Small Bus Econ (2015) 45:595–611
DOI 10.1007/s11187-015-9652-2
Firm size and spillover effects from foreign direct
investment: the case of Romania
Karolien Lenaerts • Bruno Merlevede
Accepted: 23 February 2015 / Published online: 8 March 2015
Ó Springer Science+Business Media New York 2015
Abstract This paper introduces firm size in the
analysis of the productivity spillovers of foreign direct
investment. Our analysis of a panel of Romanian firms
reveals two main findings: only medium-sized foreign
firms generate spillovers, and domestic firms’ size is of
limited importance to identify which firms absorb
spillovers. To explain these findings, we show that
large foreign firms are less embedded in the domestic
economy because they are more likely to bring their
own suppliers, import intermediate inputs and export
their output. Smaller foreign firms lack the scale to
transmit spillovers to domestic firms. Whereas foreign
firms’ size adequately proxies for these spillover
mechanisms, domestic firms’ size has an unclear
relationship with the different mechanisms.
Keywords Foreign direct investment Spillovers Firm size Supply chain
JEL Classifications
F2 D24 L23 L25 L26
1 Introduction
Electronic supplementary material The online version of
this article (doi:10.1007/s11187-015-9652-2) contains supplementary material, which is available to authorized users.
K. Lenaerts (&)
Fund for Scientific Research (FWO-Vlaanderen), Ghent
University, Tweekerkenstraat 2, 9000 Ghent, Belgium
e-mail: [email protected]
K. Lenaerts B. Merlevede
Department of General Economics, Faculty of Economics
and Business Administration, Ghent University,
Tweekerkenstraat 2, 9000 Ghent, Belgium
e-mail: [email protected]
In many countries, encouraging foreign direct investment (FDI) is one of the cornerstones of industrial
policy. Policy-makers are eager to attract multinational enterprises (MNEs), not only because MNEs bring
resources and create jobs, but also because these firms
may transfer technology and knowledge to domestic
firms. These technology transfers are indirect or
spillover effects of FDI, which are expected to
improve the productivity of domestic firms. Since
total factor productivity (TFP) is widely recognized as
a key driver of a country’s macroeconomic growth and
competitiveness, FDI spillover effects are of great
importance.1
B. Merlevede
IWH Halle, Halle, Germany
B. Merlevede
HUBrussel, Brussel, Belgium
1
Foreign presence also has a net positive impact on firm
survival (Burke et al. 2008).
123
596
There is an extensive literature that relates domestic
firms’ productivity to the presence of foreign firms.
The standard approach in this literature is to add
spillover variables as additional inputs explaining TFP
in a production function framework. Intra-industry or
horizontal spillovers emerge between MNEs and
domestic firms in a similar stage of the supply chain.
Inter-industry or vertical spillovers arise between
MNEs and domestic firms upstream or downstream
of the MNE in the supply chain. The literature
distinguishes vertical spillovers that occur between
MNEs and their domestic suppliers (backward spillovers) from those that arise between MNEs and their
downstream clients (forward effects). The first studies
on FDI spillovers focused solely on horizontal
spillovers (Caves 1974). After Javorcik (2004), empirical work on vertical effects has soared, recognizing
vertical linkages as a more likely channel for positive
productivity spillovers. Several literature surveys have
concluded that positive spillovers mainly originate
from backward linkages whereas empirical evidence
on horizontal and forward effects is much more mixed
(Crespo and Fontoura 2007; and Havranek and Irsova
2011, 2013). Researchers therefore increasingly focus
on the identification of the determinants, particularly
at the firm-level, that facilitate positive spillover
effects. Meyer and Sinani (2009) and Havranek and
Irsova (2013), for example, indicate that the sign and
magnitude of (horizontal) spillover effects systematically depend on the characteristics of the domestic
economy and the foreign investor.
This paper considers firm size, both of foreign and
domestic firms, as a determinant of FDI spillovers.
The size of domestic firms has been examined in a few
studies in the spillover literature, for example by
Sinani and Meyer (2004), but generally this is done
only as a robustness test. The role of the size of foreign
firms has—to the best of our knowledge—not yet been
explored. This may result from the implicit assumption that MNEs are large and that small firms do not
have the scale to transmit spillovers. Many studies in
the literature also lack data on smaller foreign firms
because often they rely on datasets where a cutoff is
applied for firms to be included. This is not the case in
our large firm-level dataset for Romania, and we do
find a substantial number of smaller foreign firms. We
model FDI spillovers as a function of both foreign and
domestic firms’ size to study whether firm size serves
as a proxy for the underlying spillover mechanisms
123
K. Lenaerts, B. Merlevede
that explain which foreign firms generate spillovers
and which domestic firms benefit from these effects.
Our analysis reveals two main findings: (1) micro,
small and, more surprisingly, large foreign firms do
not generate spillover effects, only medium-sized
foreign firms do; and (2) domestic firms’ size seems
unimportant in explaining spillovers. We find negative
forward and positive horizontal and backward effects,
the latter being the most robust positive spillovers in
the literature (Havranek and Irsova 2011). We provide
evidence for two mechanisms that may account for the
finding that large foreign firms do not generate
positive backward spillovers. First, we show that large
foreign firms largely import their intermediate inputs
instead of sourcing them locally. Second, we provide
support for the hypothesis that large foreign investors
‘‘bring their own supply chain’’ and therefore do not
generate spillovers along the supply chain. The latter
result also accounts for the negative forward effects
we find: although medium-sized firms source locally,
they likely produce more advanced inputs for MNEs
that are too difficult to handle for domestic firms. The
absence of horizontal spillovers originating from large
foreign firms could result from the fact that these firms
typically are exporters that are less involved in the
domestic market. We show that larger foreign firms in
Romania are indeed more likely to show higher export
intensities.
Our second main result is that domestic firms’ size
is of more limited importance. This can be explained
by the fact that domestic firm size appears to correlate
with different characteristics and mechanisms that
have been identified as determinants of spillovers.
However, these mechanisms may result both in
positive or negative spillovers. Domestic firm size is
therefore an inferior proxy to disentangle the underlying spillover mechanisms. In this regard, we show
that absorptive capability varies within rather than
between size categories. Whereas size is unimportant,
absorptive capacity does affect spillovers to domestic
firms.
The remainder of this paper is organized as follows.
Section 2 presents related literature and shows more
details on the size distribution of foreign firms in our
sample. Section 3 discusses the data and the measurement of spillover variables. Section 4 deals with the
empirical approach. Section 5 presents our findings,
and in Sect. 6, we provide an interpretation of these
results. Finally, Sect. 7 concludes.
Firm size and spillover effects from foreign direct investment
2 Foreign firms, size and FDI spillovers
Although many studies cover FDI spillover effects,
only a few contributions focus on the role of firm size.
Theoretical work on the relation between domestic
firms’ size and spillovers points to several possibilities. Aitken and Harrison (1999) indicate that small
firms are less able to cope with the ‘market stealing’
effect of MNEs, which is confirmed by their finding
that horizontal spillovers are negative only for firms
with less than 50 employees. Sinani and Meyer (2004)
further suggest that large firms are more capable of
exploiting opportunities offered by foreign technology
due to scale effects. Large firms also have better access
to finance to invest in capturing spillovers. Alfaro et al.
(2010) find a positive relation between the development of local financial markets, access to finance and
spillover effects. On the other hand, there also exists a
considerable literature pointing to small- and mediumsized firms as major sources of growth and innovation.
