of an individual plant, Y, depends on its own treatment
status, D, and the proportion of offshoring firms in the same industry:
[Y.sub.p] = [Y.sub.p] ([D.sub.p], [M.sub.s(p)]). Note that we index
plants p with i and j, if they are treated and untreated. There are
three underlying assumptions. We assume that any supplier-substitution
and business-stealing effects have the greatest impact in the same
sector as the offshoring firm. This rests on the conjecture that any
losses in contracts for domestic suppliers due to offshoring stem from
changes within industry sourcing (narrow definition of international
outsourcing). Furthermore, we expect that competition occurs primarily
within the same sector. Finally, we do not allow for spillover effects
across industries.
We are concerned about a potential violation of SUTVA for two
reasons in this study. First, since both mentioned general equilibrium
effects tend to depress employment in the control group, any positive
average treatment effect could result from a positive effect of
offshoring on offshoring firms or a negative effect on the control group
or a mixture of both. Hence, it is crucial for the interpretation of the
results, whether such general equilibrium effects of offshoring are
important in Germany. Second, from an economic policy perspective, it is
interesting to know whether such spillover effects are substantial or
not, in order to identify potential winners and losers from trade
liberalization.
C. Dataset
This article builds on the IAB Establishment Panel from the
Institute for Employment Research (IAB). This panel started in 1993 and
included roughly 16,000 establishments nationwide in 2005 (see for
instance Koelling 2000). (15) The IAB panel is drawn from a stratified
sample of the establishments included in the employment statistics
register with the stratum defined over 16 industries, 10 categories of
establishment size, and 16 German states (Lander). Large establishments
are oversampled, but the sampling within each cell is random. Survey
data are collected by professional interviewers of Infratest
Sozialforschung on account of the IAB. Participation of firms is
voluntary but the response rate of more than 80% for repeatedly
interviewed establishments is high. Our sample covers the period 1998 to
2004 and is centered around the three business years 1998, 2000, and
2002, where the establishments were asked about their use of imported
intermediate goods in their production.16 17 More precisely, we exploit
information on, whether plants have predominantly, partly or not at all
received intermediate inputs, that is, all raw materials and supplies
purchased from other businesses or institutions from abroad. Our dataset
includes manufacturing and nonmanufacturing plants. Table B2 provides
some summary statistics.
D. Implementation of Matching Estimator
Conditioning on all pretreatment plant characteristics x,0 is not
practical because of the curse of dimensionality. However. Rosenbaum and
Rubin (1983) have shown that conditioning on [x.sub.i0] can be replaced
by conditioning on the propensity score, that is, the probability that
an establishment is treated. We estimate the propensity score with a
logit model and allow the likelihood of offshoring to depend on plant
characteristics and the offshoring intensity in the sector that a given
firm is operating, P([D.sub.i1] = 1) = P([x.sub.i0], [M.sub.s1])
[equivalent to] [p.sub.i]. (17)
The choice of our selection variables is motivated by the existing
empirical and theoretical literature on offshoring. These variables
should influence both the participation decision and the outcome
variable, but need to be unaffected by the treatment itself or its
anticipation. For this reason, we include only time invariant or lagged
variables:
* [Employment.sub.p,t-1] : Total employment at plant p in time t -
1 (in logarithm) serves as a proxy for the size of the plant.
* [Wage_empl.sub.p,t-1] : Wage per employee at plant p in time t -
1 (in logarithm) captures an important fixed cost of the plant. We
explicitly control also for the share of high-skilled workers, which
might be one reason for higher average wages.
* [Technology.sub.p,t-1] : Dummy variable taking the value of one
if the plant p uses state-of-the-art technology or above-average
technology in comparison to peer-group in time t - 1. The technology
variable allows investigating, whether those firms that exhibit a
superior technology within an industry tend to incur more offshoring or
not.
* [High_skilled.sub.p,t-1] : Share of high-skilled employees as
percentage of total employees at plant p in time t - 1. Intra-firm
imports from Eastern Europe to German firms depend inter alia positively
on the size of the parent firm and its R&D intensity (Marin 2006).
Yeaple (2005) shows that firms pursuing international activities tend to
pay higher wages, have more skilled workers, and employ more advanced
technologies.
* Foreign: Dummy variable taking the value of one if a foreign
owner holds majority of plant p. This variable can be expected to be
positively correlated with multinationals. Helpman, Melitz, and Yeaple
(2004) present evidence in favor of a higher productivity of
multinationals relative to nonmultinational exporters. (18)
* Offshoring [intensity.sub.s,t] : Number of offshoring plants
relative to total plants in the industry s, where plant p operates at
time t.
Finally, we control for industry-specific ([D.sub.B]),
regional-specific ([D.sub.R]), and time-specific effects ([D.sub.T]).
The error term o is assumed to be independent of the explanatory
variables and is assumed to follow a logistic distribution.
Note that we have also experimented with lagged outcome variables
as an additional determinant of offshoring in the logit model with
little success, that is, none of them turns out significantly in any
specification and the predictive power does not increase. Hence, we
prefer to continue with the more parsimonious specification.
In the population, the selection variables are balanced between the
treatment and matched-control group conditional on the true propensity
score (Rosenbaum and Rubin 1983). This property of matching ensures that
differences in outcome do not rely on differences in characteristics
between treatment and matched control group other than the treatment
itself. (19) We employ a number of standard balancing tests to exclude
systematic differences in characteristics in the sample, as discussed in
Appendix B.
There are numerous matching estimators that basically differ in
their way of measuring similarity of the propensity scores, the set of
neighbors included in the matched control group, and the weights each of
them obtain, respectively. (20) We use a difference-in-differences
kernel matching estimator, where the ATT is calculated in the following
way (Heckman, Ichimura, and Todd 1997):
(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
whereby the nonparametric function g(x) uses a kernel-matching
algorithm given by (21)
g([p.sub.i], [p.sub.j]) = K (([p.sub.j] - [p.sub.i]) /h) /
[summation over (j[member of]A(i))] K (([p.sub.j] - [p.sub.i]) / h(
with the Epanechnikov Kernel function K(x), the set of control
group observations A(i) = {j|[absolute value of [p.sub.i] - [p.sub.j]]
< h}, and the bandwidth parameter h. This estimator includes a rather
large number of control group observations in the calculation of the
ATT, but matched control group members with propensity scores more
distant to a treated observation receive a smaller weight and the
treatment and control group only includes plants with common support
([CS.sub.p]).
In this article, we are mainly interested in the employment effects
of offshoring. Accordingly, our outcome variable of main interest is the
change in log total employment at the plant-level from before (t')
to after treatment (t), whereby [DELTA][y.sub.it+s] and [DELTA][y.sub.jt
+ s] denote these changes till time t + s', s [greater than or
equal to] 0, following the offshoring event for offshoring plant i and
nonoffshoring plant j. We will consider different time horizons, with t
= 1 being equivalent to the first year, in which the offshoring activity
has been completed. For instance, if an establishment reports a higher
share of imported intermediate inputs in the year 2000 as compared to
1998, the offshoring event must have taken place during the 1999-2000
period and we measure the outcome variable as the change between at the
end of 1998 (t') and 2000 (t = 1). Since changes triggered by the
offshoring event might not materialize immediately, estimations for t +
1 and t + 2 will be reported as well. We will discuss further outcome
variables in the relevant section.
E. Identification of Heterogeneous Offshoring Effects
Our empirical strategy rests on the estimation of different
treatment effects of offshoring. In particular, we estimate the ATT
employing various treatment and control groups. Our principal definition
for the treatment status of plant p (D) is as follows:
[Offshoring.sub.pt] takes the value of one, if the plant increases its
share of imported intermediate goods (materials and services) in overall
intermediate inputs, and zero otherwise. We are able to measure this
increase qualitatively, that is, a plant increases its intermediate
goods from abroad from "not at all" to "partly" or
from "partly" to "predominantly" in the business
years 1999-2000 and 2001-2002, respectively. (22) Our plant-level
offshoring definition is similar to the definition of international
outsourcing by Feenstra and Hanson (1996, 1999).
While measuring offshoring at the plant-level offers a number of
advantages, some caveats apply to our measure of offshoring. The main
concern is that this measure might suffer from a measurement error. We
can only crudely measure any changes in offshoring. Even worse, for a
given qualitative level of imported intermediate inputs there might be
certain input fluctuations over time which are unobserved to the
econometrician. Finally, respondents to this survey might struggle to
adequately assess the use of foreign intermediate inputs. Plant-level
measures of offshoring relying on customs-based trade data like in
Biscourp and Kramarz (2007) and Hummels et al. (2014) are clearly more
precise.
