Content uploaded by Ralf Martin
Author content
All content in this area was uploaded by Ralf Martin
Content may be subject to copyright.
Multinationals, foreign ownership and US
productivity leadership: Evidence from the
UK ∗
Chiara Criscuolo†Ralf Martin‡
April 22, 2003
Abstract
Several studies using firm level data find that foreign-owned firms
are more productive than domestic ones. This could reflect a foreign
advantage or an omitted variable bias: foreign firms are by defini-
tion multinational enterprises (MNEs), and MNEs are typically more
productive than non-MNEs. This paper attempts to discriminate be-
tween these hypotheses. We are the first to study the productivity
of foreign owned firms relative to UK firms separated into MNEs and
∗We would like to thank ONS for making the data available to us and Prabhat Vaze
for advice. Thanks to Jonathan Haskel. We have benefitted immeasurably from his
advice and support. For extremely helpful comments we thank Richard Disney, Steve
Nickell, Nick Oulton, Steve Redding and John Van Reenen. This research has been pro-
duced under contract with the ONS. CeRiBA (Centre for Research into Business Activity,
www.ceriba.org.uk) is funded by the Evidence-Based Policy Fund from the Treasury, the
DTI and the ONS. Errors and opinions are those of the authors alone.
†University College London (UCL), and Centre for Research into Business Activity
(CeRiBA); c.criscuolo@ucl.ac.uk
‡Centre for Economic Performance(CEP), London School of Economics(LSE) and Cen-
tre for Research into Business Activity (CeRiBA); r.martin@lse.ac.uk
1
non-MNEs. We obtain three main results. First, the foreign produc-
tivity advantage is mostly a multinational advantage: MNEs, foreign
and UK, are more productive than non-MNEs. Second, US owned
firms maintain a productivity advantage with respect to both UK and
other foreign owned firms. Third, examining the longitudinal dimen-
sion of our data we find no evidence that higher MNE productivity
is driven by sharing superior firm specific knowledge among affiliated
plants. Thus, the MNE advantage must lie in an ability to takeover
already productive plants or in setting up above average productivity
plants on green field.
JEL Classification: F230, L600 Keywords: Multinational Firms,
Productivity, Foreign Ownership, US leadership, Double Fixed-Effects
1 Introduction
Several studies using firm level data find that foreign-owned firms are more
productive than domestic ones. Using US data, Doms and Jensen [6] find
that, controlling for capital, age, industry and region, productivity1in foreign
owned plants is on average 11 to 13% higher than domestic plants. Griffith
et al. using UK data [10] find an advantage of 9%.
In the UK this result has been interpreted in the context of a poor aggre-
gate performance relative to other advanced market economies. O’Mahony
and de Boer [19] find that the US, French and German manufacturing as a
whole have 55, 32 and 29% higher labor productivity than UK manufactur-
ing. Commentators2have suggested that the aggregate productivity gap and
the gap between foreign and domestic firms within the UK are driven by the
same factors, namely bad management and inferior technology in UK owned
firms. This differs from earlier explanations for the aggregate productivity
gap such as low skill level of the labour force and poor institutions which
would affect both domestic and foreign firms in a similar way.
1Measured as value added per employee
2see for example Dorgan et al. [7].
2
Does the gap between foreign and domestic firms necessarily lead to such a
conclusion? Figure 1 shows a possible alternative explanation: the compari-
son of foreign owned plants with all domestic plants in a country is potentially
affected by a selection problem. Foreign owned plants are, by definition, part
of multinational firms (MNEs). However, only a small fraction of domestic
plants are part of UK MNEs. If MNEs have an intrinsic productivity ad-
vantage, the superior performance of foreign firms might simply reflect a
multinational advantage. A number of authors3have suggested that MNEs
Figure 1: The populations of firms in a country
3Hymer [15], Dunning [8], Markusen [16]
3
should have an advantage over firms which only operate in one country. The
idea is that a foreign firm will always have higher costs in setting up in
business compared to a domestic one. These additional costs might arise,
for example, from barriers due to language, unawareness of local business
networks, or from assigning workers abroad. If, nevertheless, a firm sets up
abroad it must have some particular characteristic, such as a patent, a trade-
mark or some firm-specific knowledge that allows it to achieve lower costs
of production or higher prices than rival firms and thereby stay profitable
despite higher set-up costs 4. To establish if foreign owned firms have indeed
a superior technology to domestic firms we should therefore provide a fair
comparison and compare the performance of foreign owned firms relative to
domestic MNEs.
The key innovation of the current paper is to do this for the UK. This was
only possible after merging a recently available dataset, the Annual Inquiry
into Foreign Direct Investment (AFDI), to the Annual Respondents Database
(ARD), the UK’s main dataset for productivity research at the microlevel.
We find that MNEs, both UK and foreign owned, are more productive than
non MNEs which suggests that the foreign effect is by large a multinational
effect. Doms and Jensen are - to the best of our knowledge - the only ones
who have done a similar investigation, but for the US. They find that US
MNEs are more productive than foreign owned firms in the US. US non
MNEs are less productive than both, foreign and US MNEs. To compare our
results to Doms and Jensen we control separately for US ownership. This
yields the same ranking: US owned multinationals are the most productive
followed by other foreign and UK MNEs, with domestic non MNEs being the
least productive. Our study therefore confirms but also qualifies Doms and
Jensen’s result, because it suggests that their finding of US leadership reflects
a genuine advantage of US firms and not a home advantage5. In the second
part of the paper we exploit the panel structure of our dataset to examine the
4In his OLI framework Dunning calls this ownership advantage
5i.e. MNEs might be more productive in their home country because they do not have
the additional setup costs mentioned earlier
4
nature of the MNE advantage in more detail. We try to disentangle if firms
are productive because they are multinational - we call this Generic MNE ef-
fect - or because the most productive firms and plants become multinational,
by either investing abroad themselves or by having a higher probability of
being taken over by a multinational firm. We find a small but significant
Generic MNE Effect of about two percent. We then estimate a double fixed
effects model to examine if it is mainly firm or plant specific effects which
drive the multinational effect. We find a large significant positive difference
between plants effects for plants that are part of MNEs and no significant
difference between MNE and non MNE firm effects.