Smaller firms tend to be less bureaucratic and are
likely to seize opportunities that larger firms conducting their own R&D and innovation disregard (Acs and
Audretsch 1990). Acs et al. (1994) show that large
firms are more adept at exploiting knowledge created
within the firm, whereas smaller firms have a
comparative advantage at adopting outside technology. For a sample of Estonian companies, Sinani
and Meyer (2004) detect the largest horizontal
spillovers for small firms while no significant effects
are found for large firms. Domestic firm size may
further determine the likelihood of interaction with
foreign firms, i.e., the rationale for vertical spillover
effects. A rising number of studies point to entrepreneurship capital as an important determinant of
economic growth (Audretsch and Keilbach 2004). In
these studies, entrepreneurship manifests itself mainly
through the arrival of new small firms.
Foreign firms’ size has received, to the best of our
knowledge, no attention in the spillover literature.
This reflects the literature’s implicit assumption that
MNEs are big. When we plot the size distribution of
foreign manufacturing firms in Romania in the left
panel of Fig. 1, we do observe a substantial number of
smaller foreign firms. A comparison of this size
distribution with that of domestic firms, portrayed in
the right panel, reveals that the distribution for foreign
firms does have a fatter tail of large firms. In the
remainder of the paper, we use four size classes
597
following the EU’s classification of firms as micro
(less than 10 employees), small (between 10 and 50
employees), medium (between 50 and 250 employees)
and large firms (more than 250 employees).2 Table 1
presents some statistics for different size classes of
foreign firms in Romania. Foreign firms with more
than 250 employees account for about 8 % of the total
number of foreign firms, but their share in total value
added, turnover, capital and employment (of foreign
firms) is between 60 and 65 %. Medium-sized firms
account for about 22 % of the number of firms and for
another 25 % share of value added, turnover, capital
and employment. Smaller firms are large in numbers,
but account for a very small share in value added,
turnover, capital and employment.
3 Data and spillover measurement
We use a Romanian firm-level panel dataset to analyze
FDI spillover effects from MNEs in manufacturing
and services industries on Romanian manufacturing
firms. Our data span the period 1996–2005, and there
is no restriction on firm size. Data are trimmed for
outliers by removing the top and the bottom percentiles of the annual growth rates of real operating
revenues, labor (L), real capital (K) and real material
inputs (M). Data are drawn from the Amadeus
database issued by Bureau Van Dijk Electronic
Publishing. Amadeus holds information on the ownership and financials on public and private companies
across Europe (Bureau Van Dijk 2011). Multiple
DVDs were used to construct our dataset to get a full
overview of financials and ownership through time.3
Nominal data are deflated with prices at NACE 2-digit
level. Price data were extracted from the Statistical
Yearbook of the Romanian Statistical Office (RSO
2
The EU criteria use employment and turnover thresholds, but
we focus on employment only. The European Commission
defines size classes using the following ceilings: micro firms
have less than 10 employees and a turnover or balance sheet
total of maximum 2 million Euro; small firms have less than 50
employees and a turnover or balance sheet total of maximum 10
million Euro; medium-sized firms have less than 250 employees
and a turnover or balance sheet total of maximum 50/43 million
Euro (http://ec.europa.eu/enterprise/policies/sme/facts-figuresanalysis/sme-definition/index_en.htm).
3
A single issue of the database is a snapshot of the ownership
information and firms that exit are quickly dropped from the
database.
123
598
Domestic Firms
.6
Foreign Firms
0
.2
.4
Fig. 1 Foreign and
domestic firm size
distributions in year 2005
(firms employing more than
250 employees are included
in the 250 employee
category)
K. Lenaerts, B. Merlevede
0
50
100
150
200
250
0
50
100
150
200
250
Number of Employees in 2005
Table 1 Characteristics of foreign firms across size categories (manufacturing industries, year 2005)
Employment size category
# Firms
(%)
Share in value
added (%)
Share in
employment (%)
Share in
capital (%)
Share in
turnover (%)
Micro firms
1–5 employees
28.7
1.3
0.7
1.9
1.1
5–10 employees
Small firms
11.8
1.1
0.9
1.2
1.6
10–20 employees
12.3
2.2
1.8
1.7
2.2
20–50 employees
17.0
6.3
5.6
5.3
7.4
50–100 employees
11.1
8.6
7.9
7.0
8.3
100–250 employees
10.8
17.4
17.0
18.3
15.2
Medium firms
Large firms
250–1000 employees
6.6
37.0
29.8
39.2
37.6
[1000 employees
1.6
26.1
36.2
25.5
26.5
Percentages are expressed as shares in total number of foreign firms in manufacturing, in total foreign value added
2005) and the Industrial Database for Eastern Europe
from the Vienna Institute for International Economic
Studies (WIIW 2007). To construct real output (Y),
operating revenues are deflated with producer price
indices. Labor equals the number of employees. Real
capital is tangible fixed assets deflated by the average
of the following industry deflators: machinery and
equipment (NACE 2-digit 29), office machinery and
computing (30), electrical machinery and apparatus
123
(31), motor vehicles, trailers and semitrailers (34) and
other transport equipment (35). Real material inputs
are obtained by deflating material inputs with an
intermediate input deflator, constructed on the basis of
input–output tables (IO-tables). A time-series of IOtables in a Romanian industry classification (approximately NACE 3-digit) was obtained from the
RSO, which allows us to construct time-varying
input–output coefficients. The subset of Romanian
Firm size and spillover effects from foreign direct investment
599
firms in the Amadeus database is known for its
excellent coverage (see Merlevede et al. 2014).
Summary statistics for the domestic and foreign
firms across size categories are provided in Table 2.
The stylized facts commonly found in the literature
also hold for our data: foreign firms are more
productive and larger in terms of output, capital,
employment and intermediates. In terms of total factor
productivity (TFP), larger firms do not seem necessarily more productive than smaller firms.
To calculate horizontal spillover variables, the
empirical literature applies a definition that dates back
to Caves (1974). The horizontal spillover variable
HRjt captures the degree of foreign presence in
industry j at time t as:
P
i2j Fit Yit
HRjt ¼ P
ð1Þ
i2j Yit
latter prefer to buy inputs from more productive
domestic firms.5 The forward spillover variable FWjt
is defined as:
X
FWjt ¼
djlt HRlt
ð3Þ
where Yit is the output produced by firm i in year t. HRjt
is industry j’s share of output produced by foreign
firms. Foreign firms are identified by Fit which is the
share of foreign participation in firm i in year t. To be
considered as ‘‘foreign,’’ a foreign participation by a
single investor of at least 10 % is required.4 HRjt is
then combined with input–output coefficients obtained
from input–output tables to calculate the vertical
spillover variables. We define the backward spillover
variable BKjt as in Lenaerts and Merlevede (2012):
X
BKjt ¼
cjkt HRkt
ð2Þ
k
cjkt is the proportion of industry j’s output supplied to
sourcing industry k at time t. The cs are calculated
from time-varying IO-tables for intermediate consumption. Backward spillovers capture that domestic
firms supply intermediates to foreign firms. Since
firms cannot easily nor quickly switch industries to
buy inputs, this approach avoids potential endogeneity
by using the share of industry output sold to downstream domestic markets k with some level of foreign
presence HRkt. Employing the share of firm output
sold to MNEs causes endogeneity problems if the
l
where the IO-tables reveal the proportion djlt of
industry j’s inputs purchased from upstream industries
l. HRjt, BKjt and FWjt are related to domestic firms’
TFP to infer the direction, magnitude and significance
of spillovers.
We introduce foreign firm size in Eq. (4) to
decompose the traditional horizontal spillover variable in (1) according to different size categories using
the EU definition above.