To isolate the downsizing and productivity effects of offshoring on
employment, we separate the offshoring variable into cases which
coincide with restructuring, that is, plants that have had a major
restructuring by shutting down, selling-off, or spinning-off parts of
the plant (offshoring-with-restructuring) and cases which do not
coincide with restructuring (offshoring-without-restructuring). We
expect all channels (A - B + C + D) to be at work for the former group,
but all except for the downsizing channel B for the latter group (A + C
+ D). We implicitly assume that any restructuring that coincides with
offshoring is due to offshoring. (23) Offshorers with restructuring as
compared to those without restructuring tend to be bigger and employ a
less sophisticated technology relative to their industry competitors.
Note that the share of restructures among offshorers (w) is about 10%
and that these two additional treatment definitions are exhaustive and
mutually exclusive. Hence, the two ATTs for these offshoring
subgroups--weighted by their share of treated observations--gives the
overall ATT of the baseline, that is, w(A-B+C + D) + (l - w)(A + C + D).
Furthermore, the difference between these two groups allows us to
isolate the negative downsizing channel (B).
Finally, we also consider a treatment variable, where (by
construction) no domestic suppliers are expected to suffer from the
offshoring activity of other firms. We build on the treatment variable
offshoring-without-restructuring, but confine it to those offshoring
events, where there has been no simultaneous decrease in the sourcing
activity in Germany. The lAB-panel once more reports this domestic
intermediate inputs qualitatively. The last treatment definition gives
us an idea about the existence of the productivity channel (A + D), even
though the business-stealing effect might depress the control group. In
all main specifications, nontreatment is defined as those plants that do
not increase their foreign intermediate input share during the same time
period. (24)
Table 1 summarizes all theoretical ATTs that we expect for the
various combinations of treatment and control group. We will provide
some sensitivity analyses to these treatment and control group
definitions below.
IV. EMPIRICAL RESULTS
A. Main Results
We start by looking at the baseline results, which encompass all
four channels.25 Our preferred specification based on a kernel-matching
estimator (with bandwidth h of 0.01) is presented in column 1 of Table
2. (26) We find a positive and significant employment effect of
offshoring for all three time horizons considered. The point
coefficients vary between 2 and 4.5, implying that offshoring firms
enjoy an employment growth that is up to 4.5 percentage points higher
than nonoffshoring firms. Columns 2 and 3 of Table 2 show that this main
result is not sensitive to the exact matching algorithm, with the former
varying the kernel estimator's bandwidth (to 0.001) and the latter
allowing for a different matching algorithm (nearest neighbor matching).
Furthermore, we restrict the control group in columns 4 and 5 for a
given treated observation either to the same industry or same region.
The idea behind the last two specifications is that the relevant market
concept might be the respective industry or the local labor market. All
results are very similar to the baseline results.
Table 3 displays the main results of our study with column 1
presenting the baseline results from Table 2 for comparability. In
column 2, we are shutting down one of the four channels of offshoring,
namely the downsizing channel (B). We exclude all those observations
where offshoring coincides with a restructuring event, leaving us with
the channels A + C + D.27 As we would expect from theory, the positive
point coefficients increase and lie between 3.3% and 6.6%. On the other
hand, column 3 now exactly considers the treatment subgroup excluded in
column 2 as our treatment variable. Those offshoring firms that
simultaneously restructure their plant constitute about 10% of all
offshoring firms and most likely make the headlines in the daily news.
For this subgroup in principle all four channels could be still at work,
but we know that the downsizing channel plays a more important role
here. The results indeed indicate strong and negative point
coefficients, with two of the estimates being significant at least at
the 5% confidence level. The differing results between columns 2 and 3
are consistent with heterogeneous employment effects of offshoring. Note
also that these two columns appropriately weighted give us the baseline
ATT of column I. that is, w(A - B+C + D)+ (1 - w)(A+ C + D) =
[ATT.sub.base]. This is indeed the case: For t + 1, we get 0.0424, which
is very close to the estimate of 0.0419 in column 1.
Column 4 presents the clearest case in favor of the existence of
the downsizing channel B, by employing the mutually exclusive subgroup
of other offshoring firms as the control group. Hence, within offshoring
firms those that also restructure are doing much worse in terms of
employment growth with their growth rates being between 15 and 31
percentage points lower. These estimates are statistically significant
at the 1% level for all three time periods. It seems reasonable to
believe that this subgroup of offshoring firms that simultaneously close
down or sell parts of their plant are relocating production abroad.
Wagner (2011) to the best of our knowledge is the only other offshoring
study that offers a measure for relocation abroad. His results point
into the same direction, but the estimated negative effects are less
pronounced (up to 4.5% over 4 years). (28)
A final unreported result seems worth mentioning. If we keep the
same treatment group as in column 4, but employ restructuring firms that
do not offshore as a control group instead, we will get positive albeit
not significant point coefficients between 4% and 7%. This might be
cautiously interpreted in the following way: Offshoring firms tend to
save more jobs within a group of firms that have to adjust discretely.
The last column of Table 3 aims at the identification of the
productivity channel of offshoring. Thereby, we chose a treatment group,
where we know that (by construction) the substitution-supplier effect is
very unlikely to exist. (29) We exploit some additional information in
the IAB-panel on the national sourcing behavior of German plants, which
is once more available qualitatively. Our treatment variable of interest
is now offshoring-without-restructuring that does not coincide with a
reduction in domestic outsourcing. Note that this important subgroup of
the treatment group off shoring-without-restructuring in column 2 makes
up for about two thirds of these observations. Most importantly, we
expect only the channels A + D to operate in this subgroup. There are
two important takeaways from column 5. First, the estimates indeed show
positive employment effects consistent with the productivity effect,
even though we should be aware that it is not possible to isolate A from
D in this exercise. Second, the only difference in channels at work
between columns 2 and 5 is channel C. Since point coefficients are very
similar for these two types of treatment, it is rather unlikely that the
business-stealing effect (C) plays an important role.
In Table 4, we report a number of different outcome variables that
measure firm performance. (30) We start by looking at two different
measures for the overall (domestic and foreign) use of intermediate
inputs. Column 1 shows that offshoring firms increase the value of
intermediate inputs relative nonoffshoring firms, but column 2 indicates
that this is not the case, once intermediate inputs are scaled by total
sales. Two remarks are important here. First, since we know that
offshoring firms increase their share of foreign intermediate inputs (in
total intermediate inputs) and the scaled overall intermediate inputs
stay constant, this leaves the door open for a negative effect on
domestic suppliers (supplier-substitution effect). Second, we do not
want to interpret this result as causal, because the foreign and
domestic intermediate input mix is likely jointly determined.
Columns 3, 4, and 5 of Table 4 report the results for sales (in
logarithm), exports (in percent as a share of total sales), and
productivity (as measured by average labor productivity). In particular,
the first two outcome variables are statistically robust and positive
indicating that offshoring firms tend to increase their total and
foreign turnover. If offshoring firms increased the market share at the
cost of their competitors, this would be consistent with a
business-stealing effect.
While we cannot directly test for the supplier-substitution and
business-stealing effect, Figure 2 speaks to these effects'
importance (or lack thereof). We follow the empirical strategy of
Ferracci et al. (forthcoming) who conjecture that a nonconstant average
treatment effect along (in our case) the offshoring intensity is
evidence in favor of the violation of SUTVA and consequently substantial
spillover effects. It is important to bear in mind that both types of
general equilibrium effects in our study would tend to depress the
control group due to losses in supplier contracts or market share.
We are able to compute the offshoring intensity and
industry-specific average treatment effects on the treated for 14 out of
16 sectors and two periods in time. Figure 2 plots these 28 shares of
treated observations within the same sector and time against the ATTs
for the employment effects of t, t + 1, and t + 2. If spillover effects
were important in our study, we would expect a non-negligible positive
correlation. This does not seem to be the case. According to Figure 2
there is no clear pattern--if anything a slight negative
correlation--between the sector-specific ATTs and the sector's
proportion of offshoring plants. We conclude from this empirical
exercise that there might well be some firms within the control group
that suffer from other firms' offshoring activity, but this effect
is clearly not pervasive.
[FIGURE 2 OMITTED]
B. Further Results
We now provide a number of sensitivity analyses to ensure that our
results are not driven by the way the treatment or control group is
defined, by lagged employment growth, where firms lie within the
productivity distribution, or by the sourcing country-group or industry.
The first two robustness checks are mainly concerned about whether
measurement error of offshoring might impact our results. In Table Al,
we exclude those plants that increase from "partly" to
"predominantly" from our sample and confine our treatment
variable offshoring to those events, where plants increase the foreign
intermediate inputs from "none" to "partly." It is
noteworthy that most offshorers belong to this group and that it seems
reasonable to assume that plants are well aware, whether they start to
use any foreign intermediate inputs or not. Furthermore, we eliminate in
Table A2 those plants from the control group that have a positive (but
not increasing) share of foreign intermediate inputs. The main
motivation for this exercise is that firms that claim to partly use
foreign inputs might still profit from the persistence of past increases
or might have variations in their use that are unobservable to the
econometrician. Reassuringly, the results of these two tables are very
similar to our main findings in Table 3.