The rest of the paper is organised as follows: in sections 2 and 3 we
describe our dataset. Section 4 shows that the foreign ownership effect is
rather a multinational effect. In section 5 we examine the existence of a
Generic MNE Effect and calculate firm and plant specific effects. We discuss
in depth the double fixed effects technique used for this purpose. The section
also features an illustrative model. Section 6 concludes. The Appendix A
contains more detailed descriptions of the variables in our dataset, Appendix
B details of the Model introduced in Section 5 and Appendix C robustness
checks for the results presented in Section 4.
2 Data Sources
Our dataset contains information from two sources: the Annual Respon-
dents Database (ARD) and the Annual Survey into Foreign Direct Invest-
ment (AFDI). We describe each in turn6.
6More details on the ARD data can be found in Griffith[12], Oulton[20], Disney et
al.[4], and Barnes and Martin[18]
5
2.1 The Annual Respondents Database (ARD)
The ARD is a dataset made available by the Office for National Statistics
(ONS) based on information drawn from the Annual Business Inquiry (ABI)7,
the annual survey of UK businesses. Until 1997 the ARD only included the
production sector. Since 1998 it covers the whole economy. In our study we
use the production sector data only. Response to the ABI is mandatory un-
der the 1947 Statistics of Trade Act. The ABI requires extensive operational
information on inputs and outputs, which we use to estimate productivity.
The most disaggregated unit on the ARD is a production facility at a single
mailing address referred to as local unit. The ONS keeps a register that keeps
track of all local units in the country, which also captures if a local unit is
part of a larger firm or group of firms. This register is drawn from a variety of
sources including historical records, tax returns and other surveys. However,
for at least two reasons the ARD is not actually a census of all local units.
First, businesses are required to report about their activities at the “enter-
prise level”. For the ONS an enterprises are relatively autonomous business
units which are not necessarily different units in a legal sense. Consequently
an enterprise does not necessarily correspond to a firm. Larger firms might
consist of several enterprises. Nor does an enterprise necessarily correspond
to a single plant. As a consequence the observations in our dataset corre-
spond to local units either if a firm consists of a single plant or if any of the
business units of a larger firm consists of one plant only. 80 percent of the
local units in the manufacturing part of the ARD register report at the local
unit level which makes our dataset by large a plant level dataset. Therefore,
to simplify discussion in what follows we will refer to this level as the plant
level and to the observational units as plants in what follows.
The second reason for the ARD not being a census is that smaller reporting
units - or plants as we call them now - do not have to complete the survey
7Before 1998 it was called Annual Census of Production and included the production
sector only
6
every year. Plants with employment below a certain threshold8are sampled
on a random basis. The sampled plants altogether are referred to as the
“selected sample”, while all non-sampled plants constitute the “non-selected
sample”. Each year the selected sample accounts for around 90% of total
U.K. manufacturing employment (Oulton, [20]).
The country of ownership of a foreign owned firm operating in the UK
- and thus the ability to identify foreign MNE plants in the UK - is an
information which is already part of the ARD register9. Whilst this identi-
fies foreign-owned plants, until now it has not been possible to identify UK
MNEs. To do this we use the Annual Inquiry into Foreign Direct Investment
(AFDI) described in the next section.
2.2 The Annual Inquiry into Foreign Direct Invest-
ment(AFDI)
The AFDI is an annual survey to businesses which requests a detailed break-
down of the financial flows between UK firms and their overseas parents or
subsidiaries. The AFDI is thus a survey run at the firm and not at the plant
level as the ARD. The inquiry has an “outward” part that measures foreign
direct investment (FDI) by UK firms abroad and an ‘inward’ part that mea-
sures FDI in the UK by foreign corporations.
To conduct the AFDI, the ONS maintains a register which holds informa-
tion on the country of ownership of each firm and on which UK firms have
foreign subsidiaries or branches 10. This register is designed to capture the
8The threshold was 100 employees in most years but increased to 250 in later years
9The ARD data is supplemented here with information from Dun&Bradstreet global
“Who own’s Whom” database.
10In the following we refer to subsidiaries and branches jointly as affiliates. The ONS
distinguishes between subsidiaries and branches as follows: a ‘subsidiary’ is mainly a
company where the parent company holds more than 50% of the equity share capital;
a ‘branch’ is a permanent plant as defined for UK corporation tax and double taxation
relief purposes; companies where the investing company holds between 10% and 50% of
the equity share capital, i.e. does not have a controlling interest but participates in the
7
universe of firms that are involved in foreign direct investment abroad and
in the UK11. It is drawn from (and continuously updated) using a variety of
sources including administrative records, (from HM Customs and Excise and
from Inland Revenue), Dun and Bradstreet’s ‘Worldbase’ system and ONS
inquiries on acquisitions and mergers involving UK companies.