P
P
mic
sma
Yit
i2j Fit Yit
i2j Fit
P
P
HRjt ¼
þ
Yit
Yit
P i2j med
P i2j lar
Yit
i2j Fit
i2j Fit Yit
P
þ
þ P
ð4Þ
Y
i2j it
i2j Yit
In (4), e.g., Fmic
it equals the share of foreign participation in firm i in year t conditional on firm i employing
less than 10 employees. We denote the different
components on the right-hand side of (4) as HRmic
jt ,
med
sma
HRsma
and HRlar
jt , HRjt
jt . Then HRjt , for example, is
industry j’s share of year t output that is produced by
small foreign firms. In our analysis, we use these
different components, calculated as in (5), without
restricting their coefficients to be equal. The definitions for BKsize
and FWsize
follow from Eqs. (2) and
jt
jt
(3):
P
size
i2j Fit Yit
size
P
HRjt ¼
ð5Þ
i2j Yit
BKsize
jt ¼
X
cjkt HRsize
kt
ð6Þ
djlt HRsize
lt
ð7Þ
k
FWsize
jt ¼
X
l
Table 3 shows summary statistics for all spillover
variables, and Fig. 2 illustrates the evolution of the
horizontal spillover variables. Table 3 and Fig. 2
indicate that large foreign firms account for the largest
4
We use the definition formulated in the OECD Glossary of
foreign direct investment (http://www.oecd.org/daf/inv/
investment-policy/fdibenchmarkdefinition). In the supplementary materials, we show that our results still hold if we apply a
more stringent definition of at least 50 % foreign participation.
5
In the supplementary materials, we show that results are not
qualitatively affected by applying the slightly different Javorcik
(2004) definition.
123
600
Table 2 Summary
statistics for domestic and
foreign firms across four
size categories
K. Lenaerts, B. Merlevede
Domestic firms
Obs.
Foreign firms
Mean
SD
Obs.
Mean
SD
Outputa
Micro firms
Small firms
Medium firms
Large firms
211,516
225.6
1701.7
19,048
768.4
5004.7
62,858
17,137
1352.5
5192.0
5074.6
12,593.5
15,529
9484
2672.8
9471.2
8592.0
30,561.0
7252
28,126.7
108,733.2
4020
51,170.2
98,067.9
Capitala
Micro firms
226,993
42.6
286.7
21,107
194.5
1082.1
Small firms
64,074
383.7
2611.5
16,071
894.9
6247.1
Medium firms
17,319
2114.0
6471.8
9683
4350.6
18,767.0
7273
14,210.3
37,399.2
4045
22,156.6
49,299.9
Large firms
Intermediatesa
Micro firms
214,315
158.7
1476.5
19,807
511.5
4469.5
Small firms
63,114
936.8
4382.1
15,703
1708.0
6700.7
Medium firms
17,176
3227.0
9939.6
9562
5447.9
22,495.3
7265
16,658.4
91,448.6
4031
30,674.8
67,704.6
Large firms
Employees
a
Thousands of local
currency. Two measures of
firms’ total factor
productivity (TFP) are
considered: a measure
based on the Olley and
Pakes (1996) methodology
(OP) and a measure based
on the Ackerberg et al.
(2008) methodology (ACF).
More details on the OP and
ACF estimation procedures
and our motivation to use
these methodologies are
provided in Sect. 4 below
Micro firms
183,804
3.7
3.1
16,970
4.5
3.6
Small firms
60,917
21.1
15.0
15,023
24.2
17.2
Medium firms
Large firms
16,902
7244
106.2
781.4
78.4
878.3
9311
3992
114.9
870.5
78.2
1279.0
OP TFP
Micro firms
161,966
2.13
1.02
15,420
2.19
1.27
Small firms
59,656
2.11
1.02
14,582
2.27
1.10
Medium firms
16,708
2.31
1.08
9184
2.73
1.19
7214
2.17
0.97
3961
2.42
1.16
Large firms
ACF TFP
Micro firms
114,038
5.79
1.50
10,296
6.03
1.69
Small firms
45,040
5.78
1.38
11,133
6.10
1.37
Medium firms
13,452
5.81
1.23
7452
6.09
1.11
6356
5.89
1.15
3416
6.04
1.24
Large firms
part of the spillover variable. In terms of the time
dimension, we see a clear upward trend of foreign
firms’ share in total industry output in the different size
classes, which is more pronounced for medium and
large firms. The time dimension of the vertical
spillover variables shows a similar evolution (not
reported here).
4 Empirical approach
Our empirical strategy to assess whether foreign and
domestic firms’ size function as determinants of FDI
123
spillover effects involves introducing spillover variables
in a production function framework. We use the ‘‘best
practice’’ approach of Havranek and Irsova (2011)
which entails using firm-level data, accounting for the
endogeneity of input demand in the estimation of TFP,
estimating the regression in first-differences, and including a rigorous set of industry controls. We start from
a Cobb-Douglas production function to obtain an
unbiased TFP estimate and then relate domestic firms’
TFP to horizontal and vertical spillover variables.
The estimation of TFP is complicated by endogeneity of inputs because the input choices of firms are
based on their productivity (Griliches and Mairesse
Firm size and spillover effects from foreign direct investment
601
Table 3 Summary statistics for the horizontal and vertical
spillover variables across four size categories
proxy6 while LP chooses material inputs, arguing that
investment is not a good proxy as it is lumpy and does
not respond smoothly to productivity shocks. Ackerberg et al. (2008) (ACF) developed an alternative
semi-parametric procedure to tackle potential collinearity issues in OP and LP. As the discussion on the
most appropriate estimation method is still ongoing,
we use both OP and ACF TFP. OP TFP is based on
output, and ACF TFP on value added. The correlation
between both measures is 0.65. TFP estimates are
obtained from production functions estimated by
NACE 2-digit manufacturing industry.
In a next step, firm-level TFP of firm i in industry
j at time t is related to lagged spillover variables
(FDIjt-1) and a set of industry controls (Zjt-1) as
shown in Eq. (8). Spillover variables are lagged
because the entry of foreign firms is not expected to
immediately affect domestic firms’ productivity.
TFPijt ¼ ai þ u1 f FDIjt1 þ u2 Zjt1 þ nijt
ð8Þ
Obs.
Mean
SD
Min
Max
Horizontal spillover
All firms
580
0.258
0.193
0.000
0.880
Micro firms
580
0.022
0.078
0.000
0.875
Small firms
580
0.039
0.046
0.000
0.324
Medium firms
580
0.063
0.081
0.000
0.552
Large firms
580
0.134
0.160
0.000
0.862
Backward spillover
All firms
580
0.207
0.078
0.018
0.552
Micro firms
580
0.028
0.022
0.001
0.278
Small firms
580
0.047
0.029
0.003
0.370
Medium firms
580
0.054
0.033
0.003
0.284
Large firms
580
0.078
0.055
0.000
0.441
Forward spillover
All firms
580
0.211
0.086
0.019
0.570
Micro firms
Small firms
580
580
0.019
0.043
0.016
0.020
0.002
0.007
0.222
0.162
Medium firms
580
0.054
0.035
0.003
0.232
Large firms
580
0.094
0.064
0.006
0.409
Equation (8) is first-differenced, and then, region (ar),
industry (aj) and time (at) dummies are added, which
results in Eq. (9). Using first-differences and subsequently adding these dummy variables is introduced
by Haskel et al. (2007). Taking first-differences
eliminates fixed-effects and other time-invariant factors. Region, industry and time dummies are introduced to capture unobserved factors that drive TFP
growth.7 This specification is estimated for firms that
are domestic throughout the sample period.