Table A3 reports the empirical results from another robustness
check, where the estimation of the propensity score additionally
incorporates the lagged employment growth rate as another potential
determinant of offshoring. Even though the sample is reduced markedly,
two results appear to be robust. First, in all but one case changes in
employment are not a significant explanatory variable in the logit
model.31 More importantly, all estimated treatment effects are in the
same ballpark.
As a number of theoretical models stress that only the most
productive firms can cover the fixed costs associated with offshoring,
we provide another sensitivity test along this dimension in Table A4.
Instead of simply including a productivity measure into the estimation
of the propensity score, we use a stricter test: Based on the average
labor productivity, we group all plants into a specific decile within
the productivity distribution prior to offshoring and then allow any
match between a treated plant and its control group members only within
the same productivity decile. We note in passing that the strict pecking
order suggested by theory does not hold in the data. While a larger
fraction of offshoring firms is in the higher deciles, treated firms can
be found (and their share never falls below 7%) in each decile. Our main
results once more follow a similar pattern in Table A4, but the
short-run employment effects for the baseline are smaller and not
significantly different from zero due to a (for all time horizons) more
pronounced downsizing channel.
Finally, Table A5 deals with the sourcing origin and the offshoring
sector. While the first column replicates the baseline results for
comparability, columns 2 and 3 distinguish between offshoring
originating either from European Monetary Union (EMU) or non-EMU
countries.32 Even though our measure of offshoring is due to the
aggregation of countries much cruder than Lo Turco and Maggioni (2012)
and Geishecker (2006), our results are in line with these papers.
Offshoring to EMU-countries tends to be more benign for domestic
employment than offshoring to lower-income countries. Columns 4 and 5
split the main sample into manufacturing and nonmanufacturing plants.
Since most empirical studies on offshoring are confined to manufacturing
firms, it is interesting to see that the evidence of a positive
productivity effect on employment extends beyond the boundaries of the
manufacturing sector.
V. CONCLUSIONS
From a policy perspective, the findings in this article suggest
that offshoring might have contributed to generating and safeguarding
employment during the "adjustment period" of the German
economy after the mid-1990s. In particular, we find that there is
consider able heterogeneity across firms. While there is large and
significant negative relative employment growth due to downsizing in a
small group of firms, productivity increases dominate and the overall
effect on employment is positive.
Moreover, the indirect effects from offshoring on nonoffshoring
firms (business stealing and supplier substitution) appear to be small
and insignificant. Thus offshoring firms created more employment on
average than nonoffshoring firms and there was no significant negative
indirect effect on employment in local suppliers or domestic
competitors. The later finding is relevant not only from a policy
perspective, but also methodologically since spillovers from the
treatment to the control group constitute a potential violation of the
SUTVA. This suggests that it is important to identify all direct and
indirect channels through which offshoring may affect employment.
ABBREVIATIONS
CIA: Conditional Independence Assumption
EMU: European Monetary Union
FDI: Foreign Direct Investment
MNE: Multinational Enterprise
SUTVA: Stable Unit Treatment Value Assumption
doi: 10.1111/ecin.12124
Online Early publication July 24, 2014
APPENDIX A
TABLE A1
Identification of the Different Channels of Offshoring on
Employment (Kernel Matching)-Robustness Treatment Definition
Gradual Discrete
Baseline Downsizing Downsizing
Time (1) (2) (3)
t 0.0274 ** 0.0382 *** -0.0717
(0.0097) (0.0109) (0.0492)
t + 1 0.0452 *** 0.0708 *** -0.1872 ***
(0.0124) (0.0136) (0.0661)
t + 2 0.0475 *** 0.0682 *** -0.1715 **
(0.0153) (0.0143) (0.0772)
Relocation Productivity
Abroad Effect
Time (4) (5)
t -0.1207 ** 0.0419 ***
(0.0577) (0.0115)
t + 1 -0.2778 *** 0.0780 ***
(0.0743) (0.0144)
t + 2 -0.2551 *** 0.0785 **
(0.0801) (0.0169)
w(A - B +
C + D)
Theoretical + (1 - w)
ATT (A + C + D) A + C + D A - B + C + D
Treatment group Offshoring Offshoring- Offshoring-
without- with-
restructuring restructuring
No. of 1,038 940 98
treatments
Control group Nonoffshoring Nonoffshoring Nonoffshoring
No. of controls 7,201 7.201 7,193
Total 8,239 8,141 7,291
Theoretical
ATT -B A + D
Treatment group Offshoring- Offshoring-
with- without-
restructuring restructuring
No. of 98 690
treatments
Control group Offshoring- Nonoffshoring
without-
restructuring
No. of controls 938 7,201
Total 1,036 7,288
Notes: Standard errors in parentheses. For the matched
sample standard errors are generated via bootstrapping (200
replications). Number of observations refer to time t. The
treatment-variable Offshoring in this table is defined as a
(qualitative) increase of a plant's share of foreign
intermediate inputs in total inputs from "none" to "partly"
either in the years 1999-2000 or 2001-2002 for a certain
plant. Note that those plants that increase from "partly" to
"predominantly" are excluded from the sample.
Offshoring-with-restructuring (Offshoring-without-restructuring)
comprises those plants that offshoring and (do not)
restructure their plant at the same time, that is, (no)
parts of the plant are closed down, sold-off, or spun-off.
Column 5 employs Offshoring-without-restructuring-and-
without-supplier-substitution as treatment covering all
those Offshoring-without-restructuring events, where there
was no reduction in sourcing to domestic suppliers at the
same time. The control group nonoffshoring is defined as
those plants that do not offshore during the same time
period.
*** Significant at 1%; ** significant at 5%; * significant
at 10%.
TABLE A2
Identification of the Different Channels of Offshoring on
Employment (Kernel Matching)-Robustness Control Group
Definition
Gradual Discrete
Baseline Downsizing Downsizing
Time (1) (2) (3)
t 0.0245 ** 0.0388 *** -0.1142 **
(0.0111) (0.0117) (0.0490)
t + 1 0.0535 *** 0.0788 *** -0.2116 ***
(0.0156) (0.0148) (0.0633)
1 + 2 0.0579 *** 0.0772 *** -0.1715 **
(0.0162) (0.0177) (0.0772)
Relocation Productivity
Abroad Effect
Time (4) (5)
t -0.1415 *** 0.0410 ***
(0.0532) (0.0125)
t + 1 -0.2882 *** 0.0827 ***
(0.0655) (0.0157)
1 + 2 -0.2812 *** 0.0807 ***
(0.0758) (0.0196)
w (A - B +
C + D)
Theoretical + (1 - w)
ATT (A + C + D) A + C + D A - B + C + D
Treatment Offshoring Offshoring- Offshoring-
group without- with-
restructuring restructuring
No. of 1,265 1,143 122
treatments
Control Nonoffshoring Nonoffshoring Nonoffshoring
group
No. of 4,859 4.859 4,853
controls
Total 6,124 6,002 4,975
Notes: Standard errors in parentheses. For the matched
sample standard errors are generated via bootstrapping (200
replications). Number of observations refer to time t. The
treatment-variable Offshoring is defined as a (qualitative)
increase of a plant's share of foreign intermediate inputs
in total inputs either in the years 1999-2000 or 2001-2002
for a certain plant. Offshoring-with-restructuring
(Offshoring-without-restructuring) comprises those plants
that offshoring and (do not) restructure their plant at the
same time, that is, (no) parts of the plant are closed down,
sold-off, or spun-off. Column 5 employs Offshoring-without-
restructuring-and-without-supplier-substitution as treatment
covering all those offshoring-without-restructuring events,
where there was no reduction in sourcing to domestic
suppliers at the same time. The control group nonoffshoring
is defined as those plants that do not offshore during the
same time period. Note that in this table all those plants
with a positive, but constant level of foreign intermediate
inputs (either partly or predominantly) are excluded from
the control group.
*** Significant at 1%; ** significant at 5%;
* significant at 10%.