2.3 Merging the ARD with the AFDI
The main innovation of this paper is to be able to identify UK MNEs by
merging the AFDI to the ARD.
We merge the two datasets at the firm level, so that all plants in merged firms
are marked as MNEs. We, therefore, classify an ARD plant as being part of
a UK MNE if it is owned by a firm which appears in the AFDI and is not
foreign owned. The merging procedure is subject to two measurement error
problems. First, although, the ONS register tries to include all firms engaged
in FDI, in practice, the register population has varied with the ONS’ success
and effort in identifying such firms 12. Second, to combine the information in
both datasets we have to rely on the ARD’s firm identifier. This variable has
been subject to a major coding change in 1998, which is only incompletely
documented and there appear to be minor inconsistencies and errors also in
other years. In the appendix we document in greater detail our efforts to
clean this variable.
As a consequence of these problems, a number of plants is likely to be
recorded as domestic despite being multinational. Also there may be plants
whose status changes from domestic to multinational although they have al-
ways been multinational. For more details on the AFDI and the merging of
management, are defined ‘associates’. ONS [9] p.120.
11The annual inquiry regards direct investment as an investment made abroad in order
to have an effective voice in the management of a foreign firm. For practical purposes this
is defined, since 1997, as holding a share of at least 10% (20% before 1997) in the foreign
company, whereas holdings below this threshold are considered portfolio investment.
12Particularly after 1997 the AFDI population has increased dramatically after the ONS
started to include information from the Dun&Bradstreet database
8
the two datasets refer to Criscuolo and Martin [3]
3 Descriptive Statistics
Table 1 shows the number of multinational plants that we can identify in
our sample over time. The top panel shows the total number and relative
Table 1: Number of multinationals over time
!
"##$%&&'
'($%)'(
shares of domestic, foreign and UK multinational plants in the complete
ARD population. Row 2 shows a jump of about 25 percent of UK MNEs
from 1997 to 1998. Rather than actual changes in ownership status this most
likely reflects the measurement error problems described earlier. The bottom
panel shows the same numbers for the selected plants; i.e. the plants surveyed
in a given year. Note that the jump in the number of UK MNEs is not as
9
dramatic for this subsample. This reflects the fact that the ONS is more
likely to overlook smaller firms, when building the AFDI register, which also
have a lower probability to be in the ARD selected sample. Consequently the
sample we use for our regression analysis is affected to a lesser extent by the
measurement error problems described in the last section. Also, since MNEs
are on average larger, the relative share of MNEs in the selected sample
is much higher. Whereas in the total population UK and foreign MNEs
combined take a share of a about 4 percent, in the selected sample this same
figure rises to almost 30 percent . The share of UK MNEs remains fairly
constant over time and the share of foreign owned firms has very slightly
increased.
Table 2 shows the shares that the various ownership types represent in
terms of aggregate value added and employment. The top panel shows em-
ployment shares for the whole population based on a combination of the
employment variable kept in the plant register - which is available for the
whole population - and the employment variable obtained from the returned
surveys. The second panel (rows 3 to 6) shows employment shares for the
selected sample only. The remaining 2 panels report value added shares,
first for the selected sample, unweighted, and then weighted to provide an
estimate for the value added shares of the whole population. Here, as in the
remainder of the paper, the weights are calculated on 4-digit industry, 11
region and employment band cells.
Consider first panel one. In terms of employment the importance of MNEs
is much larger than when considering the numbers of plants. In column 6 we
observe that all MNEs account on average for more than 40 percent of total
employment. The reason for this is the larger size of MNE plants.
Looking at the last panel, we see that with more than 50 percent the MNEs
are even more significant in terms of value added. These two pieces of evi-
dence hint at a superior productivity of MNEs. The time series of the shares
of both employment and value added show a slight decrease in the impor-
tance of domestic firms. However, changes are not very dramatic.
10
Table 2: Value added and employment share over time
!"#!$#%
&'(!)*+,*-*+.!(/
!!!(0!!!)
!"#$
!
#
!"#$
!
#
11
Table 3 distinguishes further the MNE group: it classifies foreign owned
plants by the country or world region of ownership13. The table shows that
among the foreign owned plants US MNEs are by far the largest group.
Column 1 reports the number of plants in the whole population and col-
umn 2 the number of plants for each group in the sample of selected plants.
Column 3 and 4 describe the distribution of plants, employment and value
added shares. The table shows that 20 percent of all plants in the UK are
US-owned, almost as much as all other foreign owned firms combined. Sim-
ilar figures hold for the share in employment, 24%, and value added, 28%.
These figures are consistent with the fact that the most productive compa-
nies should also have the highest market share14.
