Multinational companies and productivity spillovers: is there a specification error?

*
Davide Castellani
ISE-Universit di Urbino
Antonello Zanfei
ISE-Universit di Urbino

Version # 1.2
February 2002

Abstract
Recent empirical works on the within-sector impact of inward investments on domestic firms
productivity have found rather robust evidence of no (or even negative) effects. This paper argues
that a specification error might characterise those studies. In fact, measuring foreign presence as the
share of foreign to total sectoral activity (e.g. foreign employment share) in a sector, existing
literature implicitly assumes that changes in the same proportion of both foreign and aggregate
activities within a sector have no effect on the productivity of local firms. It is shown that this is a
restrictive assumption which is very likely to cause a downward bias on the estimate of the inward
FDI externality, thus increasing the probability of finding negative or non-significant effects. We
propose a more general specification, which includes the commonly adopted measure of foreign
presence as a special case. Using data on Italian manufacturing firms, we find positive externalities
only once we allow for the more flexible specification.

JEL: F23, L23

Keywords: Multinational firms, Productivity spillovers, Omitted variable bias

The authors are grateful to Roberto Simonetti for supplying part of the data utilised in this work, and to Elisabetta Andreani and
Elvio Ciccardini for research assistantship. We also wish to thank Nigel Driffield, Grazia Ietto-Gillies, Jack Lucchetti and Lucia
Piscitello, for helpful discussions. Usual disclaimers apply. Financial support from the EU project on Assessing the Impact of
Technological Innovation and Globalization: the effects on growth and employment (AITEG) is gratefully acknowledged.
Correspondence: ISE Universit di Urbino, Via Saffi, 42 - 61029 Urbino - Italy. E-mail: dcastell@guest.net and
zanfei@econ.uniurb.it

Introduction
A recent article by Gorg and Strobl (2001) reviews 21 studies addressing productivity
spillovers from multinational presence in host countries and shows that one-third of the examined
sample, corresponding to all the studies using firm-year panel data, reported negative or nonsignificant spillover effects. In this paper we claim that such evidence should be recast in the light
of a specification bias which might have increased the likelihood of zero or negative spillovers.
Productivity of domestic firms may be affected by foreign direct investments in different
ways. On the one hand, MNEs can be expected to positively impact on local productivity by training
workers and managers who may move or spin off from foreign owned firms and become available
to domestic enterprises (Fosfuri et al. 2001); by demonstrating the feasibility of new technology,
providing technical assistance, transferring patented knowledge, and generating opportunities for
imitation of technological, organisational and managerial practices (Mansfield and Romeo 1980,
Dunning 1993); by creating demand for local inputs, increasing the specialisation and efficiency of
upstream and downstream activities and generating positive externalities for local industries
(Hirschman 1958; Rodiguez-Clare, 1996; Markusen and Venables, 1999); and by exerting
competitive pressures to improve the static and dynamic efficiency of domestic firms (Caves 1974,
Cantwell 1989). On the other hand, foreign presence may negatively affect productivity of local
firms, particularly in the short run, to the extent that MNEs can monopolise markets and draw
demand from domestic firms, causing them to cut production and reduce their efficiency (Aitken
and Harrison 1999). Multinationals can also substitute local suppliers with foreign ones, disrupting
existing linkages (Lall 1978).
While the empirical literature usually avoids to tackle directly the difficult question of how,
and through which channels, spillovers take place, most contributions focus on the simpler question
of whether FDIs are associated with changes in productivity of domestic firms. The main approach
is to regress labour or total factor productivity of domestic firms on a measure of foreign presence
(usually calculated at the sectoral level) and other controls. A common feature of most studies is the
measurement of the foreign presence variable as the share of foreign to total activity (in a given
sector). Foreign and total activity are usually proxied by employment, assets or output. A positive
and significant coefficient on the foreign presence variable is taken as evidence of overall positive
spillover effects from multinational to domestic firms. Gorg and Strobl (2001) conclude that crosssectional studies tend to find positive coefficient on the sectoral FDI variable. However, they note,
following Aitken and Harrison (1999), that those studies might yield upward biased estimates of
FDI spillovers. In fact, if foreign multinationals gravitate towards more productive sectors there
may be a positive association between sectoral productivity and the presence of foreign firms even
without spillovers taking place. In other words, if there exists time invariant effects across the
individual units (either industries or firms) that are not captured in the explanatory variables but are
positively correlated with the foreign presence variable then the cross-sectional studies may produce
upward biased estimates of the spillover effect. Such time invariant effects may, however, be
purged from panel data studies using fixed effect models or by taking first-differences. Among the
cross-sectional studies one needs to mention two works which have tried to work this problem out.
Blomstrom and Sjoholm (1999) use industry dummies to control for such fixed effects in a sample
of Indonesian establishments and still find positive spillovers; while Driffield (2001) exploiting the
availability of two years of data for each sector, estimate a model in growth rates, which might help
overcome the problem of correlation between foreign presence and the error term in levels, and
does not find any evidence of spillovers.
On the contrary, most of the recent panel data studies find evidence of negative or nonsignificant spillover effects. Such studies have addressed countries as different as the Ivory Coast
(Harrison, 1996), Venezuela (Aitken and Harrison, 1999), Czech Republic (Djankov and Hoekman,
2000), India (Kathuria, 2000), Morocco (Haddad and Harrison, 1993), UK (Girma et al., 1999),
2