TABLE A3
Identification of the Different Channels of Offshoring on
Employment (Kernel Matching)--Robustness Selection
on Lagged Outcome
Gradual Discrete
Baseline Downsizing Downsizing
Time (1) (2) (3)
t 0.0176 0.0305 *** -0.0909
(0.0118) (0.0113) (0.0561)
t + 1 0.0342 ** 0.0593 *** -0.1749 ***
(0.0143) (0.0138) (0.0660)
t + 2 0.0374 ** 0.0584 *** -0.1737 **
(0.0176) (0.0157) (0.0718)
Relocation Productivity
Abroad Effect
Time (4) (5)
t -0.1509 ** 0.0319 **
(0.0682) (0.0135)
t + 1 -0.2898 *** 0.0651 ***
(0.0757) (0.0182)
t + 2 -0.2877 *** 0.0601 ***
(0.1057) (0.0204)
w (A - B +
C + D)
Theoretical + (l - w) A - B +
ATT (A + C + D) A + C + D C + D
Treatment Offshoring Offshoring- Offshoring-
group without- with-
restructuring restructuring
No. of 983 884 99
treatments
Control Nonoffshoring Nonoffshoring Nonoffshoring
group
No. of 5,601 5.601 5,470
controls
Total 6,584 6,585 5,569
Theoretical
ATT -B A + D
Treatment Offshoring- Foreign-
group with- supplier
restructuring switching
No. of 99 587
treatments
Control Offshoring- Nonoffshoring
group without-
restructuring
No. of 876 5,601
controls
Total 975 6.188
Notes: Standard errors in parentheses. For the matched
sample standard errors are generated via bootstrapping (200
replications). Number of observations refer to time t. The
treatment-variable Offshoring is defined as a (qualitative)
increase of a plant's share of foreign intermediate inputs
in total inputs either in the years 1999-2000 or 2001-2002
for a certain plant. Offshoring-with-restructuring
(Offshoring-without-restructuring) comprises those plants
that offshoring and (do not) restructure their plant at the
same time, that is, (no) parts of the plant are closed down,
sold-off, or spun-off. Column 5 employs Offshoring-without-
restructuring-and-without-supplier-substitution as treatment
covering all those offshoring-without-restructuring events,
where there was no reduction in sourcing to domestic
suppliers at the same time. The control group nonoffshoring
is defined as those plants that do not offshore during the
same time period. Note that in this table the underlying
estimation of the propensity score additionally incorporates
the lagged employment growth rate as another potential
determinant of offshoring.
*** Significant at 1%; ** significant at 5%; * significant
at 10%.
TABLE A4
Identification of the Different Channels of Offshoring on
Employment (Kernel Matching)--Matching Within Productivity Deciles
Gradual Discrete
Baseline Downsizing Downsizing
Time (1) (2) (3)
t 0.0069 0.0252 ** -0.2039 **
(0.0129) (0.0106) (0.0918)
t + 1 0.0279 * 0.0581 *** -0.3366 ***
(0.0129) (0.0140) (0.1055)
t + 2 0.0355 ** 0.0626 *** -0.4036 ***
(0.0174) (0.0156) (0.1369)
Relocation Productivity
Abroad Effect
Time (4) (5)
t -0.1317 ** 0.0232 **
(0.0617) (0.0118)
t + 1 -0.3576 *** 0.0587 ***
(0.0909) (0.0163)
t + 2 -0.4642 *** 0.0668 **
(0.1506) (0.0187)
w (A - B +
Theoretical C + D) +
ATT (1 - w) A + C + D A - B + C + D
(A + C + D)
Treatment Offshoring Offshoring- Offshoring-
group without- with-
restructuring restructuring
No. of 1,195 1,078 117
treatments
Control Nonoffshoring Nonoffshoring Nonoffshoring
group
No. of 6,795 5,859 6,586
controls
Total 7,990 6.883 6,703
Theoretical
ATT -B A + D
Treatment Offshoring- Offshoring-
group with- without-
restructuring restructuring
No. of 117 958
treatments
Control Offshoring- Nonoffshoring
group without-
restructuring
No. of 988 6,550
controls
Total 1,105 7,508
Notes: Standard errors in parentheses. For the matched
sample standard errors are generated via bootstrapping (200
replications). Number of observations refer to time t. The
treatment-variable offshoring is defined as a (qualitative)
increase of a plant's share of foreign intermediate inputs
in total inputs either in the years 1999-2000 or 2001-2002
for a certain plant. Offshoring-with-restructuring
(Offshoring-without-restructuring) comprises those plants
that offshoring and (do not) restructure their plant at the
same time, that is, (no) parts of the plant are closed down,
sold-off, or spun-off. Column 5 employs Offshoring-without-
restructuring-and-without-supplier-substitution as treatment
covering all those offshoring-without-restructuring events,
where there was no reduction in sourcing to domestic
suppliers at the same time. The control group nonoffshoring
is defined as those plants that do not offshore during the
same time period. Note that we restrict in this table the
potential members of the control group to those who share
the same productivity decile with a given treatment
observation.
*** Significant at 1%; ** significant at 5%; * significant
at 10%.
TABLE A5
Overall Effect of Offshoring on Employment-Further Results
EMU- Non-EMU
Baseline Offshoring Offshoring
Time (1) (2) (3)
t 0.0220 ** 0.0144 0.0117
(0.0103) (0.0124) (0.0193)
t + 1 0.0419 *** 0.0492 *** 0.0061
(0.0129) (0.0153) (0.0236)
r + 2 0.0447 *** 0.0653 *** -0.0022
(0.0151) (0.0182) (0.0296)
Manufacturing Nonmanufacturing
Time (4) (5)
t 0.0232 * 0.0174
(0.0139) (0.0156)
t + 1 0.0377 ** 0.0398 *
(0.0166) (0.0206)
r + 2 0.0333 * 0.0509 **
(0.0194) (0.0205)
Notes: Standard errors in parentheses are generated via
bootstrapping (500 replications). Column 1 shows the
baseline results from Table 2 for comparability. Columns 2
and 3 report results, when offshoring is confined to EMU or
non-EMU countries, respectively. Columns 4 and 5 show the
results for manufacturing and nonmanufacturing plants. The
treatment-variable Offshoring is defined as an increase in
the share of imported intermediate inputs in overall
intermediate inputs either in the years 1999-2000 or 2001-2002
for a certain plant. Nontreatment is defined as those
plants that do not increase their vertical integration
during the same time period.
*** Significant at 1%; ** significant at 5%;
* significant at 10%.
APPENDIX B
We employ a number of different balancing tests to exclude
systematic differences in characteristics in the sample. First, we
calculate the standardized difference between treatment and
matched-control group of all selection variables at a time (see, e.g.,
Rosenbaum and Rubin 1985; Smith and Todd 2005b). Rosenbaum and Rubin
(1985) consider the standardized difference large if it exceeds 20%.
Second, we perform a mean-difference t test with standard deviations
differing in treatment and matched-control group. Third, we follow Smith
and Todd (2005b) who propose a regression-based test. For each selection
variable [x.sub.it] that is used in the propensity score, the following
regression is estimated
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
for the years f = 1998 and 2000. Smith and Todd (2005b) argue that
a joint significance test over the [gamma]-coefficients
would indicate that the balancing condition is not satisfied.
Hence, we expect an insignificant Wald-test. Fourth, we perform the
Hotelling test on quintiles that tests balancing within each quintile
over all variables jointly. Tables B4 and B5 display the results for the
Hotelling test as well as the distribution over the live quintiles
considered, showing once more no significant imbalance. Finally, we
follow Imbens (2004) and Smith and Todd (2005a), who suggests a way to
indirectly test for the CIA using a test of Heckman and Hotz (1989).
We estimate the average treatment effect on the treated for an
outcome variable before treatment takes place. If this effect is zero,
it renders the CIA more plausible. Contrary to that, if it is not zero,
this test indicates that there are systematic differences in outcomes
between treatment and matched control group even before treatment,
suggesting that the ATT is not caused by treatment alone. All these
balancing tests are reported in Tables B3-B6 and indicate a good
balancing between the treatment and control group.
