Does Investing Abroad Reduce Domestic Activity?
Evidence from Italian Manufacturing Firms
Raffaello Bronzini
PRELIMINARY (AND INCOMPLETE) VERSION
5TH OCTOBER 2008
DOES INVESTING ABROAD REDUCE DOMESTIC ACTIVITY? EVIDENCE FROM
ITALIAN MANUFACTURING FIRMS
By Raffaello Bronzini*
Bank of Italy, Structural Studies Dept.
[email protected]
Abstract
The goal of this paper is to evaluate whether the firms that invest abroad reduce domestic activity.
We take advantage of a unique long firm-level panel from the Bank of Italy Survey on
manufacturing firms that provides information on international activity of a representative sample
of Italian enterprises. We use matching methods and diff-in-differences estimates to compare the
performance of the firms becoming multinationals, before and after the investment, with that of
multiple control groups of domestic firms. We supplement the counterfactual strategy by studying
the over-time correlation between domestic and foreign activity through a different econometric
methodology.
Both methods lead to the same conclusion: in the short-medium run there are no significant
effects of the investment abroad on the firm’s domestic activity, as employment, skill
composition of labor force, sales, productivity or exports. If any, the impact seems more positive
than negative. Alike, domestic and foreign activity seems more complement than substitutes.
*
Bank of Italy, Structural Studies Dept. [email protected]
1
Introduction1
The increasing internationalization of production and its effect on the world economy is
at the center of the economic debate. Some observers fear that firms investing abroad might
relocate offshore some stages of production process previously realized at home. If such
activities are not substituted by other domestic productive activities, foreign direct investment
(FDI) would lead to an impoverishment of the domestic economy. This is felt as a threat by
some politicians; some of them are even willing to introduce subsidies in order to stop firms
from delocalizing.2
The main goal of this paper is to measure the impact of investing abroad on the domestic
activity of a sample of Italian manufacturing firms that have started to produce goods and
services abroad (switching) from 1989 to 2004. This is a challenging task. The main problem is
that we cannot merely compare domestic firms and multinationals in that they are heterogeneous
enterprises: a simple comparison of the performance between the two groups is likely to be
biased because of self-selection. In this paper we adopt a strategy of identification based on a
counterfactual analysis, typical of the program evaluation literature (Angrist and Kruger, 1999;
Blundell and Costa Dias, 2000). We match switching firms with control groups of domestic
firms that are similar as much as possible to switching firms. Then we compare the performance
before and after the investment through difference-in-differences (DID) estimates. Taking
advantage of some qualitative information, we are also able to match switching firms with those
that considered the possibility to invest but did not invest yet, capturing in this way domestic
firms that are closer to switching.
We supplement the analysis carrying out an out of sample test adopting a different
empirical strategy. Using a sub-sample of firms for which data are available, we explore the
over-time correlation between domestic and foreign productions to verify if they are
complements or substitutes. Notice that in this case we do not assess explicitly the impact of
switching but the type of relationship between home and foreign activities.
Both methods lead to the same conclusion. In the short-medium term there are no
significant effects of the investment abroad on the firm’s domestic activity in terms of
1
I would like to address a special thank to David Card, Enrico Moretti and the participants in the Center for Labor
Economics lunch seminar at the UC, Berkeley (March 2008) for their comments and suggestions. The views
expressed are those of the author and do not necessarily correspond to those of the Bank of Italy.
2
See e.g. the article: “Not so exceptional. French industry is taking on more Anglo-Saxon characteristics in: The
Economist (28th March 2008). In the cited article the case concerns France, but the same issue involves many other
countries.
2
employment, skill composition of labor force, sales, productivity or exports. If any, the impact
seems more positive than negative. Alike, domestic and foreign productions seem more
complements than substitutes.
This paper is related to an established literature that studies the impact of international
activity on firms performance with similar methods. Wagner (2002), Girma et al. (2004) and De
Loecker (2007), for example, assess the impact of exporting on firms productivity using
matching methods, the last two papers with DID estimates (for a review see Greenaway and
Kneller, 2007). As regards FDI the literature is small. Barba Navaretti et al. (2006) analyze the
impact of switching for a sample of Italian and French firms by applying propensity score
matching method and DID estimates. They distinguish investments in less developed countries
from those in advanced economies to disentangle the specific effect of investing in cheap labor
countries. They find positive impact on productivity and output of Italian firms from one to three
year after the investment, both in developed and less developed countries, and almost no effect
on employment. The results on French firms are rather different: there is a positive impact on
employment and a null one on productivity; the effect on output is positive only for investment
in developed countries. Hijizen et al. (2006) carry out a similar analysis on French firms. They
find that market-seeking investment have a positive impact on employment and productivity
while for factor-seeking investment the initial negative impact is compensated by a recovery
after two years from the investment.
This paper contributes to the existing literature in various ways. First, we take advantage
of a unique data set collected by the Bank of Italy. The data set provides key information on the
international production of the sampled firms as well as on some aspects so far unexplored
within the same framework, namely exports activity and changes of workforce skill-intensity.
An additional advantage is that data allow us to know firms that considered the possibility to
invest abroad but that they did not invest yet: this is important for our identification strategy.
Finally, we supplement our counterfactual approach with a different econometric strategy
analyzing the over-time correlation between domestic and foreign employment and output for a
sample of multinationals firms.
This paper is structured as follows. In the next section we discuss the theoretical
background. In section 3 we present our main empirical strategy and data. In the following
section we discuss the results and present some robustness checks. Section 5 is dedicated to the
second empirical strategy. Section 6 concludes.
3
2
Background
The traditional theory of foreign direct investment (FDI) distinguishes between Vertical
and Horizontal FDI (see e.g. Markusen, 2002). Horizontal FDI occurs when firms produce
similar goods or services in multiple countries to overcome trade barriers, reduce transport costs
or because they benefit from being close to the final customers.3 In the Vertical FDI firms
fragment the production process into stages to take advantage of differences in input prices; thus
activities are located across countries according to the relative endowment of inputs. It is argued
that domestic and foreign productions are substitutes in the case of horizontal foreign
investment, in that multinationals produce and sell directly in the final markets, and
complements in the case of vertical FDI, because the activities located in different countries
represent stages of the same production process (Markusen and Maskus, 2001; Markusen, 2002).
This theory helps us to predict the impact of investing abroad on firm’s domestic activity
only partially.
First, because such approach has been recently questioned by the empirical evidence and
theoretical analyses. On the empirical ground it has been observed that the large majority of the
foreign direct investment cannot be fully classified into one of the two categories. Unctad (1998)
coined the terms “complex integration strategies” to illustrate how firms found new forms of
internationalization that are outside the frame of the vertical-horizontal paradigm. Firms may
broke down the production process into different stages and divide these stages into different
countries, combining market-seeking (horizontal FDI) with factor-seeking strategies (vertical
FDI); they may also produce in cheap labor countries in order to benefit from the low input costs
and sell the output in third markets, creating the so-called export-platforms. Complex strategies
are also documented by Feinberg and Keane (2006) who found that only 12 per cent of the US
multinationals with affiliates in Canada can be classified as purely horizontal (they have
negligible intra firm flows of intermediate inputs) and only 19 per cent as purely vertical (they
have negligible intra firm intermediate inputs’ flows of in one direction only). On the theoretical
ground, Yeaple (2003), for example, shows how firms in advanced country can undertake
complex strategies by investing in other advanced countries to reduce transport cost and in the
less developed countries to take advantage of price differentials. The equilibrium strategy
3
For example because in that way they are able to better meet consumers’ tastes. Similarly Nocke and Yeaple
(2007) argued that firms engage in FDI also to acquire non-mobile capabilities located in other countries, as
distribution or country-specific institutional capabilities.
4
depends on the combination of transport costs, fixed costs of investing abroad and factor prices
differentials. In his model foreign and domestic production can be either complements or
substitutes. In a similar model by Ekholm et al. (2007) also export-platform strategies can arise
in equilibrium. Grossman et al. (2006) develop this framework by introducing firms’
heterogeneity. They find that the degree of firms’ internationalization is positively correlated
with the level of productivity and that export platform strategies are undertaken mainly by the
more productive firms.
Of course, when firms pursue complex strategies the sign of the link between domestic
and foreign activity becomes unclear. Even in the case of pure vertical or horizontal strategies,
however, this theoretical paradigm can only in part predict what happens to the firms switching
from domestic to multinationals.
If we look at domestic activity as measured by sales, exports and employment the
expected effect of switching is ambiguous. Domestic production may decrease if firms move
part of the production process previously produced at home, both in vertical and horizontal FDI.