Spain (Barrios, 2000), Portugal (Flores et al., 2000). Among the works reviewed in Gorg and Strobl
(2001) only Liu et al. (2000) find positive FDI externalities, once controlled for fixed effects, in the
UK manufacturing industries.
In this paper we argue that the specification adopted in most of the exisiting literature is likely
to produce a downward bias on the spillover coefficient. In fact, when modelling within-sector
spillovers from multinational activity as a function of the foreign presence ratio (i.e. the ratio of
foreign to total sectoral activity), one implicitly assumes that an increase in the same proportion of
activities of foreign firms and of the sectoral aggregate (leaving the ratio unchanged) should cause
no effect on domestic firms productivity. In other words, as we show in subsequent sections, the
elasticities of domestic firms productivity to foreign and total activity are restricted to be equal in
magnitude but with inverted signs. When this restriction does not hold, as in the case of Italian
manufacturing firms illustrated in this paper, the spillover coefficient may be downward biased:
allowing for a more flexible specification of externalities yields larger (positive and significant)
spillover effects (even controlling for fixed effects). Indeed these results seem consistent with other
evidence provided by a few other studies using absolute measures of foreign presence (instead of
the usual share of foreign to total activities). Barrel and Pain (1997; 1999) and Hubert and Pain
(2000) estimate a CES production function where levels of aggregate and sectoral FDI are
introduced as determinants of technical progess, finding positive effects of sectoral inward FDI on
technical progress in the UK and other EU countries. Furthermore, preliminary evidence on Irish
firm-level data seems to confirm that the foreign presence ratio increases the likelihood of nonsignificant spillovers (Ugur and Ruane, 2001).
The paper is organized as follows. Section 2 sketches the empirical model and derives the bias
which might be associated with measuring spillovers from multinational firms using the foreign
activity ratio. Section 3 describes our data, variables and estimation strategy. Section 4 discusses the
results and section 5 concludes.

2.

An empirical model of externalities from multinational activities in host countries

A rather established economic tradition models a single firm production function introducing
aggregate activity as an externality, which is taken as exogenous in firms maximising decision and
increases firms total factor productivity1. We extend this formulation by adding to the usual
externality term T, denoting aggregate activity in sector j, core business of firm i, a second source of
productivity gain stemming from activities of foreign multinationals in sector j, F.
(1)

Yit = Bit K it Lit M ti

Bit = ei +t + it Fjt1T jt 2
i = 1, Nd (domestic firms)
j = 1, J (sectors)
t = 1, T (time)
Notation is as usual: Y is real output, L is the number of employees, K is the stock of capital
and M is the use of raw materials and energy. Firms TFP (B) is modelled simply as a function of
the two externality parameters, a fixed effect and an error term. We maintain a very simple structure
of the determinants of firms TFP to avoid complexities in the derivation of the bias below. We
1

For analytical simplicity we chose a Cobb-Douglas specification for the production function. However, as it will be shown shortly,
the empirical implementation we use can be derived from a logaritmic differentiation of a generic production function (among others
see Caballero and Lyons 1991, 1992; Basu and Fernald, 1996).

admit that economic applications should be able to control for other important factors affecting
firms TFP, such as firms age, R&D and innovation activities, as well as other time varying
firm/sector characteristics.
Taking logs, TFP term becomes
log Bit = c + 1 log( F ) jt + 2 log(T ) jt + i + t + it