Table 4 reports averages and standard deviations for various variables. Con-
sider employment in row 1. Comparison of column 1 and 2 of panel 1 shows
that foreign owned plants are much larger than all UK plants. When we dis-
tinguish all UK plants between non MNEs (column 3) and MNEs (column
4) we observe that UK MNE plants are on average almost as large as foreign
owned plants and more than double the size of domestic non multinational
plants. Row 2 reports labour productivity - measured as value added per
employee - for the various plant groups. Column 1 and 2 report averages for
all domestic and all foreign plants. Foreign owned plants have an advantage
of more than 50 percent in respect to UK plants. If we distinguish between
UK MNEs (column 4) and UK non MNEs (column 3) we find that UK MNEs
are more similar to foreign owned plants than to UK non MNEs. However,
foreign owned plants (column 2) still have an advantage of more than 20
percent over UK MNEs. In columns (5) and (6), we further distinguish for-
eign owned plants between US owned and non US owned foreign MNEs, and
we observe that US owned plants are the most productive. When looking
at gross output per employee (Panel 3) the foreign advantage becomes more
dramatic: UK MNEs lag behind foreign MNEs by almost 45 percent. Also,
13A detailed description of the country groups that feature in the table can be found in
the data appendix
14Note that it is also a feature of our model below
12
Table 3: Multinational types
!"#
$
%
&
'()*
+ !"#$%&'"()%*+,"
"*-
).'-"&'/"
0$1,"2,",+3#*"%"2-)4&1,"
,-&)"2-,"56*
13
the ranking has now changed: in terms of gross output per employee, foreign
non-US owned plants are the most productive, as shown in columns (5) and
(6). Do these gaps represent the “true” UK disadvantage? Panels 6 and 7 of
table 4 suggest otherwise: foreign owned plants have much higher interme-
diates to labour and capital to labour ratios than UK MNEs, with non US
foreign owned MNEs being the most capital intensive. At least part of the
gap in productivity can therefore be explained by foreign owned plants be-
ing more capital intensive and employing more intermediates. Indeed, panel
6 reports the averages of the logarithm of TFP for UK and foreign owned
plants. Foreign owned plants are still more productive but the difference is
less pronounced. Columns 3 to 6 show that non US foreign MNEs have a
slightly lower average TFP than domestic UK plants. Are these differences
due to industry, location, size or age of the plants? Table 515 addresses this
issue: it reports regression coefficients for UK MNE, US MNE and other for-
eign MNE dummies, which indicate the relative difference to UK domestic
plants. Column 1 reports for each group values without any further controls
which leads to the same qualitative result as table 4. Columns 2 report the
coefficients from regressions that controls for size, age, location of the plant
and industry. The table shows that for most variables the differences among
UK non-MNEs and MNEs found in column 1 are still significant, although
attenuated, and the ranking for the different MNE groups remains virtually
unaffected when controlling for compositional differences. A notable excep-
tion is TFP, however: controlling for compositional differences other foreign
plants turn out to be more productive than domestic non MNEs.
4 Foreign or Multinational Effect
Several studies16 have examined equations of the following type:
yit =δXit +β F ORJ(i,t)+εit (1)
15this table follows Doms and Jensen’s table 7.4
16e.g. Griffith [12] and [11], Harris [13], Doms and Jensen[6]
14
Table 4: Averages for the pooled 1996-2000 sample
!"!#$
!%&#&''#()
!"
#$"
"
"
!%&&&"
'
(
)
15
Table 5: Conditional averages
!! " #!!$
!!%&!!!"!!' (!!)")!
!*+&!)(! !!!!!( !" !&!
+&!)(!(!' +,-+&!)!"./01
!
"#$"%
16
where yis a productivity measure, typically log gross output per worker, xit
is a vector of observed explanatory variables such as the log capital labor
ratio and F ORit is a dummy equal to one if a plant is foreign owned. These
studies find large, positive and significant values for β. There are two ex-
planations for this finding. First, there is a specific domestic productivity
disadvantage: all UK plants are worse than foreign owned plants. Second,
since foreign owned plants are part of a MNE, and MNEs are more produc-
tive than non-MNEs, positive and significant values for βare just reflecting
a multinational advantage. To test between these two hypotheses, we note
that the latter implies that plants belonging to domestic MNEs should have
similar productivity advantages as foreign owned plants. Thus, a high value
for βcould be the result of an omitted variable bias. To examine this, we
include in equation 1 a dummy for MNE:
yit =δXit +β1MNEJ(i,t)+β2F ORJ(i,t)+εit (2)
where MNE takes value one if a plant is part of a MNE, be it domestic or
foreign owned.
In the UK the advantage of foreign owned plants, found in previous studies,
has often been interpreted as evidence of a UK productivity lag in the context
of an aggregate productivity gap compared to other leading economies, in
particular the US. The idea is that the same factors which make the US
economy more productive are also responsible for higher productivity of US
owned plants in the UK. To account for this and to be able to compare
our results directly with the study by Doms and Jensen [6] our preferred
specification of Equation 1 includes a separate identifier for US owned plants:
yit =δXit +β1MNEJ(i,t)+β2F ORJ(i,t)+β3U SAJ(i,t)+εit (3)
Table 6 reports results of estimation of equations 2 and 3 using the pooled
sample for the years 1996 to 2000 and real gross output per employee as de-
pendent variable. In column 1 we only include a foreign ownership dummy
and find a result which other studies have found before: foreign plants en-
joy a strong and significant labour productivity advantage of more than 56.5
17
Table 6: OLS regressions: dependent variable log gross real output
!
! "
#$%
&' !
!"# !"# !"
$%%&%%!&
18
percent17 with respect to the reference group of all UK plants. As discussed
previously, the estimates in column 1 are likely to be affected by an omitted
variable bias. Column 2 shows that once we include a separate dummy for
being part of a MNE, foreign owned plants are 25 percent more productive
than the reference group, which now includes only UK domestic plants that
are not part of an MNE. Column 2 also shows that plants that are part of a
multinational firm are 33.3 percent more productive than non MNE plants.
This result shows that about half of the foreign advantage found in previous
studies is actually a multinational effect. In column 3 we separate foreign
owned MNEs into US owned MNEs and non US owned MNEs. This column
shows that in addition to the 33 percent for being part of a MNE, plants
that are US owned have a 22.2 percent and non US foreign owned plants
a 27.1 percent additional productivity advantage with respect to UK plants
that are not part of MNEs 18.