However, it could also increase if firms through the investment are expanding their production or
if there are some complementarities between domestic and affiliates production lines; e.g. parent
firms may supply certain types of services or inputs, as management or marketing, to
subsidiaries. A positive relationship can also occur when, thanks to the investment abroad, firms
are more competitive and gains market shares.
To invest abroad might also have structural effects on the home workforce’s skillintensity. If firms displaces the less skill-intensive production stages, as it should happen for
firms located in advanced countries in the case of vertical FDI, we assist to a skill upgrading of
the workforce after the investment. This is not the only possible effect, though. Firms could also
move the more skilled stages of the production, for example some firms could locate R&D
activities in those advanced countries where skilled workers are abundant, or when birth or
acquisition of foreign firms induces to transfer managers or supervisors to guide the subsidiaries.
Even for productivity the expectation is not unambiguous. We envisage that by relocating
or acquiring some activities a firm should improve its productivity: by scale effects, rationalizing
the division of the labor across countries or save on input prices thanks to the investment, firms
should obtain efficiency gains. However, the positive effect is likely to occur in the long term.
But in the short run we can imagine also a negative impact on productivity if there are
adjustment costs or frictions in the new firm’s international organization of labor.
5
Finally, as regards investment our a priori is that owing to budget constraints domestic
investment should decrease, at least in the short run. Of course, if the budget constraint is not
binding foreign investment could also be uncorrelated with the domestic one.
Summarizing, theory is unable to accurately predict the impact of investing abroad on the
level of domestic activity. Thus, to understand the home effect of FDI remains mainly an
empirical issue.
3
Empirical strategy and data
Our goal is to evaluate the causal effect of investing abroad (switch) on the firm’s
domestic activities. For this purpose we would like to observe the same firm in two different
settings, one in which it gets multinational and another in which it remains domestic, to answer
the question: what would have been the domestic activity if the firm had not invested abroad?
Formally, let yit be our outcome variable (employment, sales, exports, etc.) of firms i at time t,
and SWITCHit={1,0} be an indicator if the firms i switches from domestic to multinationals at
time t. The causal effect of switching on the variable y at time t+s is defined as: (y1it+s-y0it+s),
where y1it+s is the value of the variable y of the switching firm i after the investment and y0it+s the
value of the same variable in the same period if the firm i had not switched. The problem is that
y0it+s is unobservable for the firms that have switched. To overcome such a problem we follow
the traditional approach of the program evaluation literature (e.g see: Angrist and Kruger, 1999;
Heckman et al., 1997). We define the average impact of switching on the variable y at time t+s
as:
E{y1it+s-y0it+s|SWITCHit=1}= E{y1it+s|SWITCHit=1}- E{y0it+s|SWITCHit=1}
(1)
and since y0it+s is unobservable we try to construct a valid counterfactual for the last term
of the equation (1) by choosing a control group of firms that had not switched. In particular, the
impact is estimated by substituting the last term of the equation (1) with: E{y0it+s|SWITCHit=0},
the average of the outcome variable for a sample of firms that did not switch.
The challenge of this strategy is the construction of a valid control group, that is to
choose, among the firms that did not switch, those that are similar to the switching firms as much
as possible. Ideally they should differ from the switching firms only for not having switched. In
our case the task is particularly challenging because the choice of investing abroad is
6
endogenously made and self-selection bias is likely to occur. For example, it is known that
multinationals are larger, more productive, with a higher export propensity and R&D outlays
than domestic firms (see e.g. Markusen, 1995; Helpman et al., 2004). Hence, merely comparing
the performance of the two groups could lead to biased results on the impact of
internationalization just because multinationals on average perform better than domestic firms.
Our strategy tries to remove the potential self-selection bias in several ways. First, we use
one-to-one matching methods, comparing only switching and domestic firms that belong to the
same 2-digits sector and that are very similar according to several observables, in levels and
trends, before the investment. To choose the domestic firms belonging to control groups we rely
on covariates and propensity score matching methods (Rosenbaum and Rubin, 1983).4 We
combine the method of matching with the difference-in-differences estimator to improve the
robustness of the results. In particular, DID method is powerful because allows for timeinvariant differences in observables and non-observables between switching and domestic firms
(Smith and Todd, 2005a). With DID the impact of the investment abroad is estimated by the
change of the difference between switching and the control group after the investment. Formally,
DID=[E(y1it*+s)-E(y0it*+s)]-[E(y1it*-s)-E(y0it*-s)], where t* is the year of the first foreign
investment.
Second, we take advantage of a key qualitative information provided by the Bank of
Italy’s survey. Using a combination of the firms’ answers we are able to select the firms that
have considered the possibility to produce abroad, but that did not invested yet. Such a
information turns out to be very important for the construction of additional control groups, since
it reduces heterogeneity between domestic and switching firms. The stronger closeness of the
two groups, documented later, encourages our supplementary identification strategy. Third, we
collect evidence from different control groups, specifications of the econometric model and
matching methods for robustness-checks purposes.
The econometric model for the DID estimates is the following:
4
The metric used for the one-to-one matching on covariates is given by the following distance function, a variant of
the Mahalanobis metric (Rosenbaum and Rubin 1985): DF=│∑k∑j ωk(Xki- Xkj)│; where i=switching firms,
j=domestic firms (of the same 2-digit sector); X=is the set of observables (employment, sales, export, annual change
of sales and employment in logarithm); ωk= weight assigned to observable k. The explanatory variables employed in
the propensity score equation are the log of: employment, export propensity (export/sales) and skill-intensity (share
of white over blue collars). For the propensity score matching we used the method of nearest neighbor with
replacement.
7
yit= αi + αt+ αr + αp + β1(Post) + β2(SWITCH) + γ(Post* SWITCH) + Σj θjZjit
(2)
where, αi, αt, αr, αp are full sets of fixed effects at level of firm, year, region of
localization of the firms, pairs of firms (each treated and its control), respectively; SWITCH=1
for switching firms and 0 for the controls; let τi be a time indicator equal to 0 in the year of the
investment abroad; we define Post=1 if τi>=0; Post=0 if τi<0; Zit is a vector of firms covariates
including sales, exports, investment and employment.5 Our outcome variables y are the log of:
employment, sales, investment, skill-intensity, labor productivity and exports; all referred to the
domestic activity. γ is the parameter of interest: it measures the change in the difference of the
outcome variable between switching and controls after the investment; it is our estimate of the
effect of switching on the domestic activity.
DID estimator is implicitly based on the common trend assumption: the validity of the
inference is undermined if the two groups show different trends in the outcome variables before
the treatment (Blundell and Costa Dias, 2000; Blundell et al., 2004). Therefore, we carefully
check that in the pre-investment period switching and control groups firms have no significant
differences among the growth rates of the main variables studied. However, even if firms show
similar trend before the treatment, it is possible that after the treatment macro changes will have
different impact across the two groups. In our case, for example, switching firms can react to
exogenous macro shocks in the world demand differently from firms that remained domestic. By
including the vector of time varying firms variables Zit in the model we should control for such
changes.
Data
Data are drawn by the Bank of Italy survey of industrial and services firms. The survey is
conducted annually from the early ‘80s on a representative sample of about 4,400 firms. In this
paper only firms of the manufacturing sector are considered. In 2004 the Bank of Italy asked
firms with more than 50 employees information on their international activity: e.g. if firm
produces goods or services abroad, if the offshore activity was mainly manufacturing and when
the firm started to produce abroad.
We use the surveys from 1984 to 2006 to select firms that during this period switched
from domestic to multinationals and the corresponding control groups. The samples of switching
5
In each equation we include only the covariates that do not appear on the left hand side. For example, when the
outcome variable are sales/employment only investment and exports are included (notice that some robustness
checks assured that outcomes are not sensitive to small change of covariates’ set).
8
and control groups are balanced over a period of four years: starting two years before the
investment and finishing one year after. Even if a long time-window can be more interesting for
policy purposes, stretching the period reduces the number of the switching firms and can also
weaken the reliability of the evaluation exercise because the identifying assumption is less likely
to hold over time, when many other things are likely to happen and confound the effect of
switching.
We illustrate our strategy to construct the group of switching firms and its controls with
the help of Figure 1. We first asked the firms if they produced goods or services abroad or if they
considered this possibility in the last two years. We denote domestic firms those that answered
no (Figure 1; Panel C). Then, we asked those that answered yes to the first question if they
actually produced goods or services in the last year (Panel B); we call the firms that answered no
“near-investing” domestic firms (Panel E).6 Next, among the firms that produced abroad (we
called them multinationals; panel D) we select the switching firms as those that started to
produce abroad in the interval 1984-2004 and that are continuously observed for four year across
the first foreign investment (Panel F) and we separate the subgroup of switching firms that
produce abroad mainly goods (Panel G). In our exercise we match switching firms (as defined in
Panel F and G) to control groups drawn by domestic and near-investing firms samples.