(2)

= c + 1 log( F / T ) jt + ( 1 + 2 ) log(T ) jt + i + t + it

As anticipated in the previous section most existing studies looking for FDI externalities
estimate variants of equation (2) where the externality term reduces to the F/T ratio2. The seemingly
innocent implication would be that an increase of the same proportion of both T and F (leaving F/T
unchanged) should not cause any effect on domestic firms productivity. However, as shown in
equation (2), this would require that we impose the important restriction that 1 + 2 = 0 . Here it is
argued that this is not such an innocent restriction, since it imposes that an increase in F and T in the
same proportion will have an impact on domestic productivity that is equal in magnitude but
opposite in direction.In other words, the restriction implies that either a positive spillover generated
by foreign activities is exactly counterbalanced by a negative spillover of total activities; or,
symmetrically, that a positive spillover generated by total activity is exactly counterbalanced by a
negative spillover of foreign activity. Neither of these circumstances needs to occur as a rule (even
though it might be the case under specific conditions). In fact, the former statement contradicts most
of existing theoretical work in the economic growth literature (e.g. Romer, 1986 and Arrow, 1962)
and recent empirical findings (Caballero and Lyons, 1991; 1992; Oulton, 1996) which would rather
suggest that increases in total activities determine positive productivity effects. The latter statement
(that positive spillovers from total activity are counterbalanced by negative spillovers of foreign
activity) implies that a negative impact of inward investment is assumed ex ante, thus predetermining the direction of the spillover effects that are estimated.

Indeed, one can easily notice that imposing 1 +2 = 0 will most likely cause a downward
biased estimate of 1. First differencing wipes out fixed effects from equation (2) and yields
log Bit = 1 log( F / T ) jt + ( 1 + 2 ) log(T ) jt + t + it

(3)

= + 1 xit + z it + it
where we simplified notation by setting xit = log( F / T ) jt and z it = log(T ) jt , = 1 + 2 .
To save notation the subscript j, indicating variables varying across sectors, have been substituted
by i. It remains that F and T have the same values for all the i belonging to sector j.
Whenever 1 +2 = 0 is imposed, i.e. zit is omitted from the regression, equation (3) can be rewritten as:
log Bit = + 1 xit + uit

(4)

Where uit = zit + it is new the error term. Equation (4) is a simplified version of what is
estimated in most of the literature using only the F/T ratio as a measure of foreign presence3. From
textbook econometrics we obtain (Greene, 1997 p.401-403)4:

Studies reviewed by Gorg and Strobl (2001) differ in the type of data used (sector vs. firm-level, panel vs. cross-section), in the
vector of control variables, and in the proxies used for aggregate and sectoral activity (employment, capital, output). However there
seems to be a general agreement on the use F/T ratio as a measure of externality.
3
As we noted above, such a simple specification is used for illustrative purposes and is required to keep tractable the analytical
derivation of the bias below (Greene, 1997 p. 402).
4
We thank Jack Lucchetti for an illuminating discussion on this point.

Cov( x, z )
Cov[( log F / T ), ( log T )]
E (1 ) = 1 +
= 1 +
(1 + 2 ) .
Var ( x )
Var[( log F / T )]
To the extent that the restriction imposed in the literature estimating only the F/T ratio is
satisfied (i.e. 1 + 2 = 0 ) no bias is produced. Otherwise, since Var ( x) > 0 , the direction of the
bias is determined by two terms: (i) the sum of the unrestricted coefficients of externalities from
foreign and aggregate sectoral activity and (ii) the covariance between log F / T and log T .
Therefore, if 1 + 2 > 0 (which is for instance the case when a positive externality from activities
of foreign affiliate occurs), the restriction imposed in the literature is likely to produce a downward
biased externality coefficient, when log F / T and log T are negatively correlated. As illustrated
below, in the case of firms active in Italy, this condition is met and FDI externalities are markedly
higher when the restriction is not imposed.
3.