0
0
DTFPijt ¼ u1 Df FDIjt1 þ u2 Zjt1 þ d1 ageit
þ d2 sizeit1 þ ar þ aj þ at þ eijrt
ð9Þ
Zjt-1 includes a Herfindahl index of industry concentration, an index for downstream demand,8 an index of
import competition and an indicator of export
Fig. 2 Box plots of horizontal spillover variables for four size
categories (NACE 3-digit industries)
1995). This implies that the estimation of CobbDouglas production functions using OLS or fixedeffects results in biased estimates of factor shares and
productivity. Several authors have therefore proposed
alternative estimation methods that yield an unbiased
TFP estimate. The semi-parametric approaches pioneered by Olley and Pakes (1996) (OP) and Levinsohn and Petrin (2003) (LP) introduce a proxy to
handle the endogeneity bias. OP use investment as a
6
We follow Amiti and Konings (2007) to calculate investment
from our data.
7
Introducing region and industry dummies is particularly
important to our context. During the sample period, Romania is
going through a transformation process because the country’s
industrial structure was distorted due to communist preferences.
8
Downstream foreign entry could increase demand for intermediate products and result in scale economies. To separate this
effect, the regression includes demand for intermediates
calcuP
lated following Javorcik (2004) as demandjt ¼ k ajk Ykt with
ajk the IO-matrix coefficient indicating that in order to produce
one unit of good k, ajk units of good j are needed. Ykt stands for
industry k output deflated by an industry-specific deflator.
123
602
intensity. We further include firm age and the lagged
level of firm size (measured by total assets) as
controls. Standard errors (SE) are clustered at the
industry-year level because the spillover variables are
defined at the industry-level while estimation is
performed at the firm-level (Moulton, 1990). Equation (9) is estimated by OLS for domestic manufacturing firms.
5 Results
Given the substantial number of smaller foreign
investors in Romania, we investigate whether foreign
firms of different size categories generate similar
spillovers. To do so, we use definitions (5)–(7) and
introduce the subcomponents of the aggregate spillover variables separately in the estimation of Eq. (9).
Table 4 presents results. We consider spillovers on all
domestic firms (columns 1 and 6) and on four size
categories of domestic firms (columns 2–5 and 7–10)
and use both OP and ACF TFP as dependent variables
(left/right panel).
The main result in Table 4 is that a single foreign
firm size class is the primary driver of spillovers.
Surprisingly, medium-sized foreign firms that employ
between 50 and 250 employees are found to be the
main source of spillover effects (more specifically,
positive horizontal and backward spillovers and
negative forward spillovers). Micro, small and large
foreign firms do not seem to generate spillovers. There
is some heterogeneity in the effects across domestic
firm size categories with micro and small firms
benefiting from positive horizontal spillovers, whereas
larger domestic firms do not. But taking into account
SE, the coefficients are not statistically different from
one another. This raises two issues to address: (1) why
do only medium-sized foreign firms transmit spillovers; and (2) why does domestic firm size have such
a negligible impact on spillovers?
6 Interpretation of results
6.1 Why large and small foreign firms
do not generate spillover effects
Notwithstanding that micro and small foreign firms on
average are more productive than their domestic
123
K. Lenaerts, B. Merlevede
counterparts (see Table 2), it is perhaps less surprising
that they do not generate spillovers. Smaller MNEs
simply lack the scale to transmit spillover effects.
Vacek (2010) confirms that only larger foreign firms
have a sufficiently large scale to generate spillovers.9
The more surprising result in Table 4, however, is that
we find little to no evidence of (positive backward)
spillovers from large foreign firms. We first refute an
obvious candidate rationale to account for this result
and then suggest other mechanisms for which we
provide some empirical evidence.
As the prime rationale for spillover effects is the
technological superiority of foreign firms, one may
argue that foreign firms’ size is related to technological superiority and that medium-sized foreign
firms outperform other foreign firms in this respect.
Figure 3 plots the distribution of the level of OP TFP
for different size categories of foreign firms in 2005.
As is clear from the Figure, the distribution for
medium-sized foreign firms does certainly not stand
out and there is considerable heterogeneity around the
mean. The correlation between log OP (ACF) TFP and
the log number of employees is -0.10 (-0.004) for
the sample of foreign firms. In Table 5, we estimate
the technical superiority of foreign firms for different
size classes using a matching approach.10 In all size
categories, foreign firms outperform local firms for
both TFP measures. However, the TFP premium of
medium-sized foreign firms is not larger than that of
small and large foreign firms. The relative level of
technological superiority across size classes therefore
is not driving our finding that only medium-sized
foreign firms transmit spillover effects.11 These results
are also important from another perspective. In their
model, Helpman et al. (2004) show that only the most
9
We find further support in the BEEPS 2005 questionnaire (cf.
infra). Of the 33 small foreign firms interviewed in Romania, 30
answered yes to the question whether they operate locally, but
only 16 answered yes to the question whether they operate
nationally.
10
Within industry-year-size categories, foreign firms are
matched to a domestic twin firm that shares characteristics with
the foreign firm. The following variables are used to estimate the
propensity score: lagged TFP level, lagged first-differenced
TFP, lagged log of the number of employees and its square,
lagged log of capital per worker and its interaction with age, age
and age squared. Arnold and Javorcik (2009) provide a detailed
analysis of the foreign TFP premium.
11
This does not preclude a role for technological superiority of
foreign firms (Lenaerts and Merlevede, 2014).
0.056
0.086
42,847
[0.554]
-0.110
[0.906]
-3.239***
[1.158]
0.585
[2.163]
-2.044
[0.619]
0.364
[0.787]
2.002**
[1.387]
1.150
[1.693]
0.307
[0.222]
-0.087
[0.488]
1.147**
[0.576]
0.197
[1.207]
0.412
10–50
0.101
11,161
[0.537]
-0.123
[0.905]
-3.237***
[1.009]
0.855
[2.301]
-0.936
[0.518]
0.311
[0.798]
2.691***
[1.125]
1.219
[1.432]
-0.794
[0.210]
-0.053
[0.502]
0.713
[0.459]
0.186
[1.173]
-0.591
50–250
0.137
5234
[0.434]
-0.147
[0.758]
-2.187***
[0.923]
-0.954
[1.969]
0.131
[0.397]
0.290
[0.549]
1.976***
[0.469]
1.176**
[0.999]
-0.601
[0.153]
-0.076
[0.332]
0.620*
[0.463]
0.508
[0.987]
-0.324
[250
0.077
120,763
[1.462]
0.053
[1.897]
-7.095***
[3.055]
2.425
[5.676]
-0.902
[1.676]
1.300
[2.229]
6.916***
[3.361]
2.507
[3.391]
-0.616
[0.573]
-0.053
[1.059]
3.068***
[1.443]
0.752
[3.156]
-0.107
0.086
74,783
[1.650]
0.332
[1.923]
-7.235***
[3.117]
2.149
[6.079]
-0.959
[1.776]
1.509
[2.337]
6.761***
[4.396]
1.695
[3.584]
-0.529
[0.595]
-0.082
[1.076]
3.349***
[1.542]
1.095
[3.323]
0.124
\10
0.108
32,262
[1.417]
-0.466
[1.882]
-7.345***
[3.094]
2.469
[5.473]
1.040
[1.828]
1.276
[2.666]
7.080***
[4.020]
3.998
[3.600]
-0.194
[0.622]
0.218
[1.107]
2.961***
[1.555]
0.261
[3.104]
-0.611
10–50
0.115
9081
[1.044]
-0.245
[1.959]
-6.275***
[2.554]
3.637
[5.179]
-3.067
[1.345]
0.665
[2.703]
8.340***
[2.684]
2.254
[2.959]
-1.822
[0.573]
-0.242
[1.127]
1.724
[1.182]
0.653
[3.021]
-1.723
50–250
0.105
4637
[1.034]
-0.472
[1.619]
-2.472
[2.980]
-2.762
[4.936]
-2.097
[1.158]
0.123
[1.498]
3.780**
[1.092]
2.224**
[2.157]
-0.400
[0.544]
-0.229
[0.874]
1.270
[1.645]
2.205
[2.520]
-0.223
[250
Domestic firms with average number of employees
***, **, * significance at 1, 5, 10 %
Spillovers from four size categories of foreign firms (micro, small, medium and larger) on all domestic firms (columns 1 and 6) and on four size categories of domestic firms
(columns 2–5 and 7–10). Results for OP TFP (columns 1–5) and ACF TFP (columns 6–10). Robust SE in brackets
0.054
R2
107,780
[0.596]
[0.572]
167,022
0.327
[0.974]
0.175
[0.955]
[1.240]
-3.210***
[1.192]
-3.216***
0.588
[2.518]
[2.380]
0.666
-2.054
-1.869
0.700
[0.608]
0.584
[0.586]
1.980**
[0.827]
2.034***
[1.489]
[1.252]
[0.776]
0.255
0.699
0.093
[1.620]
0.064
[0.217]
[0.212]
[1.575]
-0.079
-0.087
1.334***
[0.482]
1.225**
[0.661]
[0.605]
[0.477]
0.547
0.359
0.856
[1.356]
0.598
[1.278]
Obs.