In Table 1 we describe our firms’ samples. There are 1,668 firms that answered the
question on the internationalization. 270 are multinationals, around the 16 per cent, while the
switching ones are 89 (59 of them produce abroad mainly goods).7 Among the remaining
domestic firms 280 are near investing and 1,118 are other domestic firms. The table confirms the
well-known characteristics of multinationals: they are larger, with higher export propensity,
more productive, older, pay a larger wage, employ more skilled workers and invest more in
R&D activity than domestic firms. According to our samples, the investment over employees
and profits are on the contrary smaller. Switching firms do not noticeably differ on average from
multinationals, it is just worthwhile mentioning the presence of larger wages and human capital,
together with smaller productivity, in the former with respect to the latter. Among the switching
6
In principle among the firms that replied no we might find also firms that were producing abroad during 2003 but
that stopped in 2004. Therefore, as robustness exercise we take advantage of the 2006 wave of the survey when it
was asked to the firms if they have produced goods or services abroad in the period 2000-2006. Only 9 out of the 59
firms of the near investing firms answered yes and can be considered firms suspected to have stopped to produce
abroad in 2004. However, results are substantially unchanged by excluding these firms from the DID estimates
(results are not shown but available under request).
7
The 5 largest firms (more than 8,000 employees) have been excluded for the impossibility of finding an
appropriate matching.
9
firms, those that produce abroad mainly goods are larger, less productive and with less human
capital employed than the other switching firms.
Among the domestic firms, the “near-investing” ones seem nearer to the switching than
the other domestic firms. The closeness is stronger if they are compared to switching that
become multinationals in 2003-2004, the years over which the “near-investing” have thought to
invest abroad but they did not: if we contrast column (5) with column (6), in Table 1, we notice
the stronger proximity between the two groups for the majority of the variables, an evidence that
encourages our strategy.
Table 2 illustrates the distributions of switching firms by year of the first foreign
investment, sector and region of localization. The large majority of the firms switched from 1998
to 2004 (about 70 per cent). The most represented sectors are mainly those of the Italian
specialization: machinery and equipment (with electrical machinery) and some traditional
sectors (as leather products, other manufacturing industries and textile, wearing and apparel). As
regards to the region of localization, as expected, the number of switching is larger in the North
(in particular Piedmont, Lombardy, Emilia Romagna, Friuli Venezia-Giulia and Veneto); in the
Center-South the number of switching is small (with the exception of Marche and Puglia).
4
Results
We start by matching switching firms (as defined in Panel F and G of Figure 1) with the
sample of “other domestic firms” (see Figure 1, Panel C). We called this Matching #1. We use
the method of matching on covariates described before to construct the control group for the
switching firms. Since the matching is carried out on the same sector and over the same time
span, for a few firms it was impossible to find an appropriate match.8
In the Appendix (Table A1) are reported the means of several observables and their time
changes for the two groups. The comparison of the growth rates is important because DID
assumes no differences in trend before the treatment. In order to measure the similarity between
the two groups, we report the differences in means and the standardized difference (SDIFF) of
several variables between the two groups. The SDIFF of the variable y is given by the difference
in means between switching and matched controls divided by the squared root of the average
variances of the variable y in the two groups. Formally, SDIFF(y)=100(1/N)[Σi(yi)-
8
This reduce the sample of switching firms by 4 in the case of Matching #1 and by 7 in the case of Matching #2, see
further.
10
Σj(yj)]/[Var(yi)+Var(yj)/2]1/2, where i denotes switching firms and j firms of the control group
(Smith and Todd, 1995b). The lower the two balancing tests, the better is the matching in terms
of the variables under consideration. Concerning SDIFF, there is not a formal criterion for
indicating a SDIFF too large; hence, as it has become common in the literature, we follow
Rosenbaum and Rubin (1985) and consider large a value of 20%.
Overall the matching seems rather satisfying. Mean and standardized differences between
the two groups are rather small: for almost all the observables in levels SDIFF are far below 10%
(e.g. it is 0.71% for employment, -2.6 for sales, 6.5 for export and -5.0 for productivity); for
investment and profits we find larger differences (19.7% and 16.5%, respectively). Mean
differences in the growth rates are larger but still within reasonable boundaries; again,
investment is the exception. This is understandable since investment varies remarkably over time
and it is likely that switching firms follow a different investment cycle with respect to the
domestic ones; undoubtedly this reduces the reliability of the DID estimates for investment.
When we consider the sub-sample of firms that produce abroad mainly goods, the difference are
larger for employment, sales, and exports, but smaller for investment. Overall, means differences
are always not significant at the standard significance levels except that for the growth rates of
investment.
In the Figure 2-3 are reported the non-conditional means over the 4 year-time span
including the switching. The DID estimates of parameter γ for various models are reported in
Table 3. The impact of investing abroad seems more positive than negative, at least for the
whole set of switching firms. In the model with firm covariates, the change in the difference
between the two groups for employment and sales of switching is 0.9 and 2.2 per cent,
respectively. Also for productivity and exports the impact seem positive (1.4 and 3.4 per cent,
respectively); while for investment and skill share the opposite occurs. All the differences are not
very high in magnitude (with the exception of investment) and as expected none is statistically
significant.9
For the sub-samples of switching firms with offshore activity mainly in manufacturing,
the results are somewhat different. Employment in that case seems smaller (by 0.6 per cent), and
all the other variables, except investment, turns out to be larger; such differences however
remain statistically non-significant. In Table 4 we decompose the impact by year. The evidence
9
We report results with standard errors clustered by region. Other clustering tested (e.g. by firm) do not change
substantially the results.
11
is mixed for the whole sample, while for firms that produce abroad mainly goods the
improvement during the year after the investment occurs for all the variables but exports.10
In Matching #2 we compare the switching firms to a control group drawn by the sample
of the near-investing firms, as identified in Panel E of figure 1. The balancing properties are
shown in the Appendix, Table A2. Overall the two samples are very similar in the means for the
majority of the variables. Mean differences are never significant and SDIFF is much smaller than
20%. For the sub-sample of switching with mainly manufacturing offshore activity the balancing
is a little less satisfactory, though.
DID results are reported in Table 5 (see also Fig. 4-5). In that case the coefficient of the
DID for employment is negative, but again small in magnitude (0.4%). Switching firms show
higher sales and productivity after the investment, and lower skill intensity and exports. Again
these differences are small and not statistically significant at the conventional levels. For the subsample of switching with mainly manufacturing offshore activity, investing abroad seems to
have a positive impact on employment, sales, productivity and exports; negative coefficients
characterize instead investment and skill intensity. These differences are not remarkable,
however, and are statistically non-significant except that for skill intensity.
We can now try to summarize our results from DID estimates. Overall, employment
differences between switching and the control groups after the investment tends to be more
positive than negative but low in magnitude. On average there is little evidence of substitution
between domestic and foreign employment and to some extent the two figures seem
complementary. Sales and labor productivity DID coefficient are always positive. This would
denote as if the firms through the foreign investment expand the dimension of the business
activity enhancing efficiency. A second regularity found is that for switching firms’ investment
at home is substantially lower after the internationalization if compared to the domestic firms.
Even if we should interpret this figure with extreme caution because of the imperfect
comparability of investment between the two samples in the pre-treatment period, the outcome
seems reasonable: it is likely that home and foreign investment turns out to be substitute, at least
in the short term, as a consequence of the budget constraints that firms face. As regards the skill
intensity the results tend to be negative, a somewhat puzzling result; we would have expected
that firms in advanced economy move abroad mainly unskilled intensive activity and hold or
expand the skill intensive production stages at home. A possible explanation is that firms move
managers or supervisors in order to coordinate or guide the offshore activity. Such results, which
10
Notice that because no control for autocorrelation is done standard error could be biased (Bertrand et al 2004),
thus we do not pay much attention to the statistical significance of the coefficient.
12
are not new for advanced countries (e.g. see Blomstrom et al., 1997), can also reflect
international comparative advantages. In the international division of labor, Italy is relatively
scarce in some more skilled workers. As a consequence it is possible that firms moved outside
also the more skill intense phases of production, like the R&D activity. As regards exports, in the
majority of cases the DID coefficients is positive, suggesting again that foreign and domestic
production can be considered as complement. It is worth recalling however that all the
differences mentioned are small in the absolute value and almost always they are not statistical
significant at the conventional levels.
Robustness checks
We carry out a couple of exercises to check for the robustness of our results. First of all,
we employ a different method to match switching and domestic firms: the propensity score by
the nearest neighbor matching method. At the first stage we estimate a logit function with the
switching indicator as dependent variable and using as covariates the log of: employment, export
propensity (export/sales) and skill intensity (share of white over blue collars). These are among
the most relevant characteristics that differentiate multinationals from the other firms (see e.g.