Column 4 shows estimates of a Cobb Douglas specification of equation 3,
where we control for capital intensity and material usage. The productiv-
ity advantage of MNEs is still significant but lower at 3.5 percent. The US
MNEs are now the productivity leaders, and significantly so, with an addi-
tional advantage of 4.7 percent. The coefficient on foreign non-US MNEs is
now only 0.015 and is not significantly different from zero at the 5 percent
level. Column 5 extends the results of the previous column: it accounts for
age effect by including a quadratic term in age19. Column 6 controls for scale
effects. Finally, column 7 shows that, controlling for both age and scale ef-
17The percentage differences are calculated from the coefficients of the dummy variables
in Table 6 according to the formula diff = eβ−1
18Table 13 in the appendix reports estimates of a specification with real gross value
added as dependent variables. A comparison of column 3 of table 6 and Table 13 shows
that US MNE appear to be the most productive establishments when log of real value
added per employee is the dependent variable. This is can be expained by differences in
the use of material inputs
19Since our age variable is left censored in 1980, we include an age censoring dummy.
We have tried alternative specifications for the age effect, including age categories; the es-
timates do not change significantly from the ones obtained under the current specification.
19
fects MNEs are on average 4.6 percent more productive than UK non MNEs,
US MNEs are the productivity leaders with an additional advantage of 4.7
percent, while the foreign non US advantage is a non significant 1 percent.
The last four columns confirm that US MNEs are significantly more produc-
tive than all other groups of plants and that UK MNEs are as productive as
non US foreign MNEs.
Our results so far suggest the following. First, the foreign labour produc-
tivity advantage estimated in previous studied appears to be by and large
a MNE effect. Second, as shown in table 6, once we control for capital in-
tensity, material usage, scale and age effects, US MNEs appear to be the
productivity leaders, with UK and non-US foreign MNEs having a compa-
rable productivity advantage with respect to UK plants that are not part of
MNEs.
Several issues arise when estimating Equation 1. We address them in turn
and report the results of our robustness checks in Appendix C. The first issue
is whether our results are robust to the choice of the dependent variable: in
Table 6 our dependent variable is log real gross output, deflated using 4-digit
industry producer price indices. We address this issue replicating Table 6 us-
ing value added as dependent variable in Table 13 and in column 5 of Table
11 we use relative TFP as dependent variable.
The second set of problems with equation 1 is the adoption of a suitable
specification for the production function. In the previous section, we have
adopted a static Cobb-Douglas specification, but in the appendix we show
that our results are robust to the adoption of a more flexible production func-
tion, such as the translog production function (column 4 of table 11), and
a dynamic specification to capture adjustment lags in the output following
changes in the factors of productions in column 6 of table 11.
A third issue arises from the sample used. One may want to extend the
results obtained from this sample to the whole population. For this purpose
we run weighted regressions, reported in column 3 of Table 11.
The fourth concern arises from the fact that in our preferred specification,
20
we do not control for workforce skills. Thus, in column 2 of Table 11 we
include the average wage as a proxy for the average skill level in the plant 20.
Fifthly, in column 7 of table 11 we report the results of a random effects
estimation. Under the assumptions of the random effects model, this es-
timator is more efficient than OLS. Finally, the classification of the various
MNEs groups may be debatable. Table 11 in the appendix shows the effect of
variations in the definition of ‘MNE’. In column (4) we consider UK MNEs
only those that have FDI in manufacturing sectors. The rationale behind
this more restrictive definition is to exclude those UK MNEs that only have
export platforms or distributors abroad 21. Also, in column 5 we differentiate
the “other Foreign” group further into various country groups. The results
shown in table 6 seem to be robust: US MNE are the most productive with
UK MNEs and foreign Non US MNEs alternating each other in the second
position. UK plants that are not part of a MNE are the least productive.
Our results, thus, confirm that the foreign effect found in earlier studies is by
and large a multinational effect. Therefore, rather than examining further
why foreign owned firms are more productive we focus now on why MNEs are
more productive. Ideally to answer this question, one would like to have more
structural information, like R&D expenditure, skill mix, innovation activity
or management techniques, which at present is not available in the dataset.
It is part of our research agenda to construct datasets containing variables
of this type. Currently we can get additional insight into the nature of the
MNE advantage exploiting the longitudinal nature of our data. This will be
the topic of the next section.
20Previous studies; e.g. Griffith et al [11] could further distinguish between average wage
for operatives and average wage for administrative. We cannot make such a distinction
since since 1996 this information is not reported in the ARD.
21The AFDI data contains information on the sector of activity of the UK MNE’s
branches or subsidiaries abroad at the three digit level. Thus, we distinguish among the
following type of activity: manufacturing, wholesale, mining and quarrying and services.
21
5 Decomposing the MNE effect
Should we conclude from Section 4 that whenever a foreign firm takes over
a domestic plant or a domestic firm starts to invest abroad its productivity
will increase on average by 4.5%? Only if in estimating Equations 2 and 3 we
have not ignored any unobserved heterogeneity among plants and firms. To
understand in more detail which problems might have arisen let us consider
the following version of Equation 2:
yit =xitδ+bM NEit +µJ(i,t)+αi+²it (4)
where we decompose the error term εit into a firm effect µJ(i,t), a plant effect
αi, both assumed non time-varying, and ²it a statistical residual. µJ(i,t)cap-
tures factors which affect every plant in a particular firm. These include scale
effects and complementarities at the firm level or firm specific knowledge. αi
captures particular advantages of individual plants. This could include for
example the geographical features of a plant location or certain work cultures
and attitudes which occur at specific plants only.