Data and estimation strategy

In this section we shall briefly describe the sample and the estimation strategy which will be
used to test for the existence of the bias identified analytically above, in the case of Italian
manufacturing firms. The sample is drawn from Elios (European Linkages and International
Ownership Structure), a data-set constructed at ISE-University of Urbino, from the intersection of
two commercially available databases, Amadeus and Who Owns Whom5. From the former source
most of the economic and financial data used for our analysis were gathered, while from the latter
data on the ownership structure (domestic vs. foreign) of each firm were drawn. The overall sample
contains 980 firms located in Italy. A chi-squared test rejects the hypothesis that the sectoral
distribution of these firms is significantly different from the distribution of the population of firms
with more than 50 employees, as registered by the National Institute for Statistics (ISTAT). For
every firm we were able to identify the ultimate parent company, and with this information we have
distinguished foreign-owned firms (when the ultimate parent company is different from Italy) from
domestic firms. Overall we have 761 domestic firms and 219 affiliates of foreign multinationals.
Economic and financial data were available for a 6-year time span, from 1992 to 1997. Firms for
which the complete series of data was not available were preliminarily dropped, thus the sample
available for estimation is a balanced panel of 5,880 observations (of which 4,566 refer to domestic
firms).
The empirical strategy follows two steps. First, we estimate a production function without
externalities, obtain consistent estimates of the production function parameters, , and , and get
an estimate of log(B) as the difference between actual and predicted level of output. In other words,
we specify a production function of the following form:
(5)
log Yit = log Kit + log Lit + log M it + i + t + it

where Y, K, L and M are respectively output, fixed capital, employment and material use as
above6, i is a time-invariant unobserved effect, which captures cross firm differences in
productivity, and t is a time trend.
Estimation of production functions is a rather debated issues in applied econometrics7. The
key problem is that input use is very likely to be correlated with the error term. We overcome this

Amadeus and Who Owns Whom (D&B Linkages) are products of Bureau Van Dijck and Dun & Bradstreet respectively.
Real values of Y, K and M are obtained by deflating respectively nominal turnover, book value of fixed assets net of depreciation,
and costs of materials. The deflator used is the OECD-STAN implied sectoral value added deflator.
7
See Griliches and Mairess (1996) for a recent survey and Aitken and Harrison (1999), Blundell and Bond (1999), Levinshon and
Petril (2000), Olley and Pakes (1996) for some recent applications and developments.
6

problems by taking first differences, which wipe out fixed effects, and we regress the growth rate of
output on the growth rate of inputs, on a constant (resulting from differencing a time trend) 8
(6)
log Yit = log K it + log Lit + log M it + t + it
This should capture most of the correlation between input use and productivity. We also
control for endogeneity of input growth by performing a GMM-IV estimation, using 2-years lagged
values of inputs and time dummies as instruments, in the spirit of Anderson and Hsiao (1987)9. In
the Appendix, Table A.1, we provide results from the estimation of production functions using firstdifferenced OLS and GMM-IV. The former yields significantly lower parameter estimates possibly
due to the mentioned endogeneity problems. Once we obtained consistent estimates of production
function parameters by GMM-IV we computed the estimated residual from the following
expression:

log B it = log Yit ( IV log K it + IV log Lit + IV log M it )

(7)

The second step is to use this consistent estimate of log B it as a dependent variable in OLS
regressions of equation (3). Alternative measures of T and F were obtained summing up
employment (measured by the number of workers), capital (measured by fixed assets) and output
(measured by value added) of all firms in sector j, or only in foreign firms, respectively.
We are aware that the two-step strategy illustrated above might lead to a biased estimate of ,
and , and hence of logB, due to the omission of relevant variables (namely logF and logT) in the
production function. Estimation of the production function including the externality coefficients
should solve this problem. Results using this approach can be found in Appendix. Results do not
change significantly. In particular, both the production function and the externality parameters are
largely unchanged when F and T are introduced, supporting the idea that TFP estimated from the
first-step should not be any different from the estimate in one go. We prefer the two-step strategy as
it is perfectly consistent with the analytical procedure in the previous section.
4.