Markusen, 1995). Switching firms are matched with all the domestic firms, including the nearinvesting ones, that belong to the same sector. We choose not to distinguish between the “nearinvesting” and other domestic firms to leave more degrees of freedom to the estimation of the
propensity score. All the explanatory variables in the logit model have the expected positive sign
and are statistically significant. Unfortunately, the matching by propensity score reduces the
number of switching firms by about 10% because of the common support condition imposed.
The means and standardized differences between the two samples are reported in Appendix
(Table A3). We note larger differences than in Matching #1 in employment, sales and exports.
Interestingly enough, the export propensity is larger in the control group than in the switching
firms. For almost all the variables, mean differences are not significant.
DID estimates are reported in Table 6. They confirm the previous findings for
employment, investment and skill intensity; on the other hand sales and productivity show
negative coefficients. However, as in the previous estimates, all the differences are of little
magnitude and except that in one case (employment) they are not statistically significant at the
conventional levels.
One potential weakness of the analysis carried out so far derives from the survivorship
bias. The problem arises if the survivorship rates of switching and domestic firms diverge after
the internationalization. If the probability to survive decreases for switching firms, e.g. because
13
investing abroad is risky and a share of the switching firms could die after the investment as a
consequence of an unsuccessful investment, DID estimates are upward biased: owing to attrition
we are able to observe only the best switching firms (of course bias arises only if the probability
to fail for switching is larger than the probability to fail if the firm remained domestic). In such
case our mostly positive results can be undermined.11 Another source of positive bias comes
from the possibility that firms move in the other countries the whole production processes and
close the activity at home because of the foreign investment. Since we are not able to follow
these firms over time in that they exit from our panel, positive bias could affect our sample.
We address the problem of attrition in the following way. We assume that the
survivorship rate does not substantially change immediately after the investment but only
gradually over time. We consider this a reasonable assumption. If a firm made the wrong
investment it should still have the resources to survive for some years after switching. This is
even more likely for our sample that includes large switching firms. Alike, in the event that firms
move away the whole production process and close the activity at home, it is reasonable to
assume that the cessation of domestic activity occurred after some years from the first foreign
investment. This would happen because the first investment is riskier and we envisage that firms
decide to close home activities after the initial investment abroad has proved to be successful.
In the light of these considerations we replicate the estimates on a sub-sample of
switching firms. We restrict the analysis on switching firms that internationalize in a period close
to 2004, the year in which the survey was carried out. By reducing the time window closely
enough to the year of internationalization we believe to be able to reduce, if not completely
eliminate, the positive bias. The time restriction is arbitrary. We present the results of the
analysis for the sub-sample of switching that invested from 1999 onward selected from the
samples used in matching #1 (notice that changing the starting year or using the matching #2
sample does not affect our results).
The comparison of the switching and the control group is shown in the Table A4 of the
appendix. Switching firms are on average smaller than those of the control group (in term of
employment and sales) and show higher export propensity. The DID estimates results are
reported in Table 7. We notice that for the large majority of the outcome variables the results
confirm those obtained with the previous exercises. The impact is positive on sales, productivity
and exports, and negative on investment and skill-intensity. The only difference is that in this
11
Notice that the opposite case does not create serious problem in our model. In the case of a negative bias, e.g.
because investing abroad increase the survivorship rate, our (mainly) positive results would be even larger without
bias (the sign would be the same).
14
case the effect on employment is negative, although very small (-0.6% for the wholes switching
sample); for all the variables the DID coefficients are not statistically significant. From these
findings it seems that attrition can bias our results only marginally.
5
Further evidence from an alternative empirical strategy
In this section we present some evidence on the correlation between domestic and foreign
activity on a sample of manufacturing firms for which we know the level of employment in, and
the amount of sales of, foreign activities in 2000 and 2006. Notice that unlike the previous
exercise here only firms that have produced abroad goods or services during this period are
examined.
There is an established literature that studied the degree of substitution between domestic
and foreign activity at the firm level. As regard employment, Blomstrom et al. (1997) regress the
employment of US and Swedish parent firms on sales of foreign affiliates controlling for the
level of domestic output. They conclude that foreign sales are negatively correlated with
domestic employment for US firms, while the opposite occurs for Swedish multinationals. A
different group of studies estimated labor demand equations to assess the degree of substitution
of labor employed in different locations by testing cross-countries wage elasticities. For
example, Brainard and Riker (1997a,b) focused on US multinationals and found that the type of
relationship depends on the locations: labor is complementary if affiliates are located in
countries similar in factor endowments, while substitutonary relationship occurs when they are
located in countries that differ in input endowment. Braconier and Ekholm (2000) followed a
similar approach focusing on Swedish multinationals. They found evidence of substitutionary
relationships of employment in parent firms and affiliates in high income countries, but no
evidence of substitution when affiliates are in low-income countries. Harrison and McMillan
(2007) study further the impact of changes in foreign affiliate wages on US firms’ employment
distinguishing between horizontal FDI and vertical FDI. Their paper shows that in horizontal
FDI domestic and foreign employment tend to be substitute, and the opposite occurs in vertical
FDI.
Because we do not know from our data set firms’ wages in the offshore activities we are
unable to estimate cross-countries wage elasticities, thus in this paper we follow an approach
more similar to that of Blomstrom et al. (1997).
15
In the 2006 survey, 210 firms reported to have produced outputs abroad in the period
2000-2006; among these 101 were interviewed in 2000 survey. This sub-sample of firms is the
object of our analysis. We explore the dynamic of domestic employment assuming that it is a
function of the level of domestic and foreign activity:
log(E)it= αi + αt+ β1 log(Domestic Activity)it + β2 log(Domestic Activity)2it + β3 log(Foreign
Activity)it + ΣsδsTrends + ΣrδrTrendr + εit
(3)
where Eit is the domestic employment of firm i at time t. We include a full set of firms
specific and year fixed effects, to control for firms’ heterogeneity and common time shocks
affecting firm’s labor demand. We include also sector and region specific trends to allow for the
dynamics of labor markets that influence labor demand by sector, as changes in industrial
relationships or sector specific business fluctuations, and differences in regional economic
growth. As proxy for domestic activity we use sales at the current prices (unfortunately we do
not have information on the value added); sales are also squared to take account of possible non
linearity.12 Two proxies for the level of foreign activity are used: the employment in, and the
sales of, foreign affiliates.
In order to control for individual fixed effects we take the time differences from 2006 and
2000 of the model (3) and estimate the following equation:
Δlog(E)it = α + β1 Δlog(Domestic Activity)it + β2 Δlog (Domestic Activity)2it + β3Δlog(Foreign
Activity)it + δs + δr + ηit
(4)
where Δy= y2006-y2000.
The results of the regressions are presented in Table 8 and Table 9. Our coefficient of
interest is β3; it measures the partial correlation between home employment and foreign activity.
In Table 8 domestic employment is regressed on the foreign employment. Different
specifications are presented for robustness purposes. The fit of the model is rather satisfying; one
third of the variance of the domestic employment is explained by the model without fixed effects
and almost half with regional and sectoral fixed effects. We notice that in the first three columns
(with OLS estimates) the coefficient of foreign employment is always positive. According to the
12
For homogeneity purposes we do not deflate domestic sales because we know the prices’ dynamic for domestic
sales but not for the foreign ones. However, to use domestic sales at constant prices changes only marginally the
results (the results of the exercise are not shown but available under request).
16
estimation a 1 per cent increase in the foreign employment is correlated with about 0.02 per cent
increase in domestic employment. This coefficient is relatively stable across the model
specifications and turns out to be non-statistically significant at the standard confidence
intervals.
Our model is not based on a specific theory, rather it investigates the partial correlation
between domestic and foreign employment. It is also possible that changes in domestic
employment induce changes in foreign employment. In such a case foreign employment is an
endogenous variable and the estimation of the correspondent coefficient would be biased. We
deal with the potential endogeneity problem through instrumental (IV) method and 2SLS
estimates. We use the level of foreign employment in year 2000 as instrument for the changes in
the foreign employment. According to a simple preliminary exploration the instrument is
strongly correlated with the dynamic of foreign employment, with coefficient equal to -0.614,
but only weakly correlated with the dynamic of domestic employment: the correlation is equal to
-0.103 (non-significant at the conventional levels). In the column 4 of Table 8 we report the
results of the IV estimation. The coefficient of foreign employment is still positive but higher
than the precedent estimates (0.038). However, it remains statistically non-significant at the
standard level.