Productivity advantages that arise from expanding a business internationally
- captured by bin Equation 4 - we call Generic MNE effect. These could arise
from factors such as scale effects or easier access to capital as well as comple-
mentarities of combining various national advantages. If multinational status
is correlated with firm and plant effects, the OLS estimate of βin equation
2 is an upward biased estimate of b.
What could drive a correlation between firm, plant effects and MNE status?
There are at least three factors. First, multinational firms could takeover
the best plants and firms. We call this the Cherry Picking effect. Second,
MNEs start up the best greenfield sites. Third, only the best firms become
multinational, which we call the Best firm effect. This is essentially the idea
of Dunning. MNEs are those firms which have an ownership advantage (high
µJ) which allows them to overcome the obstacles of setting up abroad and
22
still be competitive 22.
We illustrate this idea with a simple model. Demand is derived from a love
of variety utility function a la Dixit-Stiglitz [5] which gives each producer a
certain market power for her products:
U=µZN
0
q
σ−1
σ
i¶σ
σ−1
(5)
where σis the elasticity of substitution between differentiated goods and N
the number of firms operating in the market. Maximizing (5) subject to a
budget constraint leads to the following demand function for each producer:
qi=³pi
P´−σR
P(6)
where Ris the total revenue of the industry and pithe price of the variety
produced by firm i.Pis a composite price index 23. Suppose there is only
one input, for example labour. Each producer has a specific productivity µi
and a fixed set up cost fi. Given an economy wide wage w,µitranslates into
a marginal unit cost of ci=w
µi. Profit maximisation
πi= max
pi
[(pi−ci)qi−fi] (8)
22Could there also be a best plant effect? To answer this question, let us illustrate two
possible scenarios. In the first, a single plant (that is not part of any larger firm) with high
αiis more likely to start investing abroad. Its productivity would still be explained by a
plant and a firm specific component. The firm specific component captures factors which
are transferable to other plants at home or abroad, i.e. Dunning’s ownership advantage.
Thus, in this case, what looks like a best plant effect is a best firm effect. By contrast
consider the following scenario: there are credit constraints so that only firms which have
enough own resources will be able to invest abroad. Then MNEs will not necessarily be
firms that can transfer some superior knowledge to some other location but rather firms
owning plants that generated sufficient profit in the past; i.e. high αiplants. In this case,
there is a best plant effect.
23
P=ÃZN
0
p1−σ
i!1
1−σ
(7)
23
leads producers to observe the markup pricing rule
pi=ci
ρ(9)
where ρ=σ−1
σ
Using (6) and (8) the equation for profits of firm ican be written as:
πi=µw
µi¶1−σµ1−ρ
ρ¶ρσPσ−1R−fi(10)
Assume now that there are 2 countries Hand Fand both with an equally
sized continuum of entrepreneurs [0, E]. The productivity of these entrepreneurs
is distributed on a support µi∈[0,¯µ]24 according to a distribution function
Ξ : [0,¯µ]→[0,1] which - for simplicity - is the same in both countries. As-
sume now that set-up costs are the same for all firms setting up in their home
market fH→H=fF→Fbut higher when setting up abroad: fH→H< fF→H
and fF→H=fH→F. Each producer has now to decide if her productivity
µjmakes it worthwhile to set up in her home market. If this is the case
then she has to decide as well if her productivity is so high that even setting
up abroad is profitable. The existence of an equilibrium in this economy is
confirmed in the appendix. The equilibrium solution is characterised by two
cut-off productivity levels µand µM. Producers with µi< µ will not produce
at all, whereas producers with µi> µMwill be multinationals that produce
in both countries. From Equation 10 it follows immediately that µM> µ. As
a consequence the average productivity of MNEs will always be larger than
that of domestic firms:
E{µi|MNE} − E{µi|nonMN E} ≥ 0 (11)
Figure 2 illustrates this idea graphically.
Is the distinction between Best Firm,Picking and Generic MNE Effect of
any relevance? The British government has handed large subsidies to multi-
nationals in the past, partly in the hope that more foreign direct investment
24 ¯µrepresents the first best technology
24
Figure 2: The productivity distribution
25
would help boost aggregate productivity. Thus far, our results show that
attracting foreign capital is not the only solution to improve productivity;
British policy-makers should switch their policy focus from nationality of
ownership to multinationality of the firm. However, if the MNE is primarily
a picking effect then policies would not lead to any welfare improvement. If
the MNE effect is rather a Generic MNE or a Best Firm Effect, encouraging
the activity of MNEs would certainly lead to a productivity increase. But
also in this case, it is far from clear that subsidies to MNEs would bring wel-
fare gains: for that to happen we must have some additional market failure,
such as technology spillovers from MNEs to other firms, or credit constraints
which prevent firms from investing abroad even when it would be profitable
for them.
Distinction between the various effects is thus relevant in the current political
debate. Is there any hope that our data allows such a distinction?
We address this issue in the next section. We proceed by treating Equation 4
as a double fixed effects model. Techniques to handle such models have been
pioneered by Abowd et al. [1] in the context of employer-employee datasets.
In our case the dimension of the employee is replaced by the plant and the
dimension of the employee by the firm. Although in principle double fixed
effects means algebraically simply to include a dummy variable for each firm
and each plant, estimation and identification are far from trivial. In the next
section we explain in detail how we address the problem.