[250
50–250
10–50
\10
FW
[250
50–250
10–50
\10
BK
[250
50–250
10–50
\10
HR
\10
All dom. firms
All dom. firms
Domestic firms with average number of employees
ACF TFP
OP TFP
Table 4 Spillover effects across domestic and foreign firm size categories
Firm size and spillover effects from foreign direct investment
603
123
604
K. Lenaerts, B. Merlevede
Fig. 3 Distributions of OP TFP of four size categories of
foreign firms in 2005
productive firms engage in FDI as only these firms are
able to cover the fixed entry costs associated with
investment. Because the productivity distributions of
foreign firms do not differ across size classes, firms of
all sizes can be expected to enter the Romanian
market. One may argue that the fixed entry costs in
Romania are lower than in larger developed economies, which facilitates the entry of smaller firms (that
cannot enter larger markets). However, following
Helpman et al. (2004), firms of all sizes can enter if
they are sufficiently productive. Further, the firm size
distributions across the foreign investors’ countries of
origin are very similar in Romania (see the supplementary materials). Spillovers thus neither seem to be
driven by firms of a specific country. Our findings for
Romania are partly context-dependent, but likely can
be generalized to other (developing) economies.
A mechanism that may account for our findings is
that—compared to medium-sized foreign firms—
large foreign firms may import (a large share of) their
intermediates, rather than engaging in local sourcing,
and export (a large share of) their output. This rules out
potential upstream and downstream linkages with
domestic firms and potentially reduces the competitive
impact in the domestic market. Whereas our dataset is
ideally suited to establish our two main findings, it is
less suited for a direct firm-level test of this
mechanism. Nevertheless, we are able to present some
industry-level evidence from our data by combining
them with external data. From Eurostat, we retrieve
input–output tables for years 2000 and 2005 that are
less detailed than the IO-tables used above (NACE
2-digit instead of the Romanian NACE 3-digit
equivalent), but do allow calculating the share of
imported intermediates in total intermediate use by
industry. Figure 4 plots the shares of imported intermediates in total intermediate use against the log of the
median number of employees of foreign firms in the
industry. We find a significant positive correlation,
suggesting that industries where larger foreign firms
are active use a higher share of imported intermediate
inputs, leaving less scope for backward spillovers.
Simple OLS regressions for both years separately
show coefficients of 0.07 and 0.09 that are significant
at the 5 % level.
The finding that large foreign firms do not generate
horizontal spillover effects could likewise result from
the fact that these firms are less involved in the
Romanian economy. Large foreign manufacturers
may use Romania as an export platform from where
the source country or other markets are served. To test
this, we retrieve the share of export in total industry
output for the years 2000 and 2005 from the Eurostat
input–output tables. We relate this to the log size of the
median foreign firm in the NACE 2-digit industry.
Figure 5 reveals a positive relationship between
Table 5 Foreign firms’ technological superiority across four firm size categories
Micro firms
OP TFP
Obs.
ACF TFP
Obs.
0.026**
Small firms
0.085**
0.110***
Large firms
0.111***
[0.013]
[0.010]
[0.014]
[0.019]
16,500
18,422
10,770
4154
0.094***
0.251***
0.134***
0.106***
[0.032]
[0.023]
[0.023]
[0.036]
8194
11,174
7356
2892
Results based on matched samples of foreign and domestic firms. SE in brackets
***, **, * significance at 1, 5, 10 %
123
Medium firms
Firm size and spillover effects from foreign direct investment
Total
.4
.6
.8
2005
.2
Share of Imported Intermediates in Total Intermediate Use
2000
0
Fig. 4 Log median number
of employees and the share
of imported intermediate
inputs in total inter-mediate
use (NACE 2-digit)
605
2
3
4
5
6
2
3
4
5
6
2
3
4
5
6
Log Median Number of Employees (NACE 2-digit)
median size and export orientation. Simple OLS
regressions for both years separately show coefficients
of 0.11 and 0.19 that are significant at the 5 and 1 %
level.
The second mechanism we consider is that large
foreign firms do not import their intermediates but
‘‘bring their own supply chain’’; i.e., foreign firms that
are already linked through the supply chain coordinate
their foreign investment before entry and do not build
local linkages. An interesting illustration of this
mechanism is Renault’s decision to buy the Romanian
car manufacturer Dacia in 1999 and the entry of
suppliers as Mittal Steel and Michelin (Javorcik and
Spatareanu 2005). This mechanism likely runs from
large firms to other large or medium-sized foreign
firms (large investors might even force their home
country suppliers to follow their investment).12 Kuo
and Li (2003) also find that one of the main
motivations for Taiwanese SMEs to invest abroad is
‘‘following their major clients.’’ Smaller investors
probably lack the power to force or persuade their
home country suppliers (or clients) to follow their
investment. To test for this mechanism, we link entry
of foreign firms in different size classes (see Merlevede et al. (2014) on how entry is identified in the
dataset) to entry of large MNEs in upstream and
downstream industries (using detailed IO-tables).
Because entry of large foreign firms is a fairly unique
event and the number of industry-year pairs with entry
of more than a single large foreign firm is very small,
we recode entry of large firms as a zero–one event.
Specifically, we define EntryLARGE
as a dummy
kt-x
variable that indicates whether at least a single large
foreign firm entered industry k at time t.13 We then run
probit regressions to test whether entry of foreign
firms of different size categories in a given industry j is
correlated with entry of large foreign firms in supplying or sourcing industries -j, as such testing whether
large foreign firms ‘‘bring their own supply chain.’’
We construct industry-level variables as in (6) and (7)
but replace HRsize
with EntryLARGE
. The dependent
kt
kt-x
variable in the probit regressions is Entrysize
kt , a zero–
one variable indicating whether we observe entry of
foreign firms of a specific size class. We use a zero–
one variable because very likely there is a considerable
12
Although initially Renault set out to continue cooperating
with Dacia’s local suppliers, eleven foreign suppliers of the firm
were ‘asked’ to enter the Romanian market quickly to take over
from the local suppliers which did not meet Renault’s
expectations.
13
Entry of large firms occurs in only 17 % of industry-year
observations; of these 70 % is a single large foreign entrant,
another 15 % is two large foreign entrants, and the remainder
varies between 3 and 7 large foreign entrants.