In column (5)-(8) we present the estimates for the sub-sample of firms that have a foreign
activity mainly manufacturing. In such a case the coefficient is still positive but higher than that
found for the full sample; for the most complete model, with both OLS and IV estimations, it is
also statistically significant.
In Table 9 we replicate the exercise using the foreign sales instead of employment as
proxy for foreign activity. In that case the instrument is the foreign sales in 2000 (the correlation
between the instrument and the growth of foreign sales is equal to -0.722; with the dependent
variable is equal to -0.06). We find that the coefficient of the foreign activity is always positive,
but smaller in the magnitude than those previously estimated. The coefficient appears
statistically significant only in the sub-sample of firms producing abroad mainly goods and
estimated by IV.
Overall, these results seem to confirm the previous ones. The relationship between
domestic and foreign activity is relatively weak and tends to be more positive than negative.
6
Conclusions
To be done…
17
References
Angrist J. and Kruger A. (1999), Empirical Strategy in Labor Economics, in Ashenfelter O. and
Card D. (eds.), Handbook of Labor Economics, Vol. IIIA, 1277-1366.
Barba Bavaretti G., Castellani D. and Disdier A.C (2006), How Does Investing in Cheap Labour
Countries Affect Performance at Home? France and Italy, CEPR working paper 5765.
Bertrand, M., Duflo E. and Mullainathan S. (2004), How Much Should We Trust Difference-inDifferences Estimates?, Quarterly Journal of Economics, 119, 249-275.
Blomström M., Fors G. and Lipsey R. (1997), Foreign Direct Investment and Employment:
Home Country Experience in the United States and Sweden, Economic Journal, 107,
1787-1797.
Blundell R. and Costa Dias M. (2000), Evaluation Methods for Non-Experimental Data, Fiscal
Studies, 21, 427-468.
Blundell R., Costa Dias M., Meghir C. and Reneen J. (2004), Evaluating the Employment
Impact of a Mandatory Job Search Program, Journal of European Economic Association,
2, 569-606.
Braconier H. and Ekholm K. (2000), Swedish Multinationals and Competition form High- and
Low-Wage Countries, Review of International Economics, 8, 448-461.
Brainard S. and Riker D. (1997a), Are Multinationals Exporting US Jobs?, NBER working paper
5958.
Brainard S. and Riker D. (1997b), US Multinationals and Cometition from Low wage
Countries?, NBER working paper 5959.
De Loecker J. (2007), Do Exports Generate Higher Productivity? Evidence from Slovenia,
Journal of International Economics, 73, 69-98.
Ekholm K., Forslid R. and Markusen J. (2007), Export-Platform Foreign Direct Investment,
Journal of the European Economic Association, 5, 776-795.
Feinberg S. and Keane M. (2006), Accounting for the growth of MNC-based Trade Using a
Structural Model of U.S. MNCs, American Economic Review, 96, 1515-1558.
18
Girma S. Greenaway D. and Knelleer R. (2004), Does Exporting Increase Productivity? A
Microeconometric analysis of Matched Firms, Review of International Economics, 12,
855-866.
Greenaway D. and Kneller R. (2007), Firm Heterogeneity, Exporting and Foreign Direct
Investment, Economic Journal, 117, 134-161.
Grossman G., Helpman E. and Szeidl A. (2006), Optimal Integration Strategies for Multinational
Firm, Journal of International Economics, 70, 216-238.
Harrison A. and McMillan M. (2007), Offshoring Jobs? Multinationals and US Manufacturing
Employment, mimeograph.
Heckman J., Ichiumura H. and Todd P. (1997), Matching as an Econometric Evaluation
Estimator, Review of Economic Studies, 65,261-294.
Helpman E., Melitz M. and Yeaple S. (2004), Exports versus FDI with Heterogeneous Firms,
American Economic Review, 94, 300-316.
Hijzen A., Jean S. and Mayer T. (2006), The Effects at Home of Initiating Production Abroad:
Evidence from Matched French Firms, mimeograph.
Markusen J. (1995), The Boundaries of Multinational Enterprises and the Theory of
International Trade, Journal of Economic Perspectives, 9, 169-189.
Markusen J. (2002), Multinational Firms and the Theory of International Trade, MIT Press.
Markusen J. and Maskus K. (2001), General-equilibrium Approaches to the Multinational Firm:
A Review of Theory and Evidence, NBER working paper 8334.
Nocke V. and Yeaple S. (2007), Cross-border Mergers and Acquisitions vs. Greenfield Foreign
Direct Investment: The Role of firm Heterogeneity, Journal of International Economics,
72, 336-365.
Rosenbaum P. and Rubin D. (1983), The Central Role of Propensity Score in Observational
Studies for Causal Effect, Biometrika, 70, 41-55.
Rosenbaum P. and Rubin D. (1985), Constructing a Control Group Using Multivariate Matched
Sampling Methods that Incorporate the Propensity Score, American Statistician, 39, 3338.
Smith J. and Todd P. (1995a), Does Matching overcome LaLonde’s Critique of
Nonexperimental Estimators?, Journal of Econometrics, 125, 305-353.
Smith J. and Todd P. (1995b), Rejoinder, Journal of Econometrics, 125, 365-375.
Unctad (2008), World Investment Report: Trends and Determinants. United Nations
Conferences on Trade and Development, New York.
19
Wagner J. (2002), The Causal Effects of Exports on Firm Size and Labor Producitivy: First
Evidence from a Matching Approach, Economic Letters, 77, 287-292.
Yeaple S. R. (2003), The Complex Integration Strategies of Multinationals and Cross Country
Dependencies in the Structure of Foreign Direct Investment, Journal of International
Economics, 60, 293-314.
20
Tables and figures
Table 1
All samples: Means in 2004
Multinationals (MN)
Domestic Firms (DF)
All MN
MN producing
abroad mainly
goods
(1)
(2)
(3)
Switching
producing
abroad
mainly
goods
(4)
639.2
683.4
627.6
Sales
215017
207120
Exports
93234
Exports/Sales
Switching
(within all
MN)
Switching
in 20032004
Nearinvesting
abroad
Other DF
(5)
(6)
(7)
727.7
326.6
332.6
238.9
155560
182976
69870
89924
69877
88748
73555
88524
33079
40946
19593
0.478
0.467
0.498
0.497
0.512
0.399
0.291
Investment
6077
6838
5845
7060
2019
3187
2731
Skill intensity
0.384
0.354
0.368
0.313
0.371
0.363
0.296
Sales/Employe
es
322.7
263.7
242.8
254.4
230.2
260.1
251.1
Investment/E
mployees
8.496
8.512
9.027
9.440
7.285
8.634
10.011
Start year
1963.8
1963.9
1959.9
1960.1
1964.2
1969.5
1970.4
Profits
2.31
2.30
2.43
2.42
2.54
2.39
2.38
Wages
25523
24822
24612
23848
24747
25181
24276
Wages of
skilled
31662
31546
30653
30740
27925
31207
30330
Wages of
unskilled
21440
20986
20529
20198
19966
21360
21230
Research and
development
3515
2069
3547
3539
2019
1111
499
Number of
firms
270
183
89
59
26
280
1118
Employees
(1) All MN are firms that produce abroad goods or services; (2) MN that produce abroad mainly goods; (3) Switching firms are firms that switch from
domestic to MN and that are in the survey from 2 years before up to 1 year after the investment abroad; (4) Switching firms producing abroad mainly
goods; (5) “Near-investing abroad” are firms that have thought to invest abroad but that have not invested yet; (6) “Other DF” are those that have neither
invested nor thought to invest abroad.
All nominal variables are in thousands of euros, except of wages that are in euros. Skill intensity is the ratio of non-worker employees to the total
employment (white collar/total employment). Profits range from 1 (strongly positive) to 5 (strongly negative). Research and development are the outlays
in R&D activities. Only manufacturing firms are in the samples. The five largest firms (more than 8000 employees) have been excluded.