Results

Table 1 looks at productivity spillovers from multinational firms to host country firms,
comparing the different specifications derived in section 2 for a sample of Italian manufacturing
firms. In particular, it tests whether the measure of foreign presence used in most of the existent
literature is indeed likely to produce estimates of productivity spillovers biased towards zero. We
use measures of foreign presence based on employment, fixed assets and value added at the 3-digit
SIC aggregation level. For each of the three measures we estimate OLS regression of equation (3).
In column (1), (4) and (7) we omit the last term, log T , thus providing estimates similar to the
ones reviewed in Gorg and Strobl (2001), i.e. obtained by using relative measures of foreign
presence, hence implicitly assuming 1 + 2 = 0 . Results are rather different using employment,
capital or output, confirming a result also highlighted in Gorg and Strobl (2001), but in two out of
three cases, the spillover coefficient turns out negative or non-significant. The specification test
proposed in section 2 can be performed adding log T and testing whether its coefficient ( 1 + 2 )
is significantly different from zero. Columns (2), (5) and (8) show that the restriction is soundly
rejected with all the measures of foreign and total activity. Furthermore, since the correlation
8

Notice that this specification for the production function parallels the one used by other authors (Basu and Fernald, 1996; Caballero
and Lyons, 1991; 1992; Oulton, 1996) who derived it from a log-differentiation of a generic production function
Y = f ((K , L, M ), (T , F )) .
9 The estimator has been developed in the context of dynamic panel analysis, but it can be easily applied in this context even if we
do not have any lagged dependent variable. The idea is that productivity shocks can determine current input use, but past values of
input use should not be determined by current productivity shocks. In other words, inputs at t-2 can be thought as exogenous and can
be used as instruments.

between log F / T and log T is negative, the coefficient on log( F / T ) is higher in these
specifications, as expected. As a consistency check we estimate equation (2) in first-differences and
test for the hypothesis that 1 + 2 = 0 . Columns (3), (6) and (9) confirm that 1 is larger than in
the first specification and 1 + 2 is significantly different from zero, albeit with differences in
intensity according to the measures of foreign activity adopted 10.
In Table 2 we perform a sensitivity analysis. We allow for persistence in productivity and we
estimate (3) adding a lagged dependent variable and using the dynamic panel data GMM-IV
estimator proposed by Arellano and Bond (1991), which uses all the available orthogonality
conditions. Results do no change much. The autoregressive specification is strongly supported in
the data, with a coefficient on log(TFP ) i ,t 1 between .24 and .34 and always very significant and
Sargan test and serial correlation tests support the specification of the model. Values of 1 turn out
significantly higher when we adopt the specification proposed in this paper: when F is measured as
foreign employment or foreign capital it turns out significantly different from zero, while the
foreign presence ratio is not significant. The hypothesis that 1 + 2 = 0 is rejected in two out of
three cases.
5.

Conclusions

In recent years most studies on the impact of foreign firms on domestic firms productivity
obtained negative or non-significant results. In particular, after controlling for unobserved fixed
factors affecting TFP using firm-level panel data, the foreign presence ratio turns out at most as
non-significant. This paper provides an explanation for those results, relying on the idea that the
correct identification of spillovers from within-sector multinational activity should allow different
estimates for externalities coming from foreign firms and externalities from aggregate activity We
derived the estimator of such externalities and we found that under mild conditions (in particular
negative correlation between the growth rates of the foreign presence ratio and of total sectoral
activity) modelling multinational presence as the share of foreign to total activities could produce
estimates of externalities biased towards zero. Using data on manufacturing firms active in Italy, we
strongly reject the hypothesis that spillover parameters are equal and opposite in sign in the case of
foreign and of total activities. We also show that allowing the different parameters to vary, positive
externalities are found in the case of Italy. These results are robust to the introduction of a lagged
dependent variable, allowing for persistence in domestic firms productivity. In the light of
analytical and empirical results shown in this paper, one should reconsider the evidence on negative
or non-significant spillovers from multinational firms on host countries. Of course, we are not
saying that, if the specification we propose is used, positive spillovers will necessarily show up. In
fact, location-specific (as well as industry specific) differences remain, and need to be explained: in
some host economies (or sectors), competition and market stealing effects outweigh positive effects
through demostration, linkage effects, labour mobility and technology transfer, while the opposite
can occur elsewhere. In a companion paper we explore the effects of multinational presence on
productivity of domestic firms in France, Italy and Spain, and confirm that national specificities
matter (Castellani and Zanfei 2002). However, further analysis is required to understand the
underlying factors enabling productivity spillovers and the channels through which multinational
firms can foster productivity growth of domestic firms.

10
Explaining the different magnitude and significance of spillover coefficients according to the various measures of foreign presence
is beyond the scope of this paper and would make grounds for future research. One may venture suggesting that the highest and most
significant coefficients we find when foreign presence is measured in terms of fixed capital assets may signal that spillovers of
inward investments in Italy are most likely to be associated with transfer of knowledge embodied in machinery and other capital
goods.