TABLE B1
Logit Estimates of Propensity Score
Gradual Discrete
Baseline Downsizing Downsizing
(1) (2) (3)
Employment 0.129 *** 0.087 *** 0.457 ***
(t - 1) (0.228) (0.024) (0.064)
Wage/employee 0.228 *** 0.205 *** 0.702 ***
(t - 1) (0.073) (0.076) (0.270)
Technology 0.220 *** 0.296 *** -0.516 **
(t-1) (0.071) (0.075) (0.202)
High-skilled 0.361 *** 0.268 ** 1.067 **
(t-1) (0.130) (0.135) (0.423)
Foreign Q 4 J 1 *** 0.045 *** 0.161
ownership (0.120) (0.126) (0.297)
Offshoring 0.071 ** 0.057 * 0.175 **
intensity (0.028) (0.030) (0.078)
in industry
Industry Yes Yes Yes
dummies
Regional Yes Yes Yes
dummies
Times dummies Yes Yes Yes
Treatment Offshoring Offshoring- Offshoring-
group without- with-
restructuring restructuring
Control Nonoffshoring Nonoffshoring Nonoffshoring
group
Pseudo 0.06 0.06 0.15
[R.sup.2]
Observations 8,466 8,344 7,323
Relocation Productivity
Abroad Effect
(4) (5)
Employment 0.501 *** 0.105 ***
(t - 1) (0.079) (0.028)
Wage/employee 0.076 0.208 **
(t - 1) (0.307) (0.090)
Technology -0.765 *** 0.260 ***
(t-1) (0.225) (0.089)
High-skilled 1.033 0.196
(t-1) (0.478) (0.162)
Foreign -0.281 0.220
ownership (0.321) (0.157)
Offshoring 0.158 * 0.055
intensity (0.091) (0.036)
in industry
Industry Yes Yes
dummies
Regional Yes Yes
dummies
Times dummies Yes Yes
Treatment Offshoring- Offshoring-
group with- without-
restructuring restructuring
without
supplier
substitution
Control Offshoring- Nonoffshoring
group without-
restructuring
Pseudo 0.16 0.05
[R.sup.2]
Observations 1,265 7,951
Notes: Standard errors in parenthesis. Definition of
variables included in the matching: Employment: log of
number of total employees per plant, Wage per employee: log
of average wage per employee, Technology: Dummy = 1 if plant
has above average or state-of-the-art technology,
High-skilled: share of high-skilled workers of total
employment, Foreign ownership: Dummy = 1 if a foreign owner
holds the majority of the plant; Offshoring intensity in
industry: Number of offshoring plants relative to total
plants in a given industry at time t. Industry and regional
dummies are employed but not reported. The
treatment-variable Offshoring is defined as an increase in
the share of imported intermediate inputs in overall
intermediate inputs either in the years 1999-2000 or
2001-2002 for a certain plant. Nontreatment is defined as
those plants that do not increase their vertical integration
during the same time period. Offshoring-with-restructuring
imposes the following additional restriction on the
offshoring definition above: the plant is restructured
during the offshoring event, that is, parts of the plant are
closed down, sold-off, or spun-off.
*** Significant at 1%; ** significant at 5%; * significant
at 10%.
TABLE B2
Summary Statistics
Offshoring Plants Nonoffshoring Plants
Mean Std. Dev. Mean Std. Dev.
Employment 3.7507 1.7280 3.1600 1.6884
Wage per employee 7.3817 0.5661 7.2091 0.5870
Technology 0.7296 0.4443 0.6760 0.4680
High-skilled 0.3837 0.2792 0.3769 0.2965
Foreign ownership 0.0996 0.2996 0.0415 0.1995
Sales 15.3133 2.1337 14.4096 2.0797
Exports 11.2265 21.2526 5.3577 15.4371
Intermediate inputs 54.8162 22.1118 49.4416 23.7047
Share of offshoring
firms 17.2888 5.0728 14.5299 5.9645
Number of
observations 1265 7201
Notes: Employment: log of number of employees per plant,
Wage per employee: log of average wage per employee,
Technology: Dummy = 1 if plant has above average or state-
of-the art technology, High-skilled: share of high-skilled
workers of total employment, Foreign ownership: Dummy = 1 if
a foreign owner holds the majority of the plant, Sales: log
of total turnover of the plant, Exports: ratio of turnover
aboard to total turnover at the plant, Intermediate inputs:
ratio of intermediate inputs to output, Share of offshoring
firms: proportion of offshoring firms (in percent) in
industry j at time t.
TABLE B3
Balancing Tests from Kernel Matching
Mean Mean
Treatment Control Percent
Covariate Group Group Bias
Employment 3.7338 3.7180 0.9
Wage per employee 7.3771 7.3789 -0.3
Technology 0.7389 0.7338 1.1
High-skilled 0.3802 0.3822 -0.7
Foreign ownership 0.0969 0.0880 3.5
Offshoring
firms (in %) 17.395 17.414 -0.3
Regression-
Mean- Based Tests
Difference Wald
t-Statistic Statistic
Covariate (p-Value) (p-Value)
Employment 0.21 (0.83) 0.73 (0.57)
Wage per employee -0.08 (0.94) 1.11 (0.35)
Technology 0.27 (0.79) 0.93 (0.45)
High-skilled -0.17(0.87) 1.21 (0.30)
Foreign ownership 0.71 (0.48) 1.51 (0.20)
Offshoring
firms (in %) -0.09 (0.93) 7.25 (0.00)
Notes: Definition of variables included in the matching:
Employment: log of number of employees per plant, Wage per
employee: log of average wage per employee, Technology:
Dummy = 1 if plant has above average or state-of-the art
technology, High-skilled: share of high-skilled workers of
total employment, Foreign ownership: Dummy = 1 if a foreign
owner holds the majority of the plant. Offshoring firms (in
%): proportion of offshoring firms (in percent) in industry
j at time t. Balancing of industry, regional and time
dummies is not reported. All dummies have a percent bias
below 3. Mean-difference is mean difference test with
standard deviations differing between treatment and control
group. Regression based Wald test statistic follows Smith
and Todd (2005b).
TABLE B4
Hotelling's T-Squared Tests by Propensity
Score Quintile
T-Squared E-Test
Quintile Statistics Statistics p-Value
First 30.330 0.899 0.633
Second 31.345 0.901 0.632
Third 44.675 1.275 0.132
Fourth 32.091 0.895 0.645
Fifth 35.715 0.997 0.475
TABLE B5
Frequency Distribution of Treated and Nontreated
Plants by Propensity Score Quintile
Quintile Offshoring Nonoffshoring
Plants Plants
First 75 1,381
Second 124 1,331
Third 208 1.248
Fourth 288 1.167
Fifth 389 1.066
TABLE B6
Heckman and Hotz (1989): Evidence for Self-Selection into
Offshoring? Growth Differences in Employment, Sales,
Exports, and Productivity (t = -1 to t = 0)
Employment Sales Exports Productivity
Time (1) (2) (3) (4)
Kernel -0.0084 0.0086 0.5565 0.0190
matching (0.0128) (0.0176) (0.5674) (0.0173)
Notes: Standard errors in parentheses. For the matched
sample standard errors are generated via bootstrapping (500
replications). The treatment-variable Offshoring is defined
as an increase in the share of imported intermediate inputs
in overall intermediate inputs between in the years
1999-2000 or 2001-2002 for a certain plant. Nontreatment is
defined as those plants that do not increase their vertical
integration during the same time period.
*** Significant at 1%; ** significant at 5%; * significant
at 10%.
REFERENCES
Abadie, A., and G. Imbens. "Large Sample Properties of
Matching Estimators for Average Treatment Effects." Econometrica,
74, 2006, 235-67.
--. "On the Failure of the Bootstrap for Matching
Estimators." Econometrica, 76, 2008, 1537-57.
Abadie, A., D. Drukker, J. Herr, and G. Imbens. "Implementing
Matching Estimators for Average Treatment Effects in Stata." The
Stata Journal, 4, 2004, 290-311.
Balsvik, R., and S. Birkeland. "Offshoring and Wages: Evidence
from Norway." Norwegian School of Economics and Business
Administration, September 2012, mimeo. 2012.
Barba Navaretti, G., D. Castellani, and A.-C. Disdier.
"Investments Abroad and Performance at Home: Evidence from Italian
Multinationals." Oxford Economic Papers, 62, 2009, 234-60.
Baumgarten, D., I. Geishecker, and H. Gorg. "Offshoring,
Tasks, and the Skill-Wage Pattern." European Economic Review, 61,
2013, 132-52.
Becker, S., and M.-A. Muendler. "The Effect of FDI on Job
Security." B.E. Journal of Economic Analysis and Policy, 8, 2008,
1-44.
--. "Margins of Multinational Labor Substitution."
American Economic Review, 100, 2010, 1999-2030.
Becker, S., K. Ekholm, and M.-A. Muendler. "Offshoring and the
Onshore Composition of Tasks and Skills." Journal of International
Economics, 90, 2013, 90-106.
Biscourp, P., and F. Kramarz. "Employment, Skill Structure and
International Trade: Firm-Level Evidence for France." Journal of
International Economics, 72, 2007, 22-51.
Buch, C., and A. Lipponer. "Volatile Multinationals? Evidence
from the Labor Demand of German Firms." Labour Economics, 17, 2010,
345-53.
Caliendo, M., and S. Kopeinig. "Some Practical Guidance for
the Implementation of Propensity Score Matching." Journal of
Economic Surveys, 22, 2008, 31-72.
Crino, R. "Offshoring, Multinationals and Labour Market: A
Review of the Empirical Literature." Journal of Economic Surveys,
23, 2009, 197-249.
Crino, R. "The Effects of Offshoring on Post-displacement
Wages: Evidence from the United States," The World Economy, 33,
2010, 1836-69.
Debaere, P., H. Lee, and J. Lee. "It Matters Where You
Go--Outward Foreign Direct Investment on Multinationals' Employment
Growth at Home." Journal of Development Economics, 91, 2006,
301-09.