21
Table 2
Distribution of switching firms
Year
%
Sector
%
Region
%
1989
1.2
Food, beverages and tobacco
3.5
Piedmont and Aosta Valley
15.3
1990
1.2
Textile, wearing and apparel
7.1
Lombardy
22.4
1991
1.2
Leather products
Liguria
3.5
1992
4.7
Wood and wood products
1.2
Trentino-alto Adige
1.2
1993
1.2
Paper, printing and publishing
3.5
Veneto
7.1
1994
7.1
Chemical products
4.7
Friuli-Venezia Giulia
12.9
1995
9.4
Rubber and plastic products
4.7
Emila-Romagna
10.6
1996
2.4
Non-metallic mineral products
7.1
Tuscany
1997
1.2
Basic metal industries
1998
7.1
1999
10.6
Machinery and equipment
Electrical machinery, accounting
and computing machinery
2000
9.4
Transport equipment
2001
8.2
Other manufacturing industries
2002
8.2
Total
2003
7.1
2004
20.0
Total
100.0
10.6
4.7
Umbria
2.4
21.2
Marche
5.9
16.5
Lazio
1.2
5.9
Abruzzo
2.4
9.4
Campania
2.4
Puglia
5.9
Basilicata
1.2
100.0
Sardinia
Total
22
4.7
1.2
100.0
Table 3
Matching # 1: Switching vs. domestic firms
DID estimation results
(robust standard errors clustered by region in brackets)
(1)
(2)
(3)
(4)
All firms
Firms producing abroad mainly
goods
log Employment
0.014
(0.022)
0.009
(0.021)
-0.001
(0.023)
-0.006
(0.021)
log Sales
0.034
(0.042)
0.022
(0.037)
0.037
(0.055)
0.027
(0.052)
log Investment
-0.107
(0.131)
-0.115
(0.130)
-0.196
(0.176)
-0.197
(0.184)
log Skill intensity
-0.002
(0.008)
-0.002
(0.008)
0.005
(0.006)
0.006
(0.007)
log Sales/Employees
0.020
(0.038)
0.014
(0.037)
0.037
(0.044)
0.028
(0.050)
log Exports
0.058
(0.122)
0.034
(0.113)
0.085
(0.232)
0.051
(0.024)
Region, year, pair
and firm dummies
Yes
Yes
Yes
Yes
Firm covariates
no
Yes
no
Yes
Obs.
680
680
472
472
Firms` covariates are: Sales, Exports, Investment and Employment. In each equation only covariates that do not
appear in the outcome variable are included. For example, when the outcome variable is sales/employment we
include only investment and exports.
23
Table 4
Matching # 1: Switching vs. domestic firms - By year DID estimation results
(robust standard errors clustered by region in brackets)
(1)
(2)
(3)
(4)
Firms
producing
mainly
goods
All
abroad
log Employment
log Sales
log Investment
log Skill intensity
log Sales/Employees
log Exports
Region, year, pair
and firm dummies
Firm covariates
Obs.
t
t+1
t
t+1
0.002
(0.017)
0.018
(0.029)
-0.011
(0.018)
-0.001
(0.030)
0.015
(0.061)
0.038
(0.045)
0.028
(0.044)
0.016
(0.039)
-0156
(0.144)
-0.072
(0.165)
-0.243
(0.195)
-0.151
(0.211)
-0.006
(0.009)
0.002
(0.008)
0.000
(0.008)
0.011
(0.006)
0.024
(0.046)
0.005
(0.036)
0.017
(0.060)
0.037
(0.043)
0.040
(0.103
0.029
(0.133)
0.118
(0.224)
-0.014
(0.271)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
680
680
472
472
The firms` covariates are: Sales, Export, Investment and Employment. In each equation only covariates that do not appear
in the outcome variable are included. For example, when the outcome variable is sale/employment we include only
investment and exports.
24
Table 5
Matching # 2: Switching vs. Near investing - DID estimation results
(robust standard errors clustered by region in brackets)
(1)
(2)
(3)
(4)
Firms producing abroad mainly
goods
All firms
log Employment
-0.002
(0.020)
-0.004
(0.016)
0.032
(0.039)
0.020
(0.036)
log Sales
0.015
(0.057)
0.021
(0.043)
0.043
(0.050)
0.019
(0.047)
log Investment
0.012
(0.178)
-0.011
(0.158)
0.007
(0.245)
-0.032
(0.226)
log Skill intensity
-0.006
(0.005)
-0.008
(0.005)
-0.020**
(0.008)
-0.017*
(0.008)
log Sales/Employees
0.018
(0.046)
0.021
(0.041)
0.011
(0.053)
0.007
(0.053)
log Exports
-0.168
(0.233)
-0.184
(0.218)
0.118
(0.188)
0.048
(0.180)
Region, year, pair
and firm dummies
Yes
Yes
Yes
Yes
Firm covariates
No
Yes
no
Yes
Obs.
656
656
440
440
Firms` covariates are: Sales, Exports, Investment and Employment. In each equation only covariates that do not appear
in the outcome variable are included. For example, when the outcome variable is sales/employment we include only
investment and exports.
25
Table 6
Matching # 3: Propensity score sample - DID estimation results
(robust standard errors clustered by region in brackets)
(1)
(2)
(3)
(4)
Firms producing abroad mainly
goods
All firms
log Employment
0.019
(0.029)
0.028
(0.020)
0.037
(0.025)
0.047*
(0.023)
log Sales
-0.047
(0.089)
-0.078
(0.070)
-0.053
(0.135)
-0.140
(0.139)
log Investment
-0.099
(0.140)
-0.083
(0.136)
-0.037
(0.143)
-0.043
(0.157)
log Skill intensity
-0.002
(0.012)
-0.002
(0.011)
-0.004
(0.013)
-0.002
(0.011)
log Sales/Employees
-0.061
(0.068)
-0.058
(0.068)
-0.082
(0.116)
-0.094
(0.128)
log Exports
-0.053
(0.142)
-0.035
(0.132)
0.108
(0.222)
0.167
(0.232)
Region, year, pair
and firm dummies
Yes
Yes
Yes
Yes
Firm covariates
no
Yes
no
Yes
Obs.
616
616
416
416
The firms` covariates are: Sales, Export, Investment and Employment. In each equation only covariates that do not
appear in the outcome variable are included. For example, when the outcome variable is sale/employment we include
only investment and exports. The control group include “near switching” firms and other domestic firms.
26
Table 7
Matching # 4: Recent switching firms - DID estimation results
(robust standard errors clustered by region in brackets)
(1)
(2)
log Employment
-0.006
(0.027)
-0.021
(0.025)
log Sales
0.058
(0.035)
0.047
(0.036)
log Investment
-0.064
(0.166)
-0.115
(0.176)
log Skill intensity
-0.005
(0.010)
-0.006
(0.010)
log Sales/Employees
0.064
(0.041)
0.053
(0.041)
log Exports
0.204
(0.147)
0.159
(0.147)
Region, year, pair
and firm dummies
Yes
Yes
Firm covariates
No
Yes
Obs.
432
432
Firms` covariates are: Sales, Export, Investment and Employment. In each equation only covariates that do not appear
in the outcome variable are included. For example, when the outcome variable is sales/employment we include only
investment and exports. Recent switching firms are those that invested from 1999 onwards.
27
Table 8
Alternative empirical strategy
Dependent variable: Δ log Domestic Employment (2000-2006)
(1)
(2)
(3)
(4)
All MN firms
OLS
(5)
(6)
(7)
(8)
MN producing abroad mainly goods
IV
OLS
IV
Δ log Foreign Empl.
0.018
(0.012)
0.017
(0.010)
0.018
(0.016)
0.038
(0.029)
0.014
(0.017)
0.025
(0.009)
0.028*
(0.015)
0.078***
(0.027)
Δ log Domestic Sales
0.437**
(0.152)
0.436**
(0.164)
-0.037
(0.866)
-0.255
(0.917)
0.403*
(0.207)
0.382*
(0.208)
-0.883
(0.626)
-1.433*
(0.733)
Δ log Domestic Sales2
_
_
0.023
(0.041)
0.032
(0.043)
_
_
0.057
(0.033)
0.081**
(0.034)
Sector fixed effects
No
Yes
Yes
Yes
No
Yes
Yes
Yes
Region fixed effects
No
No
Yes
Yes
No
No
Yes
Yes
R2
0.34
0.41
0.50
0.49
0.26
0.39
0.50
0.44
Obs.
101
101
101
101
68
68
68
68
Δ y is the difference of the variable y between 2006 and 2000. In columns (4) and (8) we report the results of 2SLS estimates taking log
foreign employment2000 as instrument for Δ log foreign employment. Since some firms had zero foreign employment in 2000 a unit
constant has been added to the variables to calculate the logarithm. Robust standard errors in brackets. *, **,***: significant at 10%, 5%,
1% respectively.
28
Table 9
Alternative empirical strategy
Dependent variable: Δ log Domestic Employment (2000-2006)
(1)
(2)
(3)
(4)
All MN firms
(5)
(6)
(7)
(8)
MN producing abroad mainly goods
OLS
IV
OLS
IV
Δ log Foreign Sales
0.005
(0.003)
0.005
(0.004)
0.006
(0.007)
0.013
(0.013)
0.006
(0.007)
0.011
(0.008)
0.014
(0.014)
0.037*
(0.020)
Δ log Domestic Sales
0.435**
(0.148)
0.435**
(0.162)
0.026
(0.853)
-0.128
(0.848)
0.400*
(0.203)
0.381*
(0.208)
-0.892
(0.645)
-1.407**
(0.651)
Δ log Domestic Sales2
_
_
0.021
(0.041)
0.027
(0.040)
_
_
0.057
(0.033)
0.079**
(0.031)
Sector fixed effects
No
Yes
Yes
Yes
No
Yes
Yes
Yes
Regional fixed effects
No
No
Yes
Yes
No
No
Yes
Yes
R2
0.33
0.41
0.50
0.49
0.33
0.38
0.50
0.50
Obs.