5.1 How to implement double fixed effects
Various identification issues arise in the estimation of double fixed effects
and of the parameter b. First, estimation of fixed effects is only possible
for plants that are present in the selected sample at least twice. Second,
separate identification of firm and plant effects is only possible to the extent
that plants change owner, or, using the matched employer-employee jargon,
that ‘plants move between firms’. Third, to be able to identify bwe need
the presence in the sample of domestic firms that start investing abroad (i.e.
26
Table 7: MNE status and ownership changes
!"#$!"#$ % &
%'( ') &&(
'* )
27
become an MNE) over the sample period. Table 7 reports the occurrence
of all these changes in our dataset. The upper panel reports the number of
status changes for each possible transition between UK non MNE, UK MNE
and Foreign. For example the cell in row 1, column 2 reports that there are
581 transitions from UK non MNE to UK MNE in the sample of selected
plants. The lower panel reports only the number of status changes that also
involved an ownership change. Therefore, the cell in row 4 column 2 reports
that 249 of the 581 UK plants that became multinational did so by means
of an ownership change, and thus a “move to a new firm”. This implies that
332 plants became part of a UK MNE because the firm they belonged to
became itself an MNE. This is the variation we use to identify b. In total,
the upper panel shows that we have 1686 changes between non MNE and
MNE status25. The lower panel shows that 1264 of those involved a change
in ownership. How many and which fixed effects can we identify from these
changes? To answer this question, we follow Abowd et al. [1] and define
sets of ‘double fixed effect groups’ (DFG). We define a DF group DF Ggas
the set of all firms and plants which interact over the sample period. A firm
and a plant interact simply if the plant is owned by the firm. Two plants
interact if they are both owned by the same firm at some but not necessarily
the same point in time. Two firms interact if they own the same plant at
different points in time.
Abowd et al. [1] show that for each plant and each firm in a DFG one can
identify a fixed effect which is informative about its productivity relative to
the group average, where the group average includes the fixed effect of an
omitted reference firm, µR, and an omitted reference plant αr. Thus, any
estimated fixed effect has to be interpreted as relative to the omitted plant
and firm. Table 8 reports various statistics concerning these groups in our
dataset. In total there are 7518 DF groups in our dataset. The columns of
Table 8 report statistics on the number of observations, firms, plants and
MNEs across these groups. For example from the third panel of column 1
25i.e. summing the off diagonal elements of row 1 and column 1 in the upper panel
28
Table 8: Descriptive statistics for DF groups
!"
"
#!
"
!""
"#$%$$ &"#' !
$!#%$$$&!
"#' !!!!#"
"# $
" $(
"! !""!
$( #)
29
we see that 699 is the largest number of observations in any single group.
In principle, one could estimate Equation 4 by least squares including a
dummy for each group, firm and plant and dropping a reference firm and a
reference plant per group. From the last three rows of Table 8 we see that
this would lead to the inclusion of 10517 + 10616 −7518 = 13618 dummy
variables. As well known from classical panel data applications, the inclusion
of so many variables is computationally unfeasible. We therefore proceed
with the following two-stage estimation procedure. In the first stage, we
apply a special kind of within transformation on Equation 4. For example
for y, we define:
eyit =yit −1
# [iJ(i, t)] X
τs.t.J(i,τ)=J(i,t)
yiτ (12)
and for all other variables analogously, where # [iJ] is the number of years
plant iis owned by firm J. Thus, we take deviations from within plant-
firm cell means. This transformation allows us to estimate all time varying
coefficients and - in particular b- consistently by applying least squares to
the following equation:
eyit =e
Xitδ+^
MNEitβ+eεit (13)
In the second stage we first estimate the sum of residual and fixed effects as:
cηit =yit −Xitb
δ−b
βM N EJ(i,t)(14)
We then run, for each DF group gseparately, a least squares regression of
cηit on a set of dummy variables for the firms and plants in the group and a
constant
cηit =Zitgγ(15)
where Zitg is a row vector with 1 + Fg+Dgelements, Fg, the number of
firms and Dgthe number of plants in DF group g. This is only possible if
the number of firms and plants in any given group is not too large. Table 8
confirms that the largest group contains 57 firms and 212 plants, a total of
30
269 which is still computationally feasible. The second stage nature of this
regression implies a non standard covariance formula for the estimated fixed
effects γ:
Σγ=σ2
ε(Z0
gZg)−1
+(Z0
gZg)−1Z0
gXgΣβ,δ X0
gZg(Z0
gZg)−1
−σ2
ε(Z0
gZg)−1Z0
g(Xg(e
X0e
X)−1e
X0Qg)Zg(Z0
gZg)−1
−σ2
ε(Z0
gZg)−1Z0
g(Q0
ge
X(e
X0e
X)−1X0
g)Zg(Z0
gZg)−1
(16)
where Qgis a block diagonal matrix of dimension N×Ng. The blocks consist
of idempotent transformation matrices Qg,iJ 26 of dimension #iJ ×#iJ for
each combination of firm Jand plant iin group g.Nis the total number
of observations in the dataset Ngthe number of observations in group g.
Further, let iJgthe total number of firm-plant combinations in group g.
Then
Qg=
Qg,10. . . 0
0 Qg,2. . . 0
.
.
..
.
..
.
..
.
.
0. . . 0 Qg,iJg
0. . . . . . 0
.
.
..
.
..
.
..
.
.