Defevre, F., and F. Toubal. "Productivity and the Sourcing
Modes of Multinational Firms: Evidence from French Firm-level
Data." Journal of Economic Behavior and Organization, 94, 2013,
345-57.
Desai, M., F. Foley, and J. Hines. "Domestic Effects of the
Foreign Activities of US Multinationals." American Economic
Journal: Economic Policy, 1, 2009, 181-203.
Ebenstein, A.. A. Harrison, M. McMillan, and S. Phillips.
Forthcoming. "Estimating the Impact of Trade and Offshoring on
American Workers Using the Current Population Surveys." Review of
Economics and Statistics.
Egger, H., and P. Egger. "Outsourcing and Skill-specific
Employment in a Small Economy: Austria after the Fall of the Iron
Curtain." Oxford Economic Papers, 28,2003, 525-643.
--. "The Determinants of EU Processing Trade." The World
Economy, 28, 2005, 147-68.
Egger, H., U. Kreickemeier, and J. Wrona. "Offshoring Domestic
Jobs." University of Bayreuth, January 2013, mimeo, 2013.
Egger, P., M. Pfaffermayr, and A. Weber. "Sectoral Adjustment
of Employment to Shifts in Outsourcing and Trade: Evidence from a
Dynamic Fixed Effects Multinomial Logit Model." Journal of Applied
Econometrics, 22, 2007, 559-80.
Feenstra, R. "Offshoring in the Global Economy, Microeconomic
Structure and Macroeconomic Implications." Ohlin Lectures Series.
Cambridge, MA: MIT Press, 2010.
Feenstra, R. C., and G. H. Hanson. "Globalization,
Outsourcing, and Wage Inequality." American Economic Review, 86,
1996, 240-45.
--. "The Impact of Outsourcing and High-Technology Capital on
Wages: Estimates for the U.S., 1972-1990." Quarterly Journal of
Economics, 114, 1999,907-40.
Ferracci, M., G. Jolivet, and G. J. van den Berg. Forthcoming.
"Evidence of Treatment Spillovers Within Markets." Review of
Economics and Statistics.
Geishecker, I. "Does Outsourcing to Central and Eastern Europe
Really Threaten Manual Workers' Jobs in Germany?" The World
Economy, 29, 2006, 559-83.
--. "The Impact of International Outsourcing on Individual
Employment Security: A Micro-level Analysis." Labour Economics, 15,
2008, 291-314.
Geishecker, I., and H. Gorg. "Winners and Losers: A Microlevel
Analysis of International Outsourcing and Wages." Canadian Journal
of Economics, 41, 2008, 243-70.
Geishecker, I., M. Riedl, and P. Frijters. "Offshoring and Job
Loss Fears: An Econometric Analysis of Individual Perceptions."
Labour Economics, 19, 2012, 738-47.
Gorg, H. "Globalization, Offshoring and Jobs," in Making
Globalization Socially Sustainable, edited by M. Bacchetta and M.
Jansen. Geneva, Switzerland: International Labour Office and World Trade
Organization, 2011.21-48.
Gorg, H.. A. Hanley, and E. Strobl. "Productivity Effects of
International Outsourcing: Evidence from Plant-level Data."
Canadian Journal of Economics, 41,2008, 670-88.
Grossman, G. M., and E. Rossi-Hansberg. "Trading Tasks: A
Simple Theory of Offshoring." American Economic Review, 98(5),
2008, 1978-97.
Harrison, A., and M. McMillan. "Offshoring Jobs?
Multinationals and US Manufacturing Employment." Review of
Economics and Statistics, 93, 2011, 857-75.
Head, K., and J. Ries. "Offshore Production and Skill
Upgrading by Japenese Manufacturing Firms." Journal of
International Economics, 58, 2002, 81-105.
Heckman, J., and J. Hotz. "Choosing Among Alternative
Nonexperimental Methods for Estimating the Impact of Social Programs:
The Case of Manpower Training." Journal of the American Statistical
Association, 84, 1989, 862-74.
Heckman. J., H. Ichimura, and P. Todd. "Matching as an
Econometric Evaluation Estimator: Evidence from Evaluating a Job
Training Programme." Review of Economic Studies, 64, 1997, 605-54.
--. "Matching as an Econometric Evaluation Estimator."
Review of Economic Studies, 65, 1998, 261-94.
Helpman, E. "Trade, FDI, and the Organization of Firms."
Journal of Economic Literature, 44, 2006, 589-630.
Helpman, E., M. Melitz, and S. Yeaple. "Export Versus FDI with
Heterogeneous Firms." American Economic Review, 94, 2004, 300-316.
Hijzen, A. "International Outsourcing, Technological Change,
and Wage Inequality." Review of International Economics, 15, 2007,
188-205.
Hijzen, A., H. Gorg, and R. C. Hine. "International
Outsourcing and the Skill Structure of Labour Demand in the United
Kingdom." The Economic Journal, 115, 2005, 861-79.
Hijzen, A., T. Inui, and Y. Todo. "Does Offshoring Pay?
Firm-level Evidence from Japan." Economic Inquiry, 48, 2010,
880-95.
Hijzen, A., S. Jean, and T. Mayer. "The Effects at Home of
Initiating Production Abroad: Evidence from Matched French Firms."
Review of World Economics, 147, 2011, 457-83.
Hsieh, C.-T., and K. T. Woo. "The Impact of Offshoring to
China on Hong Kong's Labor Market." American Economic Review,
95, 2005, 1673-87.
Hummels, D., R. Joergensen, J. Munch, and C. Xiang. "The Wage
Effects of Offshoring: Evidence from Danish Matched Worker-Firm
Data." American Economic Review, 104, 2014, 1597-629.
Imbens, G. "Nonparametric Estimation of Average Treatment
Effects under Exogeneity: A Review." Review of Economics and
Statistics, 86, 2004, 4-29.
Imbens, G., and J. Wooldridge. "Recent Developments in the
Econometrics of Program Evaluation." Journal of Economic
Literature, 47, 2009, 5-86.
Jabbour, L. "Offshoring and Firm Performance: Evidence from
French Manufacturing Industry." The World Economy, 33,2010,507-24.
Keuschnigg, C., and E. Ribi. "Outsourcing, Unemployment and
Welfare Policy." Journal of International Economics, 78, 2009,
168-76.
Klein, M. W., C. Moser, and D. Urban. "Exporting, Skills and
Wage Inequality." Labor Economics, 25, 2013, 76-85.
Koelling, A. "The IAB-Establishment Panel." Schmollers
Jahrbuch/Journal of Applied Social Science Studies, 120, 2000, 291-300.
Kohler, W. "International Outsourcing and Factor Prices with
Multistage Production." The Economic Journal, 114, 2004, C166-85.
Kohler, W., and J. Wrona. "Offshoring Tasks, Yet Creating
Jobs?" Working Paper No. 12, University of Tuebingen, 2011.
Kramarz, F. "Offshoring, Wages, and Employment: Evidence from
Data Matching Imports, Firms and Workers." CREST-INSEE, July 2008.
mimeo, 2008.
Leuven, E., and B. Sianesi. "PSMATCH2: Stata Module to Perform
Full Mahalanobis and Propensity Score Matching, Common Support Graphing,
and Covariate Imbalance Testing." 2003. http://ideas.repec.org/
c/boc/bocode/s2432001 .html.
Lo Turco, A., and D. Maggioni. "Offshoring to High and Low
Income Countries and the Labor Demand. Evidence from Italian
Firms." Review of International Economics, 20, 2012, 636-53.
Marin, D. "A New International Division of Labor in Europe:
Offshoring and Offshoring to Eastern Europe." Journal of the
European Economic Association, 6, 2006, 612-22.
Mitra, D., and P. Rajan. "Offshoring and Unemployment: The
Role of Search Frictions Labor Mobility." Journal of International
Economics, 81, 2010, 219-29.
Munch, J., and J. Skaksen. "Specialization, Outsourcing and
Wages." Review of World Economics, 145, 2009, 57-73.
Olsen, K. "Productivity Impacts of Offshoring and Outsourcing:
A Review." OECD Working Paper No. 2006/01, 2006.
Pfluger, M., U. Blien, J. Moller, and M. Moritz. "Labor Market
Effects of Trade and FDI--Recent Advances and Research Gaps."
JahrbUcherfiir Nationalokonomie und Statistik, 233, 2013. 86-116.
Rosenbaum, P. R., and D. B. Rubin. "The Central Role of the
Propensity Score in Observational Studies for Causal Effects."
Biometrika, 70, 1983, 41 -50.
--. "Constructing a Control Group Using Multivariate Matched
Sampling Methods that Incorporate the Propensity Score." The
American Statistician, 39, 1985, 33-38.
Sethupathy, G. "Offshoring, Wages, and Employment: Theory and
Evidence." European Economic Review, 62, 2013,73-97.