101
101
101
101
68
68
68
68
Δ y is the difference of the variable y between 2006 and 2000. In columns (4) and (8) we report the results of 2SLS estimates taking log
foreign sales2000 as instrument for Δ log foreign sales. Since some firms had zero foreign employment in 2000 a unit constant has been added
to the variables to calculate the logarithm. Robust standard errors in brackets. *, **,***: significant at 10%, 5%, 1% respectively.
29
Appendix
Table A1
Matching # 1: Switching vs. domestic firms
Means over the two year before the investment abroad (one year before for growth rates)
(1)
(2)
(3)
(4)
Mean diff.
(st-error)
Mean diff.
(st-error)
Standard.
Difference
Standard.
Difference
All
switching
firms
Controls of
all
switching
firms
Switching
firms
producing
abroad
mainly
goods
Controls of
switching
firms
producing
abroad
mainly
goods
(1)-(2)
(3)-(4)
(1)-(2)
(3)-(4)
615.5
621.4
716.9
596.4
-5.9
(78.1)
120.5
(107.9)
-0.7
12.2
Sales
105567
109350
118263
101307
-3782
(14165)
16956
(17332)
-2.6
10.3
Exports
48787
44311
55836
44899
4476
(6528)
10937
(8550)
6.5
13.7
Exports/sales
0.496
0.475
0.490
0.500
0.021
(0.024)
-0.009
(0.027)
8.4
-4.1
Investment
5917
4309
6995
7545
1608
(977)
-550
(2642)
16.5
-2.6
Sal/employees
193.7
200.0
195.2
190.1
-6.3
(8.3)
5.1
(12.9)
-5.0
4.3
Skill intensity
0.343
0.332
0.296
0.301
0.012
(0.0191)
-0.005
(0.0194)
6.2
-3.5
Investment/employees
9.561
8.980
10.121
8.733
0.581
(1.255)
1.387
(1.293)
4.6
13.7
Start year
1960.7
1961.7
1961.3
1962.7
-0.986
(2.835)
-1.381
(2.727)
-4.2
-6.4
Profits
2.16
1.97
2.22
2.07
0.18
(0.12)
0.14
(0.15)
19.7
14.8
Δ Employment %
1.18
1.29
0.24
1.57
-0.11
(1.27)
-1.33
(1.57)
-1.3
-15.8
Δ Sales %
9.86
11.46
9.66
8.91
-1.60
(4.47)
0.74
(6.89)
-5.1
1.9
Δ Investment %
4.79
29.79
-2.91
25.75
-24.99*
(13.95)
-28.67*
(14.90)
-28.9
-28.8
Δ Exports %
14.40
10.03
14.54
-0.54
4.37
(5.27)
15.09
(9.10)
13.1
32.0
Δ Sales/Employment %
8.67
10.16
9.42
7.34
-1.48
(4.73)
2.08
(6.95)
-4.6
5.4
Δ Skill intensity %
0.83
5.04
-0.00
-1.01
-4.20
(38.18)
1.00
(3.84)
-16.8
5.4
85
68
59
51
Employees
Number of firms
The one-to-one matching is carried out within firms of the same 2-digit industry, minimizing a loss function that has the followings arguments: the percentage
differences of: employment, exports, sales and the growth rate of: sales, employment and investment between switching and non- investing firms, the year just
before the investment abroad. Δ y is the growth rate of y the year before the investment. The means are calculated over the two years before the investment
abroad (for the growth rate the year before).
All nominal variables are in thousands of euros. Skill intensity is the ratio of non-worker employees to the total employment (white collar/total employment).
Profits range from 1 (strongly positive) to 5 (strongly negative). Only manufacturing firms are in the samples.
30
Table A2
Matching # 2: Switching vs. Near-investing domestic firms
Means over the two years before the investment abroad (one year before for growth rates)
(1)
(2)
(3)
(4)
Mean diff.
(st-error)
Mean diff.
(st-error)
Standard.
Difference
Standard.
Difference
All
switching
firms
Controls of
all
switching
firms
Switching
firms
producing
abroad
mainly
goods
Controls of
switching
firms
producing
abroad
mainly
goods
(1)-(2)
(3)-(4)
(1)-(2)
(3)-(4)
Employees
454.7
465.4
486.4
435.4
-10.7
(53.6)
51.03
(47.80)
-2.1
12.5
Sales
85913
86158
91497
79786
-245
(10422)
11710
(11461)
-0.2
12.3
Exports
41897
34684
47283
35997
7212
(5703)
11285*
(6601)
12.8
20.5
Exports/sales
0.487
0.455
0.491
0.537
0.031
(0.022)
-0.046
(0.028)
12.0
-18.0
Investment
4126
3678
4431
3427
448
(485)
1003*
(580.2)
8.4
18.5
Sal/employees
196.8
207.2
205.6
217.4
-10.44
(13.34)
-11.82
(18.39)
-7.7
-7.5
Skill intensity
0.350
0.361
0.301
0.325
-0.011
(0.018)
-0.024
(0.017)
-6.7
-16.5
Investment/employ
ees
9.54
9.74
10.16
8.60
-0.21
(1.61)
1.567
(1.202)
-1.4
16.5
Start year
1961.5
1958.6
1963.3
1957.9
2.86
(3.124)
5.38
(4.34)
10.6
18.5
Profits
2.233
2.049
2.19
2.09
0.173
(0.146)
0.095
(0.162)
15.9
9.2
Δ Employment %
0.94
1.03
0.35
0.70
-0.09
(1.41)
-0.34
(1.59)
-1.0
-4.1
Δ Sales %
8.72
7.92
9.38
4.49
0.79
(2.62)
4.88
(3.45)
4.1
24.7
Δ Investment %
3.87
-9.47
-4.55
-13.29
13.34
(11.04)
8.73
(14.27)
16.4
10.5
Δ Exports %
16.25
18.73
16.69
12.52
-2.48
(8.54)
4.17
(7.74)
-4.6
10.3
Δ
Sales/Employment
%
7.787
6.89
9.02
3.79
0.88
(2.58)
5.23
(3.72)
4.7
27.6
Δ Skill intensity %
1.41
5.05
0.51
-0.46
-3.64
(3.73)
0.96
(4.47)
-15.4
4.1
82
59
55
42
Number of firms
The matching is carried out within firms of the same 2-digit industry, minimizing a loss function that has as arguments: the percentage differences of:
employment, exports, sales and the difference of growth rate of: sales, employment and investment between switching and near-investing firms, the
year before the investment abroad. Δ y is the growth rate of y the year before the investment. The means reported are calculated over the two years
before the investment abroad (for the growth rate the year before).
All nominal variables are in thousands of euros. Skill intensity is the ratio of non-worker employees to the total employment (white collar/total
employment). Profits range from 1 (strongly positive) to 5 (strongly negative). Only manufacturing firms are in the samples.
31
Table A3
Matching # 3: Propensity Score sample
Means over the two years before the investment abroad (one year before for growth rates)
(1)
(2)
(3)
(4)
Mean diff.
(st-error)
Mean diff.
(st-error)
Standard.
Difference
Standard.