123
606
K. Lenaerts, B. Merlevede
Fig. 5 Log median number
of employees and the share
of exports in total industry
output (NACE 2-digit)
2005
Total
.5
0
Share of Exports in Total Industry Output
1
2000
2
3
4
5
6
2
3
4
5
6
2
3
4
5
6
Log Median Number of Employees (NACE 2-digit)
heterogeneity in the number of suppliers a foreign firm
brings, e.g., depending on its activity. For identification, we exploit the fact that entry of a large foreign
firm is an isolated event in terms of industry-year pairs
and that in terms of timing supply chain entry should
be swiftly following the large foreign firm’s entry in
order to reduce potential spillovers. We therefore
consider concurrent and last year’s entry in linked
industries. For these reasons, our strategy works for
the effect of large foreign firms’ entry. It is difficult,
however, to set up a similar analysis for potential
supply chain entry of foreign firms of other size classes
because entry of such foreign firms is much more
frequent and nearly always multiple firms enter.
Combined with the heterogeneity in the number of
suppliers an MNE brings, the identification of a ‘‘bring
your own supply chain’’ effect is no longer possible
using a similar strategy.
Results are listed in Table 6 and indicate that the
mechanism described above is at work. The probability to observe entry of large, medium, or ‘‘large or
medium’’14 foreign firms positively correlates with
entry of large firms (this year or last year) in sourcing
industries. These findings are also consistent with the
negative forward-level effect and the positive forward
14
This is a dummy variable that equals one when either a large
or a medium-sized foreign firm enters, or both at the same time.
123
absorptive capability interaction effect found below.
When foreign firms bring their supply chain with
medium and large firms who produce most of their
output for their large foreign client, a domestic firm
will need a sufficient level of absorptive capability to
translate the availability of more advanced inputs into
productivity gains. The entry of micro or small firms
does not seem to be related to large firms’ investment
in related industries.15
We present some firm-level evidence on the basis of
the Business Environment and Enterprise Performance Survey (BEEPS). BEEPS is a questionnaire
on the business environment in transition countries
organized jointly by the European Bank for Reconstruction and Development and the World Bank.16 In
Romania, 45 foreign firms responded to the survey in
2002 and 73 in 2005 (five firms participated in both
waves).17 For these firms, we know whether they
employ between 2 and 49, between 50 and 249, and
between 250 and 9999 employees. In Table 7, we
15
Industry-year pairs with zero entry of micro or small firms are
limited. This caveat should be kept in mind for these results.
16
Gashi et al. (2014) for example use the full BEEPS dataset to
study export behavior of SMEs in transition countries.
17
The 1999 survey is structured differently, and corresponding
questions are hard to find; the 2009 survey is well beyond the
time-period of our dataset.
Firm size and spillover effects from foreign direct investment
607
Table 6 Entry along the supply chain—a simple exploration
Entry of type of foreign firms
In 2005
Large
2.547*
In 2001–2005
Medium
Large or
medium
[1.320]
Concurrent and last year entry of large
firms in supplying industries
[1.106]
[0.892]
[0.838]
[0.805]
[0.806]
Obs.
61
61
61
61
61
Concurrent entry of large firms in
sourcing industries
0.831
[1.471]
1.701
[1.422]
0.982
[1.366]
0.509
[1.344]
1.140
[1.377]
[1.197]
-0.301
0.051
2.074*
0.238
[1.168]
[1.104]
Micro
Concurrent and last year entry of large
firms in sourcing industries
1.113
2.421**
Small
-0.117
-0.005
-0.427
-0.155
[1.117]
0.845
Concurrent entry of large firms in
supplying industries
-0.131
0.042
0.173
[0.379]
[0.319]
[0.306]
[0.300]
[0.294]
Obs.
61
61
61
61
61
Large
1.088**
[0.452]
-0.128
[0.456]
305
0.881**
[0.375]
-0.031
[0.229]
305
Medium
0.849**
[0.408]
-0.258
[0.391]
305
0.560
[0.348]
0.031
[0.195]
305
Large or
medium
0.985**
[0.400]
-0.294
[0.377]
305
0.805**
[0.352]
-0.108
[0.193]
305
Probit (columns 1–5) and random effect probit estimates (columns 6–8)
relate these size categories to different questions that
relate to the mechanisms introduced above. Columns 1
and 2 in Table 7 analyze the answers to the question
‘‘What percentage of your material inputs and
supplies are imported directly?’’. The first mechanism
yields the expectation that large foreign firms are more
likely than other foreign firms to import their
intermediate inputs. This is exactly what we find in
column 1 where, for the subsample of foreign firms,
we regress firms’ intermediate importer status (which
equals one if the firm imports any intermediates and
zero otherwise) on a dummy for medium firms and a
dummy for large firms (small firms being the excluded
category). Large foreign firms are significantly more
likely to import intermediates. This is confirmed in
column 2 where we find that large foreign firms on
average import about a 33 percentage-points larger
share of their imports than their small and mediumsized counterparts.
The question ‘‘What percentage of your sales are
exported directly?’’ allows us to verify whether large
foreign firms produce less for the local market than
other foreign firms. Column 3 in Table 7 confirms that
large foreign firms are more likely to be exporters than
small foreign firms, whereas medium foreign firms are
not. Column 4 suggests that on average, large foreign
firms export a 20 percentage-points larger share of
their output than small foreign firms. Although the
evidence is not decisive, combined with Figs. 4 and 5
above, it does lend support to the idea that larger
foreign firms are less involved in the local market.
The survey does not contain a question to directly
test our claim that large foreign firms bring their own
supply chain. However, based on the answer to the
question ‘‘What percentage of your domestic sales are
to multinationals located in your country?’’, we
produce some further indirect support. If large foreign
firms do bring their own supply chain, foreign firms
should be more likely to supply MNEs than domestic
firms. The BEEPS data reveal that 11 % of domestic
firms supply MNEs, whereas 27 % of foreign firms do.
In column 5 of Table 7, we find that—controlling for
firm size—foreign firms are more likely to supply
other foreign firms, and column 6 suggests that on
average they supply about a 7.5 percentage-points
larger share of output to MNEs than domestic firms.
Combined with our findings on entry of foreign
suppliers following the entry of large foreign firms,
these results further support the claim that the lack of
backward spillover effects from large foreign firms is
(partly) caused by the fact that they bring their own
suppliers.
6.2 Domestic firm size versus absorptive capacity
Our second main finding is that domestic firm size is of
limited importance to understand spillovers. In the
literature, firm size typically serves as a proxy for
123
608
K. Lenaerts, B. Merlevede
Table 7 Firm-level evidence of the different mechanisms using BEEPS data for Romania (covering years 2002 and 2005)
Intermediate
importer
Probit
Medium firms
Large firms
Share imp.
intermediates
OLS
Exporter
Probit
Export
share
OLS
0.408
8.247
0.419
13.663*
[0.275]
[7.617]
[0.268]
[7.150]
1.208***
[0.347]
32.873***
0.933***
[9.327]
[0.326]
19.630**
[8.645]
Foreign firms
Supplier
to MNE
Probit
0.266**
[0.131]
0.682***
[0.159]
0.536***
[0.148]
113a
Obs.