Slaughter, M. "International Trade and Labor--Demand
Elasticities." Journal of International Economics, 54, 2001,27-56.
Smith, J., and P. Todd. "Does Matching Overcome LaLonde's
Critique of Nonexperimental Estimators?" Journal of Econometrics,
125, 2005a, 305-53.
--. "Rejoinder." Journal of Econometrics, 125, 2005b,
365-75.
Wagner, J. "Offshoring and Firm Performance: Self-Selection,
Effects on Performance or Both?" Review of World Economics, 147,
2011, 217-47.
Yeaple, S. "A Simple Model of Firm Heterogeneity,
International Trade, and Wages." Journal of International
Economics, 65, 2005, 1-20.
(1.) Geishecker, Riedl, and Frijters (2012) show that offshoring is
indeed associated with job loss fears for a representative longitudinal
survey of private households in Germany.
(2.) Other recent theoretical contributions on offshoring include
Kohler (2004), Grossman and Rossi-Hansberg (2008), Keuschnigg and Ribi
(2009), Mitra and Rajan (2010), and Egger, Kreickemeier, and Wrona
(2013).
(3.) This term was coined by Sethupathy (2013).
(4.) We will use firm, plant, and establishment interchangeably in
this article, except for if we refer to a specific study. Our empirical
results will mainly rely on German plant-level data.
(5.) Thus, we use the same definition as Helpman (2006).
(6.) Unfortunately, this effect cannot be traced in existing data,
because domestic suppliers of offshoring firms cannot be tracked in the
data.
(7.) Feenstra and Hanson (1996, 1999) distinguish between two forms
of international outsourcing. While the broad measure considers any
imported intermediate inputs, the narrow measure confines to imported
intermediate inputs from the same two-digit industry.
(8.) The literature on the employment and wage effects of
offshoring based on industry-level offshoring measures includes Feenstra
and Hanson (1996,1999), Slaughter (2001), Hijzen, Gorg, and Hine (2005),
Hsieh and Woo (2005), Egger and Egger (2003, 2005), Egger, Pfaffermayr,
and Weber (2007), Hijzen (2007), Geishecker and Gorg (2008), Munch and
Skaksen (2009), Crino (2010), Senses (2010), Ebenstein et al.
(forthcoming), and Baumgarten, Geishecker, and Gorg (2013).
(9.) Due to very detailed trade data, they are able to compute not
only a continuous firm-level measure (broad and narrow definition), but
also time-varying instruments at the firm level. These instruments
exploit the firm's trade structure and the world supply (minus
Denmark) of a sourcing country's specific product in time.
(10.) Klein, Moser, and Urban (2013) find in contrast to this study
that exports are not a rising tide that lifts all boats, but that export
activity in Germany is associated with within and between skill group
wage inequality.
(11.) Kramarz (2008) shows that French firms facing strong unions
substitute owns production for imports after the introduction of the
Single Market Program of the European Union.
(12.) There are a number of studies on labor market outcomes or
firm performance with firm-level measures of (mostly the extensive
margin of) FDI, including Head and Ries (2002), Barba Navaretti,
Castellani, and Disdier (2009) Desai, Foley, and Hines (2009), Buch and
Lipponer (2010), Debaere, Lee, and Lee (2006), Harrison and McMillan
(2011), Hijzen, Jean, and Mayer (2011).
(13.) For this reason, several empirical studies documenting an
impact of offshoring on average labor productivity or total factor
productivity may or may not imply such positive employment effects (see
for instance Gorg, Hanley, and Strobl 2008; Hijzen, Inui, and Todo 2010;
Jabbour 2010).
(14.) Our data does not allow us to differentiate between buying
intermediate goods through arms-length trade or from an own plant
abroad. Hence, we can capture offshoring as defined for instance by
Helpman (2006). At the same time, it should not matter for the
employment effects, who owns the production factors abroad, but where
the international sourcing comes from.
(15.) The IAB-Establishment Panel data is confidential but not
exclusive. They are available for noncommercial research by visiting the
Research Data Centre (FDZ) of the Federal Employment Agency at the
Institute of Employment Research in Nuremberg, Germany. For further
information, we refer to http://fdz.iab.de/en.
(16.) For simplification we refer here to the business years the
data is covering. These questions were asked in the survey years 1999,
2001, and 2003.
(17.) Heckman, Ichimura, and Todd (1998) compare the efficiency of
conditioning kernel matching estimators on p([x.sub.10]) rather than on
xi0 and do not find any one of them dominating, but conjecture that the
small sample efficiency of conditioning on the propensity score is
superior.
(18.) We have also experimented with the level of exports, but this
variable did not enter significantly due to the high correlation with
our size measure.
(19.) In the sample, a lack of balancing may be due to a
misspecification of the estimated propensity score or due to a mismatch
of propensity scores of treatment and matched control observations or
due to an unfortunate draw of the sample (Rosenbaum and Rubin 1985).
(20.) For a survey on alternative matching algorithms, see Caliendo
and Kopeinig (2008).
(21.) See, e.g., Heckman, Ichimura, and Todd (1997).
(22.) We pool the two time periods for which we are able to define
offshoring in order to profit from efficiency gains. Pooling tests
confirm that this empirical strategy is valid. Furthermore, an increase
from "partly" to "predominantly" as compared to an
increase from "not at all" to "partly" does not
yield significantly different effects on the outcome variables. Hence,
the results reported below will be based on the pooled sample.
(23.) It seems reasonable that firms take such important
organizational decisions like offshoring and restructuring not
independent of each other in a given time period, but this does not need
to be the case.
(24.) A particular case occurs if a complete plant of a multiplant
company is closed and substituted for foreign intermediate inputs. We
still measure offshoring correctly, since the other plants of the
company will increase their foreign input share and reduce their
domestic one. We do not capture though the employment loss from the
closed plant--just as we will not be able to track employment losses in
German supply industry if domestic supply contracts are replaced by
foreign intermediate inputs (channel C in Figure 1). However, this is
not the purpose of this article, since we are interested in identifying
the employment effects on plants that increase their share in foreign
intermediate inputs (offshoring). For this purpose, it is important that
the employment loss from the closed plant does not appear in the control
group, which is not the case in our data.
(25.) We use psmatch2 provided by Leuven and Sianesi (2003) for
most of our matching estimations.
(26.) Abadie and Imbens (2008) conjecture that bootstrapping yields
valid inference for kernel, but not nearest neighbor estimators. Hence,
our standard errors for kernel matching estimators will be based on
bootstrapping and our standard errors for nearest neighbor matching
estimators are analytically derived in Abadie and Imbens (2006) and
calculated using the STATA-modul NNMATCH from Abadie et al. (2004).
(27.) Note that to the extent that some restructuring events might
be independent of the simultaneous offshoring activity in a given
period, the results in columns 2, 3, and 4 of Table 3 are biased away
from zero, with the baseline result in column 1 and the identification
of the productivity effect in column 5 being unaffected.
(28.) While both studies cover a similar time period in Ger many,
Wagner (2011) focuses on manufacturing firms with at least 100 employees
and his relocation measure also includes horizontal FDI. While it is
hard to say where the differences exactly come from, the firms in his
sample might--on average--adjust their employment more smoothly.
(29.) In an earlier version of the article, we offer an alternative
strategy to identify the productivity effect. We look at the employment
effects of firms that switch between European Monetary Union (EMU) and
non-EMU suppliers, whereby the overall foreign input usage does not
increase. Similar to our preferred specification, we cannot disentangle
A from D, but find positive and significant employment effects
consistent with the existence of the productivity channel.
(30.) It is not possible for us to consolidate plants at the firm
level, but in another unreported robustness check we have restricted the
sample to single plants, that is, independent firms with largely similar
results as in Table 3.
(31.) The only exception is column 4. Within the group of
offshoring firms, those firms with lower employment growth prior to the
offshoring decision are more likely to restructure at the same time.
(32.) During the sample period, the European Monetary Union
consisted (beyond Germany) of Austria, Belgium, Finland, France,
Ireland, Italy, Luxembourg, Netherlands, Portugal, and Spain. Greece
joined the EMU at the beginning of 2001.