Table 1 The impact of sectoral foreign presence on domestic firms productivity in Italy,
1992-1997 (OLS estimates)
Dependent Variable: Log(TFP)
Sample: Only Domestic Firms
Estimation method OLS OLS
Measure of F and T** L
L
1
2
.02
.03
Log(F/T)*jt
(1.16) (1.37)
Log(F)jt
Log(T)jt
Constant
N. obs
N. firms
F-test H0: 1+2=0
Corr( log F / T , log T )

OLS
L
3

.04**
(2.44)
.17** .13**
(7.59) (4.84)
.001 -.002
-.002
(-.80) (-1.12) (-1.03)
3805 3805
3805
761
761
761
66.31**
-.01

OLS OLS
K
K
4
5
.06** .07**
(4.76) (5.68)
.

-.002
(-.87)
3805
761

OLS
K
6

OLS
OLS
VA
VA
7
8
-.06**
-.02
(-2.31) (-1.06)

.07**
(6.88)
05**
.007
(4.13)
(.51)
-.001
-.002 -.002
(-.67) (-.90) (-.94)
3805
3805 3805
761
761
761
40.99**
-.19

OLS
VA
9

-.01
(-.84)
.28**
.29**
(15.62) (12.79)
-.001
-.001
(-.71)
(-.68)
3805
3805
761
761
213.11**
-.09

t statistics are in brackets below estimates. Asterisks indicates significance values (**: p < 0.05; *: p
< 0.1). Standard error are computed using the Huber/White sandwich estimator.
*to avoid log of zeros in sectors were no foreign firm was registered, F/T have been multiplied by
100 and added one.
** Alternative measures of foreign and total sectoral activity:
L: Employment as a measure of foreign and total sectoral activity
K: Capital (fixed assets) as a measure of foreign and total sectoral activity
VA: Output (value added) as a measure of foreign and total sectoral activity

Table 2 Persistence and the impact of sectoral foreign presence on domestic firms
productivity in Italy, 1992-1997 (GMM-AB)
Dependent Variable: Log(TFP)
Sample: Only Domestic Firms
Estimation method GMM
AB
Measure of F and T** L
.34**
Log(TFP)t-1
(3.50)
.03
Log(F/T)*
(.14)
Log(F)
Log(T)
Constant
N. obs
N. firms
Sargan Test
A-B I
A-B II
F-test H0: 1+2=0

GMM GMM GMM GMM GMM GMM GMM GMM

AB AB
AB
AB
AB
AB
AB
AB
L
L
K
K
K
VA
VA
VA
.28** .29** .31** .32** .30** .34** .24** .24**
(2.85) (3.17) (3.50) (3.69) (3.37) (3.78) (2.74) (2.73)
.16
.12
.10*
.0007
.07
(1.11)
(1.40) (1.79)
(.01) (1.19)
.15**
.11**
.06
(2.47)
(2.20)
(1.57)
.26
.03
.08*
-.01
.25** .18**
(1.32) (.23)
(1.87) (-.27)
(2.95) (2.62)
.04** -.007 -.0009 .006 .05** .02** .02** -005
.007
(15.26) (-.66) (-.12) (3.27) (6.22) (6.60) (11.32) (1.02) (1.21)
3044 3044 3044 3044 3044 3044 3044 3044 3044
761
761
761
761
761
761
761
761
761
18.31 19.89 20.81 25.27 33.83 32.72 25.92 37.06 30.79
-7.83** -8.09** -7.49** -7.64** -7.63** -7.50** -7.29** -7.76** -7.40**
.50
.73
.26
.26
.52
.68
.48
.56
.47
1.97
5.40**
12.95**

GMM-AB denotes estimates obtained using the dynamic panel data estimator (DPD) proposed by
Arellano and Bond (1991)
z-statistics are in brackets below estimates. Asterisks indicates significance values (**: p < 0.05; *:
p < 0.1). Standard error are computed using the robust covariance matrix.
A-B I and II indicates tests for first and second order serial correlation in the first differenced
residuals. Sargan test is distributed as a chi-squared under the null of no overidentification
*to avoid log of zeros in sectors were no foreign firm was registered, F/T have been multiplied by
100 and added one.
** Alternative measures of foreign and total sectoral activity:

L: Employment as a measure of foreign and total sectoral activity

K: Capital (fixed assets) as a measure of foreign and total sectoral activity
VA: Output (value added) as a measure of foreign and total sectoral activity

Appendix Estimates of the production function of domestic firms in Italy, 1992-1997

Dependent Variable: Log(Real Output)
Sample: Only Domestic Firms
Estimation method OLS
Measure of F and T**
log(F)it
log(T)it
log(M)it
Log(K) it
Log(L) it
t2
Constant
N. obs
N. firms
R-squared
Hansen J 2(2)

.37**
(9.53)
.03**
(4.69)
.37**
(13.06)
-.005**
(-4.11)
.03**
(3.05)
3805
761
.60

OLS

OLS

OLS

L
.06**
(3.35)
.23**
(5.49)
.36**
(9.36)
.03**
(4.27)
.29**
(12.03)
-.01**
(-6.36)
.07**
(5.42)
3805
761
.62

K
.06**
(5.60)
.05**
(3.42)
.37**
(9.47)
.03**
(3.59)
.36**
(12.94)
-.005**
(-4.08)
.03**
(3.10)
3805
761
.61

VA
-.004
(-.21)
.33**
(12.08)
.34**
(9.33)
.02**
(3.23)
.30**
(12.26)
-.005**
(-4.62)
.03**
(3.80)
3805
761
.66

GMM GMM
-IV*
-IV*
L
.06**
(1.93)
-.05
(-.78)
.40** .43**
(11.28) (13.51)
.16** .16**
(2.68) (2.90)
.22** .29**
(2.02) (2.74)
-.02** -.02**
(-5.83) (-6.25)
.24** .20**
(-5.29) (5.83)
2283 2283
761
761
1.76

4.59

GMM
-IV*
K
.04**
(3.73)
-.06**
(-2.28)
.42**
(12.84)
.17**
(2.74)
.31**
(3.12)
-.02**
(-6.31)
.21**
(5.49)
2283
761

GMM
-IV*
VA
-.01
(-.21)
.18**
(6.11)
.45**
(14.68)
.12**
(2.30)
.26**
(2.75)
-.01**
(-3.70)
.12**
(3.27)
2283
761

5.89

2.81

t (OLS) and z (GMM-IV) statistics are in brackets below estimates. Asterisks indicates significance values (**: p < 0.05; *: p < 0.1).
Standard errors are robust to heteroschedasticity. OLS utilises the Huber/White sandwich estimator, while GMM-IV is robust to
heteroschedasticity of unknown form. Hansen J is the test of overidentifying restriction.
2
** Instruments used are: t , t, log(M)i,t-2, log(K)i,t-2, log(L)i,t-2

10

References
Aitken B., Harrison A. (1999) "Do Domestic Firms Benefit from Direct Foreign Investment? Evidence from
Venezuela", American Economic Review, 89, 3, 605-18.
Arellano M., Bond S. (1991) Some tests of specification of panel data: Monte Carlo evidence and an
application to employment equations, Review of Economic Studies, 58, 277-297.
Arrow K. (1962) The Economic Implications of Learning by Doing, Review of Economic Studies, June,
pp. 155-173.
Barrel R., Pain N. (1997) Foreign Direct Investment, Technological Change, and Economic Growth within
Europe Economic Journal, 107, November, 1770-1786
Barrel R., Pain N. (1999) Domestic institutions, agglomerations and foreign direct investment in Europe,
European Economic Review, 43, pp. 925-934.
Barrios S. (2000) "Foreign Direct Investment and Productivity Spillovers. Evidence from the Spainsh
Experience", FEDEA Documento de Trabajo, 2000-19.
Basu S., Fernald J. (1995) Are apparent productive spillovers a figment of specification error, Journal of
Monetary Economics, 36, 165-188.
Blomstrom M., Kokko A. (1998) "Multinational Corporation and Spillovers" Journal of Economic Surveys
n.12, 247-277
Blomstrom M., Sjoholm F. (1999) "Technology Transfer and Spillovers: Does Local Participation with
Multinationals Matter?", European Economic Review, 43, 915-23.
Blundell R., Bond S. (1999) GMM estimation with persistent panel data: an application to production
functions, IFS Working Papers W99/04.
Blundell R., Bond S. (1999) GMM estimation with persistent panel data: an application to production
functions, IFS Working Papers W99/04.
Caballero R., Lyons R. (1991) Internal versus external economies in European industry, European
Economic Review, 34, 805-830.
Caballero R., Lyons R. (1992). External Effects in U.S. Procyclical Productivity. Journal of Monetery
Economics, 29, 209-226.
Cantwell J. (1989) Technological Innovation and Multinational Corporation, Basil Blackwell, Oxford
Castellani D., Zanfei A. (2002) Technology Gaps, Absorptive Capacity and the Impact of Inward
Investments on Productivity of European Firms, mimeo, Universit di Urbino.
Caves R. (1974) "Multinational Firms, Competition and Productivity Convergence in Host-Country
Markets", Economica, May, 176-193.
Djankov S., Hoekman B. (2000): "Foreign Investment and Productivity Growth in Czech Enterprises".
World Bank Economic Review, Vol. 14, pp. 49-64.
Driffield N. (2001) "The impact on domestic productivity of inward investment in the UK", The Manchester
School, 69, 1, January 2001, 103-119.
Dunning (1993) Multinational Enterprises and the Global Economy, Addison Wesley, Wokingham,
England.
Flores R., Fontoura M., Santos R. (2001) Foreign direct investment spillovers: what can we learn from
Portuguese data?, mimeo.
Fosfuri A., Motta M., Ronde T. (2001) Foreign direct investment and spillovers through workers
mobility, Journal of International Economics, 53, 205-222.