(Pseudo) R
2
v2
Prob. [ v2
0.09
113a
118a
0.10
12.9
0.00
118a
0.03
0.06
8.76
819b
0.05
MNE supply
share
OLS
2.319**
[1.177]
5.654***
[1.609]
7.416***
[1.498]
819b
0.05
34.2
0.03
0.00
Data from BEEPS 2002 and 2005. Firm-level estimates for firms responding to the questions detailed in the text. BEEPS size class
definitions: small firms employ between 2 and 49 employees, medium firms between 50 and 249 employees, and large firms between
250 and 9999 employees
***, **, * significance at 1, 5, 10 %
a
Only foreign firms are used in the estimations reported in columns 1–4
b
Foreign and domestic firms are used in the estimations reported in columns 5–6. SE in brackets
different underlying channels and mechanisms. However, the correlation of these channels is not necessarily positive, and even the interpretation of a single
mechanism does not necessarily point in a single
direction. Here, we demonstrate that firm size is a poor
proxy for domestic firms’ absorptive capacity (technology level) that is often cited as an important
determinant of the direction and magnitude of
spillovers.18
Firms’ technology level can be interpreted in two
opposing ways that result in different outcomes.
Findlay (1978) argues that there is a positive connection between the distance to the world’s technological
frontier and economic growth. In his model, productivity spillovers are an increasing function of the
technology gap between foreign and domestic firms.
Measures of the technology level are often used as a
measure of the ability of firms to assimilate outside
knowledge. Blomström (1986) finds that foreign firms
are more likely to wipe out local competitors if the
initial technology level is low and human capital is
poor (i.e., a low absorptive capability). Kokko et al.
18
The example of Dacia’s former suppliers unable to live up to
Renault’s expectations is also an indication of the limited
absorptive capacity of these firms. This prevents them from
benefiting from supply chain linkages with Renault.
123
(1996) report that horizontal spillovers are positive
and significant only for plants with small or moderate
technology gaps vis-à-vis foreign firms. Therefore,
there is no theoretical ground for a clear interpretation
of the relationship between technology and FDI
spillovers. Findlay (1978) suggests that spillovers are
a negative function of the technology level, while the
absorptive capacity interpretation suggests a positive
relation.
A measure of absorptive capability needs to reflect
the relative technical capabilities of a domestic firm
vis-à-vis the foreign firms in the same industry, either
to compete with them or to use (produce) similar
inputs (output). ACit is defined in Eq. (10) as the
distance between firm i’s lagged TFP level (tfpit-1)
and the lagged ‘‘foreign frontier’’ in its industry j. The
latter is defined as the average TFP level of foreign
firms between the 75th and 99th percentile of the TFP
distribution of foreign firms in industry j (tfp_avjt1,FOR).
ACit ¼
tfpit1
tfp avjt1;FOR
ð10Þ
Figure 6 illustrates how the absorptive capability
distribution (according to the OP TFP definition)
varies across domestic firm size categories. Following
(10), a value of one indicates that the domestic firm is
Firm size and spillover effects from foreign direct investment
609
as productive as the average foreign firm between the
75th and 99th percentile of the TFP distribution of
foreign firms in the same industry. The distributions
are similar across firm size classes, and the bulk of
domestic firms are less than half as productive as the
foreign frontier. Figure 6 clearly suggests that firm
size only makes a poor proxy for a domestic firm’s
ability to benefit from spillovers (also see Sinani and
Meyer 2004).
We integrate absorptive capability in our analysis
by considering a simple interaction effect of AC with
the respective spillover variables. The level effects of
the spillover variables for the OP and ACF TFP
definitions of absorptive capacity in Table 8 are in line
with those reported in Table 4. The level effect of AC
is negative, which implies a lower productivity growth
for domestic firms with a higher AC. Girma (2005)
obtains a similar finding. This negative effect also
corresponds with the reduced scope effect described
by Findlay (1978). The interaction effects confirm that
absorptive capacity potentially has a role to play, as a
non-negligible number of them are significant. It is
difficult, however, to draw clear-cut conclusions from
these results as some effects are negative and some are
positive. In some cases, there are also differences
between the OP and ACF TFP definitions. This could
be due to multicollinearity or to the fact that the impact
of AC is nonlinear (cf. Girma 2005, who uses
threshold analysis to properly identify the effects). A
more detailed analysis of this relationship is beyond
the scope of this paper, however.
For the purpose of this paper, we conclude that
domestic firm size is of minor importance because it
does not function well as a proxy for underlying
mechanisms such as the role of absorptive capacity.
That is why other identification strategies are required
Table 8 Absorptive capability as determinant factor (OP and
ACF TFP)
OP TFP
Level
AC
ACF TFP
AC-inter
-0.018***
Level
AC-inter
-0.345***
[0.004]
[0.019]
HR
\10
10–50
50–250
0.770
0.600
[0.361]
0.582
-0.422**
0.904
[0.610]
[0.190]
[1.467]
0.983**
[0.484]
[250
-0.503
[1.283]
-0.023
0.430***
[0.161]
-0.138
[0.211]
[0.092]
0.385
-0.727**
[1.582]
[0.351]
[3.221]
-2.708
[1.710]
-0.363
[0.920]
3.635*** -1.712***
[1.155]
0.071
[0.572]
[0.636]
-0.379
[0.385]
BK
\10
10–50
0.112
[1.232]
50–250
[250
2.328***
1.135***
[0.399]
0.025
-1.059
[3.572]
1.149
[3.374]
1.531
[1.229]
4.676**
[1.878]
6.388*** -0.240
[0.783]
[0.142]
[2.219]
0.568
-0.461*
1.399
[0.813]
1.821*
[0.587]
[0.238]
[1.653]
[1.015]
FW
\10
-2.568
[2.406]
10–50
1.658**
[0.758]
[5.799]
1.931
[3.025]
0.649
0.136
1.497
2.055
[1.194]
[0.285]
[3.197]
[1.318]
50–250 -3.020*** -0.411*
[0.946]
[0.213]
[250
-1.850
-7.355***
[2.034]
0.780
[1.001]
0.384
-0.330*
0.036
0.125
[0.587]
[0.196]
[1.519]
[0.953]
Obs.
167,022
120,763
R2
0.073
0.101
Results for all domestic firms
Fig. 6 Distribution of absorptive capability of domestic firms
in different size classes in 2005
Spillovers from four size categories of foreign firms (micro,
small, medium and larger) on all domestic firms. Results for
OP TFP (columns 1–2) and ACF TFP (columns 3–4). Robust
SE in brackets
***, **, * significance at 1, 5, 10 %
123
610
to uncover these spillover mechanisms and to determine which local firms are able to benefit from
positive productivity spillovers. In this regard, Lenaerts and Merlevede (2014) present a more thorough
analysis of the effect of foreign and domestic firms’
technology level.
K. Lenaerts, B. Merlevede
supported by complimentary evidence based on the
BEEPS dataset. Small foreign firms do not generate
spillovers, which is likely to be caused by scale effects.
References
7 Conclusions
The literature on FDI spillover effects has neglected the
size of foreign firms as a potential determinant factor of
spillover effects. The main reason is that foreign firms
implicitly are assumed to be large and small firms are
assumed to play no role in the mechanisms behind these
spillovers. This paper explicitly analyzes the role of
foreign and domestic firms’ size for a sample of
Romanian manufacturing firms. We first show that a
large number of small foreign investors are present in
Romania: 70 % (40 %) of foreign firms employ less
than 50 (10) employees. Though large in numbers,
small foreign firms only account for a small share in
total foreign value added (10.9 %) and turnover
(12.3 %). Our spillover analysis reveals two main
findings: (1) small and large foreign firms do not
generate spillovers to domestic firms, only mediumsized foreign firms do; and (2) the size of domestic firms
is of limited importance in explaining spillover effects.
Domestic firm size is of limited importance because
it serves as a proxy for different underlying spillover
mechanisms and interpretations that imply different
spillover effects. These mechanisms should be identified through more in-depth analysis. As an illustration, we show that domestic firms’ absorptive
capability varies within rather than between firm size
categories.