CHRISTOPH MOSER, DIETER URBAN and BEATRICE WEDER DI MAURO *
* The authors would like to thank Steffi Bansbach, Daniel
Baumgarten, Martin Biewen, Irene Brambilla, Matilde Bombardini, Joseph
Francois, Marfa Garcia-Vega, Pierre-Olivier Gourinchas, Jorn Kleinert,
Marc Melitz, Niklas Potrafke, Michael Siegenthaler, Steffen Sirries,
Jeffrey Wooldridge, and Maurizio Zanardi for helpful comments. We are
grateful to the IAB for their hospitality and, especially Jorg Heining,
Peter Jakobebbinghaus, Dana Muller, and Alex Schmucker for support with
the IAB establishment panel as well as David Liechti for excellent
research assistance. For helpful suggestions and comments we thank
participants of the annual meeting of the Austrian Economic Association
in Vienna, the European Economic Association in Milano, the Verein fuer
Socialpolitik in Graz, the Third IAB-BA-User Conference in Nuremberg,
the KOF Brown Bag Seminar in Zurich, the Fifth Macroeconomics Research
Meeting in Tuebingen, the Volkswirtschaftliche Workshop at the
University of Tuebingen, the IO and Finance seminar at the LMU Munich,
the Second BBQ-Conference at ETH Zurich, the Otago Workshop on
International Trade, the EIIE conference in Ljubljana, the XI Conference
on International Economics in Barcelona, the European Trade Study Group
in Warsaw, the 8th ELSNIT conference in Paris, Swiss Society of
Economics and Statistics in Neuchatel, and the faculty seminar at the
Universite de Bruxelles and the University of Bayreuth. Obviously, all
remaining errors are our own. Moser and Weder di Mauro dedicate this
article to the memory of their friend and coauthor, Dieter Urban.
([dagger]) This is a completely revised paper that formerly
circulated under the title "Offshoring, Firm Performance and
Establishment-level Employment--Identifying Productivity and Downsizing
Effects."
Moser: Senior Researcher, ETH Zurich, KOF Swiss Economic Institute
and CESifo, Zurich 8092, Switzerland. Phone 0041-44-632-8311, Fax
0041-44-632-1218,
E-mail
[email protected]
Urban: Deceased.
Weder Di Mauro: Professor, University of Mainz and CEPR, Mainz
55128, Germany. Phone 0049-6131-39-20126, Fax 0049-6131-39-25053, E-mail
Beatrice. Weder@ uni-mainz.de
TABLE 1
Interpretation of Different ATTs
Treatment Group Control Group Theoretical ATT
Baseline
Offshoring Nonoffshoring w (A - B + C + D)
+ (l - w)
(A + C + D)
Identifying downsizing
channel
Gradual downsizing: Nonoffshoring A + C + D
Offshoring-without-
restructuring
Discrete downsizing: Nonoffshoring A - B + C + D
Offshoring-with-
restructuring
Relocation abroad: Offshoring- -B
Offshoring-with- without-
restructuring restructuring
Identifying productivity
channel
Productivity Nonoffshoring A + D
effect: Offshoring-
without-restructuring-
without-supplier-
substitution
Notes: A, B, C, and D refer to Figure 1. w is the share of
restructurers among offshorers. The treatment-variable
Offshoring is defined as an increase in the share of
imported intermediate inputs in overall intermediate inputs
either in the years 1999-2000 or 2001-2002 for a certain
plant. Offshoring-with-restructuring (Offshoring-without-
restructuring) comprises those plants that offshoring and
(do not) restructure their plant at the same time, that is,
(no) parts of the plant are closed down, sold-off, or spun-
off. The variable Offshoring-without-restructuring-and-
without-supplier-substitution covers all those Offshoring-
without-restructuring events, where there was no reduction
in sourcing to domestic suppliers at the same time. The
control group nonoffshoring is defined as those plants that
do not offshore during the same time period.
TABLE 2
Overall Effect of Offshoring on Employment
Baseline Alternative Nearest Matching Matching
(Kernel (Kernel Neighbor Within Within
Matching) Matching) Matching Industry Regions
Time (1) (2) (3) (4) (5)
t 0.0220 ** 0.0221 ** 0.0196 * 0.0282 ** 0.0223 **
(0.0103) (0.0106) (0.0107) (0.0125) (0.0110)
t+1 0.0419 *** 0.0419 *** 0.0410 *** 0.0450 *** 0.0440 ***
(0.0129) (0.0136) (0.0137) (0.0161) (0.0138)
t+2 0.0447 *** 0.0447 *** 0.0415 *** 0.0550 *** 0.0451 ***
(0.0151) (0.0148) (0.0157) (0.0188) (0.0154)
Notes: Standard errors in parentheses. For the baseline
sample (1) and the alternative kernel matching estimation
with a smaller bandwidth of 0.001 (2) standard errors are
generated via bootstrapping (500 replications). Column 3
shows the nearest-neighbor matching with 10 neighbors and
caliper = 0.05. For the matched sample heteroskedasticity-
consistent standard errors are generated with NNMatch from
Abadie et al. (2004). (4) Kernel matching, whereby matches
are only allowed between plants within the same industry (16
industries) and the average treatment effect on the treated
is equivalent to the average ATT's over the 16 industries.
(5) Kernel matching, whereby matches are only allowed
between plants within the same region (17 regions) and the
average treatment effect on the treated is equivalent to the
average ATT's over the 17 regions. The treatment-variable
Offshoring is defined as an increase in the share of
imported intermediate inputs in overall intermediate inputs
either in the years 1999-2000 or 2001-2002 for a certain
plant. Nontreatment is defined as those plants that do not
increase their vertical integration during the same time
period.
*** Significant at 1%; **significant at 5%; *significant at
10%.
TABLE 3
Identification of the Different Channels of Offshoring on
Employment (Kernel Matching)
Gradual Discrete
Baseline Downsizing Downsizing
Time (1) (2) (3)
t 0.0220 ** 0.0332 *** -0.0792
(0.0103) (0.0100) (0.0495)
t + 1 0.0419 *** 0.0661 *** -0.1792 ***
(0.0129) (0.0124) (0.0632)
t + 2 0.0447 *** 0.0656 *** -0.1595 **
(0.0151) (0.0150) (0.0718)
Relocation Productivity
Abroad Effect
Time (4) (5)
t -0.1475 *** 0.0373 ***
(0.0559) (0.0107)
t + 1 -0.2930 *** 0.0721 ***
(0.0704) (0.0149)
t + 2 -0.3070 *** 0.0696 ***
(0.0957) (0.0182)
w (A - B
+ C + D)
Theoretical + (1 - w)
ATT (A + C + D) A + C + D A-B+C+D
Treatment Offshoring Offshoring- Offshoring-
group without- with-
restructuring restructuring
Domestic Yes Yes Yes
substitution
No. of 1,265 1.143 122
treatments
Control Nonoffshoring Nonoffshoring Nonoffshoring
group
No. of 7,201 7,201 7.201
controls
Total 8,466 8,344 7,323
Theoretical
ATT -B A + D
Treatment Offshoring- Offshoring-
group with- without-
restructuring restructuring
Domestic No No
substitution
No. of 122 750
treatments
Control Offshoring- Nonoffshoring
group without-
restructuring
No. of 1,143 7.201
controls
Total 1,265 7,951
Notes: Standard errors in parentheses. For the matched
sample standard errors are generated via bootstrapping (500
replications). Number of observations refer to time t.
Observations that have to be dropped due to a lack of common
support never exceed ten observations. The baseline
treatment-variable Offshoring is defined as a (qualitative)
increase of a plant's share of foreign intermediate inputs
in total inputs either in the years 1999-2000 or 2001-2002
for a certain plant. In columns 2 and 3, Offshoring-with-
restructuring (Offshoring-without-restructuring) comprises
those plants that offshoring and (do not) restructure their
plant at the same time, that is, (no) parts of the plant are
closed down, sold-off, or spun-off. Column 5 employs
Offshoring-without-restructuring-and-without-supplier-
substitution as treatment covering all those offshoring-
without-restructuring events, where there was no reduction
in sourcing to domestic suppliers at the same time. The
control group nonoffshoring is defined as those plants that
do not offshore during the same time period.
*** Significant at 1%; ** significant at 5%;
* significant at 10%.
TABLE 4
Indirect Evidence on Supplier Substitution and
Productivity Channel
Intermediate Intermediate
Input Value Input/Sales Sales
Time (1) (2) (3)
t 0.0717 *** 0.2594 0.0511 ***
(0.0234) (0.6189) (0.0141)
t + 1 0.0491 ** -0.5830 0.0599 ***
(0.0238) (0.7160) (0.0175)
t + 2 0.0668 ** -0.7587 0.0729 ***
(0.0308) (0.7811) (0.0200)
Exports Productivity
Time (4) (5)
t 0.9608 *** 0.0368 ***
(0.3550) (0.0135)
t + 1 1.2009 *** 0.0304 *
(0.4367) (0.0168)
t + 2 1.3356 ** 0.0233
(0.5528) (0.0190)
Notes: Standard errors in parentheses. For the matched
sample standard errors are generated via bootstrapping (500
replications). The treatment-variable Offshoring is defined
as an increase in the share of imported intermediate inputs
in overall intermediate inputs either in the years 1999-2000
or 2001-2002 for a certain plant. Nontreatment is defined as
those plants that do not increase their vertical integration
during the same time period.
*** Significant at 1%; **significant at 5%; *significant at
10%.