11

Girma S.; Greenaway D., Wakelin K. (1999): "Wages, Productivity and Foreign Ownership in UK
Manufacturing". Centre for Research on Globalisation and Labour Markets Research Paper 99/14,
University of Nottingham.
Gorg H., Gorg E. (2001) Multinational companies and productivity spillovers: a meta-analysis, Yhe
Economic Journal, 11, November, F723-F739.
Haddad M., Harrison A. (1993) "Are There Positive Spillovers from Foreign Direct Investment? Evidence
from Panel Data for Morocco", Journal of Development Economics, 42, 51-74.
Harrison A. (1996) "Determinants and Effects of Direct Foreign Investment in Cote d'Ivoire, Morocco, and
Venezuela", in Roberts M., and Tybout J., editors, Industrial Evolution in Developing Countries,
Oxford University Press, 1996.
Hirschman A.O. (1958) The Strategy of Economic Development, Yale University, New Haven.
Huibert F., Pain N. (2000) Inward investment and technical progress in the UK manufaturing sector, OECD
Economics Department Working Papers 268.
Kathuria V. (2000): "Productivity Spillovers from Technology Transfer to Indian Manufacturing Firms".
Journal of International Development, 12, pp. 343-369.
Lall S. (1978) Transnational, Domestic Enterprises and Industrial Structure in Host LDCs. A Survey
Oxford Economic Papers , vol. 30, n.2, (217-248)
Levinsohn J, Petril A. (1999) "When Industries Become More Productive, Do Firms? Investigating
productivity dynamics", NBER Working Paper 6893, January.
Liu X., Siler P., Wang C. and Wei Y. (2000): "Productivity Spillovers from Foreign Direct Investment:
Evidence from UK Industry Level Panel Data". Journal of International Business Studies, 31, 3,
pp.407-423.
Mansfield E., Romeo A. (1980) Technology Transfer to Overseas Subsidiaries by US Based Firms
Quarterly Journal of Economics, 95(4) (737-750)
Markusen J., Venables A. (1999) Foreign Direct Investment as a Catalyst for Industrial Development,
European Economic Review, 43, 335-356.
Olley G. , Pakes A. (1996) The Dynamics of Productivity in the Telecommunications Equipment Industry,
Econometrica, pp 1263-98.
Oulton N. (1996) Increasing returns and externalities in UK manufacturing: myth or reality?, The Journal
of Industrial Economics, XLIV, 1, March 1996, 99-113.
Rodriguez-Clare A. (1997) "Multinationals, Linkages and Economic Development", American Economic
Review, 86, 4, 852-73 .
Romer P. (1986) Increasing Returns and Long Run Growth, Journal of Political Economy, October, pp.
1002.1037.
Ugur A. Ruane F. (2001) Foreign Direct Investment And Productivity Spillovers In The Irish Manufacturing
Industry: Evidence From Firm Level Panel Data, Paper presented at the Network of Industrial
Economists Christmas Conference, Birmingham, 14/12/2001.

12