Difference
All
switching
firms
Controls of
all
switching
firms
Switching
firms
producing
abroad
mainly
goods
Controls of
switching
firms
producing
abroad
mainly
goods
(1)-(2)
(3)-(4)
(1)-(2)
(3)-(4)
Employees
426.6
378.0
438.8
377.1
48.6
(30.97)
61.7*
(35.1)
11.8
13.6
Sales
76066
66016
76211
68894
10049
(7693)
7317
(9795)
12.6
8.8
Exports
33244
27751
35029
30829
5493
(3087)
4199
(4317)
14.4
9.8
Exports/sales
0.47
0.51
0.47
0.51
-0.033
(0.018)
-0.037*
(0.020)
-12.6
-14.4
Investment
3609
10142
3541
12461
-6532
(4389)
-8920
(6420)
-16.7
-19.0
Sal/employees
194.3
209.4
201.2
242.2
-15.1
(26.7)
-41.03
(3809)
-5.5
-12.6
Skill intensity
0.355
0.361
0.304
0.341
-0.006
(0.017)
-0.036
(0.019)
-3.3
-23.5
Investment/employ
ees
9.26
14.21
9.56
16.91
-4.95
(4.15)
-7.36
(5.98)
-13.4
-16.8
Start year
1962.3
1965.2
1964.6
1965.2
-2.97
(2.70)
-0.55
(2.80)
-12.4
-2.5
Profits
2.231
2.175
2.21
2.24
0.055
(0.146)
-0.026
(0.171)
4.8
-2.3
Δ Employment %
0.89
0.11
0.37
0.02
0.78
(2.06)
0.35
(2.41)
6.2
2.6
Δ Sales %
9.62
8.36
10.52
14.96
1.26
(8.72)
-4.43
(12.49)
2.3
-6.7
Δ Investment %
3.36
4.93
-2.08
13.75
-1.57
(14.05)
-15.84
(16.67)
-1.8
-17.4
Δ Exports %
17.47
7.95
18.58
6.25
9.52*
(4.81)
12.32**
(5.54)
22.8
26.9
Δ
Sales/Employment
%
8.73
8.25
10.15
14.93
0.48
(7.52)
-4.78
(10.75)
1.0
-8.5
Δ Skill intensity %
0.81
-2.03
0.26
-5.51
2.84
(4.51)
5.77
(5.62)
11.3
22.5
77
71
52
51
Number of firms
The one-to-one matching is carried out within firms of the same 2-digit industry. The control group is chosen using the propensity score method with
the technique of the nearest neighbor matching. Δ y is the growth rate of y the year before the investment. The means are calculated over the two years
before the investment abroad (for the growth rate the year before).
All nominal variables are in thousands of euros. Skill intensity is the ratio of non-worker employees to the total employment (white collar/total
employment). Profits range from 1 (strongly positive) to 5 (strongly negative). Only manufacturing firms are in the samples.
32
Table A4
Matching # 4: Recent switching firms
Means over the two years before the investment abroad (one year before for growth rates)
(1)
(2)
Mean diff.
(st-error)
Standard.
Difference
Switching firms
Controls of all
switching firms
(1)-(2)
(1)-(2)
Employees
422.8
588.4
-165.6**
(78.7)
-24.6
Sales
89837
119651
-29814*
(17341)
-21.2
Exports
46048
45956
92
(8985)
0.1
Exp/sales
0.51
0.44
0.069***
(0.030)
27.6
Investment
4687
4854
-166
(790)
-2.3
Sal/employees
224.91
235.70
-10.79
(10.23)
-8.0
Skill intensity
0.347
0.355
-0.008
(0.023)
-4.1
Investment/employees
11.67
10.87
0.79
(1.89)
5.3
Year of foundation
1964.5
1966.6
-2.10
(3.40)
-9.3
Profits
2.23
2.0
0.23
(0.14)
23.6
Δ Employment %
1.79
2.02
-0.22
(1.46)
-2.8
Δ Sales %
10.05
5.58
4.46
(2.74)
26.5
Δ Investment %
5.37
29.53
-24.16
(19.26)
-26.2
Δ Exports %
16.34
7.67
8.67
(7.06)
24.2
Δ Sales/Employment %
8.25
3.56
4.68
(2.97)
28.5
Δ Skill intensity %
-0.05
7.77
-0.078
(0.057)
-26.0
54
47
Number of firms
The one-to-one matching is carried out within firms of the same 2-digit industry, minimizing a loss function that has the followings arguments: the
percentage differences of: employment, exports, sales and the growth rate of: sales, employment and investment between switching and non- investing
firms, the year just before the investment abroad. Δ y is the growth rate of y the year before the investment. The means are calculated over the two
years before the investment abroad (for the growth rate the year before). Recent switching firms are those that invested after 1998.
All nominal variables are in thousands of euros. Skill intensity is the ratio of non-worker employees to the total employment (white collar/total
employment). Profits range from 1 (strongly positive) to 5 (strongly negative). Only manufacturing firms are in the samples.
33
Figure 1
The choice of the switching firms and the control groups
(A )
Did you produce,
or consider the possibility to produce,
goods or services abroad in 2003-04?
(B) Yes:
Did you produce goods or services
abroad in 2004?
(D) Yes:
Multinationals (MN)
(C) No:
Domestic Firms (DF)
(E). No:
“Near-investing” domestic firms
(F) Switching:
Firms switching from domestic to MN
in the interval 1984 and 2004 and that
are continuously in the survey from two
years before to one year after the
internationalization.
(G)
Switching firms producing abroad
manly goods
34
Fig. 2. Matching 1 – All switching vs. other domestic firms
Unconditional means. Index=1 at (t*-1)
Employment
Sale
1.40
1.20
Switching
Controls
1.20
Difference
1.00
1.00
0.80
Switching
Controls
0.80
Difference
0.60
0.60
0.40
0.40
0.20
0.20
0.00
-2
-1
0
1
0.00
-2
-1
0
1
-0.20
Exports
Skill intensity
1.20
1.40
Switching
Controls
1.20
Difference
1.00
Switching
1.00
0.80
Controls
Difference
0.80
0.60
0.60
0.40
0.40
0.20
0.20
0.00
0.00
-2
-2
-1
0
-1
0
1
1
-0.20
-0.20
Investment
Sales/Employment
1.20
1.20
1.00
1.00
0.80
0.80
Switching
Controls
Difference
Switching
0.60
Controls
0.60
Difference
0.40
0.40
0.20
0.20
0.00
0.00
-2
-0.20
-1
0
1
-2
-0.20
35
-1
0
1
Fig. 3. Matching 1 – Switching producing abroad mainly goods vs. other domestic firms
Unconditional means. Index=1 at (t*-1)
Employment
Sales
1.40
1.20
1.20
1.00
Switching
1.00
Controls
0.80
Difference
0.80
Switching
Controls
0.60
Difference
0.60
0.40
0.40
0.20
0.20
0.00
-2
-1
0
0.00
1
-2
-0.20
-1
0
1
-0.20
Exports
Skill intensity
1.20
2.00
Switching
1.00
Controls
Difference
1.50
Switching
0.80
Controls
Difference
1.00
0.60
0.50
0.40
0.00
-2
-1
0
0.20
1
0.00
-0.50
-2
-1
Investment
0
1
Sales/Employment
1.20
1.60
1.40
1.00
1.20
1.00
0.80
0.80
Switching
0.60
0.60
Switching
Controls
Difference
Controls
Difference
0.40
0.40
0.20
0.20
0.00
-2
-1
0
1
-0.20
0.00
-2
-0.40
-0.60
-0.20
36
-1
0
1
Fig. 4. Matching 2 – All switching vs. near-investing domestic firms
Unconditional means. Index=1 at (t*-1)
Employment
Sales
1.40
1.20
Switching
1.20
Controls
1.00
Difference
1.00
Switching
0.80
Controls
Difference
0.80
0.60
0.60
0.40
0.40
0.20
0.20
0.00
-2
-1
0
0.00
1
-2
-0.20
-1
0
1
-0.20
Exports
Skill intensity
1.20
1.40
Switching
1.20
1.00
Controls
Difference
1.00
0.80
Switching
Controls
Difference
0.80
0.60
0.60
0.40
0.40
0.20
0.20
0.00
0.00
-2
-2
-1
0
-1
0
1
1
-0.20
-0.20
Investment
Sales/Employment
1.20
1.40
Switching
1.20
Controls
1.00
Difference
1.00
0.80
Switching
Controls
0.80
Difference
0.60
0.60
0.40
0.40
0.20
0.20
0.00
-2
-1
0
1
0.00
-2
-0.20
-0.40
-0.20
37
-1
0
1
Fig. 5. Matching 2 – Switching producing abroad mainly goods vs. Near-investing domestic firms
Unconditional means. Index=1 at (t*-1)
Employment
Sales
1.40
1.20
Switching
1.20
Controls
1.00
Difference
1.00
Switching
Controls
0.80
Difference
0.80
0.60
0.60
0.40
0.40
0.20
0.20
0.00
-2
0.00
-2
-1
0
-1
0
1
-0.20
1
Exports
Skill intensity
1.40
1.40
Switching
Controls
1.20
Switching
1.20
Difference
Controls
Difference
1.00
1.00
0.80
0.80
0.60
0.60
0.40
0.40
0.20
0.20
0.00
0.00
-2
-1
0
1
-2
-0.20
-1
0
1
-0.20
Investment
Sales/Employment
1.20
1.20
1.00
1.00
0.80
Switching
0.80
Switching
Controls
Controls
Difference
Difference
0.60
0.60
0.40
0.40
0.20
0.20
0.00
0.00
-2
-0.20
-1
0
-2
1
-0.20
38
-1
0
1