The size of foreign firms, on the other hand, is an
important determinant factor of spillover effects. We
provide evidence that foreign firm size is a good proxy
for the intensity of foreign firms’ linkages with
domestic firms. We show that—rather than engaging
in local sourcing—larger foreign firms import (a large
share of) their intermediates and export (a large share
of) their output. This rules out potential upstream and
downstream linkages with domestic firms and reduces
the horizontal (within-industry) effect in the domestic
market. We further show that large foreign firms do
not necessarily import their intermediates but ‘‘bring
their own supply chain.’’ These mechanisms are also
123
Ackerberg, D., Caves, K., & Frazer, G. (2008). Structural
Identification of Production Functions. mimeo.
Acs, Z., & Audretsch, D. (1990). Innovation and Small Firms.
Cambridge, MA: MIT Press.
Acs, Z., Audretsch, D., & Feldman, M. (1994). R&D spillovers
and recipient firm size. The Review of Economics and
Statistics, 76(2), 336–340. doi:10.2307/2109888.
Aitken, B. J., & Harrison, A. E. (1999). Do domestic firms
benefit from direct foreign investment? Evidence from
Venezuela. American Economic Review, 89(3), 605–618.
doi:10.1257/aer.89.3.605.
Alfaro, L., Chanda, A., Kalemli-Ozcan, S., & Sayek, S. (2010).
Does foreign direct investment promote growth? Exploring
the role of financial markets on linkages. Journal of
Development Economics, 91(2), 242–256. doi:10.1016/j.
jdeveco.2009.09.004.
Amiti, M., & Konings, J. (2007). Trade liberalization, intermediate inputs, and productivity: Evidence from Indonesia. American Economic Review, 97(5), 1611–1638.
doi:10.1257/aer.97.5.1611.
Arnold, J., & Javorcik, B. S. (2009). Gifted kids or pushy parents? Foreign direct investment and plant productivity in
Indonesia. Journal of International Economics, 79(1),
42–53. doi:10.1016/j.jinteco.2009.05.004.
Audretsch, D., & Keilbach, M. (2004). Entrepreneurship capital
and economic performance. Regional Studies, 38(8),
949–959. doi:10.1080/0034340042000280956.
Blomström, M. (1986). Foreign investment and productive efficiency: The case of Mexico. Journal of Industrial Economics, 35(1), 97–112. doi:10.2307/2098609.
Bureau Van Dijk (2011). Amadeus database. http://www.
bvdinfo.com/Products/Company-Information/International/
Amadeus.aspx
Burke, A., Görg, H., & Hanley, A. (2008). The impact of foreign
direct investment on new firm survival in the UK: Evidence
for static versus dynamic industries. Small Business Economics, 31(4), 395–407. doi:10.1007/s11187-007-9065-y.
Caves, R. (1974). Multinational firms, competition and productivity in host-country markets. Economica, 41(162),
176–193. doi:10.2307/2553765.
Crespo, N., & Fontoura, M. (2007). Determinant factors of FDI
spillovers—What do we really know? World Development,
35(3), 410–425. doi:10.1016/j.worlddev.2006.04.001.
Findlay, R. (1978). Some aspects of technology transfer and
direct foreign investment. American Economic Review,
68(2), 275–279.
Gashi, P., Hashi, I., & Pugh, G. (2014). Export behaviour of
SMEs in transition countries. Small Business Economics,
42(2), 407–435. doi:10.1007/s11187-013-9487-7.
Girma, S. (2005). Absorptive capacity and productivity spillovers from FDI: A threshold regression analysis. Oxford
Firm size and spillover effects from foreign direct investment
611
Bulletin of Economics and Statistics, 67(3), 281–306.
doi:10.1111/j.1468-0084.2005.00120.x.
Griliches, Z., & Mairesse, J. (1995). Production functions: The
search for identification. Harvard Institute of Economic
Research working papers 1719, Harvard - Institute of
Economic Research.
Haskel, J., Pereira, S., & Slaughter, M. (2007). Does inward
foreign direct investment boost the productivity of domestic firms? Review of Economics and Statistics, 89(3),
482–496. doi:10.1162/rest.89.3.482.
Havranek, T., & Irsova, Z. (2011). Estimating vertical spillovers
from FDI: Why results vary and what the true effect is.
Journal of International Economics, 85(2), 234–244.
doi:10.1016/j.jinteco.2011.07.004.
Havranek, T., & Irsova, Z. (2013). Determinants of horizontal
spillovers from FDI: Evidence from a large meta-analysis.
World Development, 42(C), 1–15. doi:10.1016/j.worlddev.
2012.07.001.
Helpman, E., Melitz, M. J., & Yeaple, S. R. (2004). Exports versus
FDI with heterogenous firms. American Economic Review,
94(1), 300–316. doi:10.1257/000282804322970814.
Javorcik, B. (2004). Does foreign direct investment increase the
productivity of domestic firms? In search of spillovers
through backward linkages. American Economic Review,
94(3), 605–627. doi:10.1257/0002828041464605.
Javorcik, B., & Spatareanu, M. (2005). Disentangling FDI
spillover effects: What do firm perceptions tell us? In T.
H. Moran, E. M. Graham & M. Blomström (Eds.), Does
foreign direct investment promote development? US: Institute for International Economics. http://bookstore.piie.
com/book-store/3810.html
Kokko, A., Tansini, R., & Zejan, M. (1996). Local technological
capability and productivity spillovers from FDI in the Uruguayan manufacturing sector. Journal of Development
Studies, 32(4), 602–611. doi:10.1080/00220389608422430.
Kuo, H., & Li, Y. (2003). A dynamic decision model of SMEs’
FDI. Small Business Economics, 20(3), 219–231. doi:10.
1023/A:1022819206036.
Lenaerts, K., & Merlevede, B. (2012). Supply chain fragmentation and spillovers from foreign direct investment. Ghent
University working paper: No. 12/822.
Lenaerts, K., & Merlevede, B. (2014). FDI spillovers and
multinational firm heterogeneity. Ghent University working paper: No. 14/879.
Levinsohn, J., & Petrin, A. (2003). Estimating production
functions using inputs to control for unobservables. Review
of Economic Studies, 70(2), 317–341. doi:10.1111/1467937X.00246.
Merlevede, B., Schoors, K., & Spatareanu, M. (2014). FDI
spillovers and time since foreign entry. World Development, 56, 108–126. doi:10.1016/j.worlddev.2013.10.022.
Meyer, K. E., & Sinani, E. (2009). When and where does foreign
direct investment generate positive spillovers? Journal of
International Business Studies, 40(7), 1075–1094. doi:10.
1057/jibs.2008.111.
Moulton, B. (1990). An illustration of a pitfall in estimating the
effects of aggregate variables on micro units. The Review of
Economics and Statistics, 72(2), 334–338. doi:10.2307/
2109724.
Olley, S., & Pakes, A. (1996). Dynamics of productivity in the
telecommunications equipment industry. Econometrica,
64(6), 1263–1297. doi:10.2307/2171831.
RSO. (2005). Statistical Yearbook. Bucharest: Romanian Statistical Office.
Sinani, E., & Meyer, K. E. (2004). Spillovers of technology
transfer from FDI: The case of Estonia. Journal of Comparative Economics, 32(3), 1–22. doi:10.1016/j.jce.2004.
03.002.
Vacek, P. (2010). Panel data evidence on productivity spillovers
from foreign direct investment: Firm-level measures of
backward and forward linkages. IES working paper
19/2010. IES FSV. Charles University.
WIIW. (2007). Industrial Database Eastern Europe. Vienna:
Vienna Institute for International Economic Studies.
123