0. . . . . . 0
(17)
and
Qg,iJ =I#iJ −1
#iJ e#iJ e0
#iJ (18)
5.2 Testing for various MNE effects
Testing for a Generic MNE Effect, b > 0, follows from the first stage regres-
sion (Equation 13). To test for the best firm effect we need an estimator of
26compare with Hsiao [14] p31
31
the statistic in Equation 11, the difference between a MNE and a non MNE
firm effect:
∆F=E{µi|MN E} − E{µi|nonM N E}
The obvious sample analog is the difference between estimated MNE and
non MNE fixed effects:
1
#MFX
J∈
M
FbµJ−1
#DFX
J∈
D
FbµJ(19)
where MFis the set of all firms in our sample that are multinational at some
point in the sample period and DFits complement. The problem with this is
that any fixed effect we can estimate will always be relative to its DF group’s
reference firm; i.e. we cannot estimate µJbut only µJ−µRg(J), where µRg(J)
denotes the fixed component of the reference group productivity. This leads
to the following test statistic:
b
∆F=1
#
M
FPJ∈
M
F
\
µJ−µRg(J)
−1
#
D
FPJ∈
D
F
\
µJ−µRg(J)
(20)
Both, b
∆Fwill be an unbiased estimators of (20) if there is no systematic
relationship between the reference group for a particular firm and its multi-
national status, which implies:
E{µRg(J)|J∈MF}=E{µRg(J)|J∈DF}(21)
Since the choice of the firm and plant within each DF group that become the
reference group is random, no correlation might be introduced in this way.
Yet, the groups differ considerably in size and in the presence of MNEs.
Also it could be possible that multinational firms with higher productivity
are more likely to exchange plants with other high productivity firms. As a
consequence we expect that multinationals have a higher probability to be
in groups with a high productivity reference firm so that:
E{µRg(J)|J∈MF}> E{µRg(J)|J∈DF}(22)
32
This would bias b
∆Fdownward, which implies that if we were to reject the
hypothesis that there is no multinational firm effect on the biased statistic,
then we would also reject it for a non biased version. In other words: If we
find any positive MNE firm effect in this way than we can be quite sure that
it is really there.
We can compute a similar statistic for the plant level:
b
∆P=1
#MPX
J∈
M
P
\
αi−αrg(i)−1
#DFX
J∈
D
F
\
αi−αrg(i)(23)
Again it might be downward biased if multinationals tend to be in groups
with above average plants. In the following section we describe the results
of the double fixed effect estimation and these various statistics.
33
Table 9: Is there a Generic MNE Effect?
!""
#$ %&
'($)!
*$+$$
$"#,-$"#.,-$"#,
5.3 Decomposition Results
Is there any evidence for a Generic MNE Effect? Table 9 reports regression
results for Equation 13, the first stage of our Double Fixed Effects procedure
described in Section 5.
We find a significant value of about 2 percent for the coefficient bon the
multinational dummy. This finding suggests that there is a significantGeneric
MNE Effect; on average, becoming multinational boosts a firm’s productivity
by 2 percent. Table 10 reports average values for the firm and plant fixed
effects along with the test statistics discussed in Section 5.1. Consider first
the firm effects displayed in the upper panel. The point estimates reported
in the second row suggest the following ranking: non US MNE come first
followed by non MNEs and US MNEs are last. However, rows 2 and 3 reveal
that any differences between the three groups are not significant. This means
that we cannot find any evidence for best firms effects whatsoever.
Panel 2 shows the the results for plant effects. The ranking here is different.
34
Table 10: The evidence on Best Firm Effects and Plant Picking Effects
!"#!$!%&!%#'!!$#!!!(!)!!
%'!*!!!!!!!
!!+$!!'!!%!!,)!)!!
!+$!)!,!!!!!!)!,!
!!!!!$#
$#!
#%#
!#%#
--
!!!$
-
35
Plants owned by US MNEs turn out to be the most productive ones. They
have a significant advantage of about two percent over other MNEs. Equally
other MNEs are signficiant two percent more productive than non MNE
plants.
Is the empirical evidence therefore suggesting that there is no Best firm
effect and the multinational effect is essentially driven by cherry picking of
the best plants? Maybe, but not necessarily. There are other explanations
which equally fit the facts. First are the biases discussed in Section 5.1. If
MNEs self select themselves into DF groups with other high productivity
firms then we might not detect a multinational effect even though there is
one. Secondly, we should remember on what the identification of the firm
effects rests: our estimator allocates a high firm fixed effect µJto a firm
J, if the productivity of plants that are taken over by J rises subsequent to
the takeover. If the beneficial impact of a firm’s intangible assets on the
productivity of its plants27 does not affect all plants in the same way then
this could well lead to the results we get. A notable example of this latter
case is that MNEs can only achieve high productivity in green field startups
and not in existing plants they takeover.
Even if we take this last point into consideration our hold nevertheless a clear
message for policy makers: There is no evidence that encouraging MNEs to
takeover existing plants is a policy which will have dramatic direct effect
on the UK’s productivity performance. From our results we would expect a
modest improvement of about two percent as a consequence of the Generic
MNE Effect 28
27i.e. Dunning’s ownership advantage
28Note that the Generic MNE effect, the firm effect and plant effect need not add up to
the overall level effect of about 4 percent found in Table 6, to the extent that µJand αi
are correlated. Also, if there is either a strong firm or plant effect then the MNE effect
found in the pooled level regresssion is lower because high performance plants or firms
that become multinational only at the end of the sample contribute to a higher average
performance of the non MNE group earlier in the sample.
36
6 Conclusions
We started by conjecturing that what has been considered up to now a for-
eign effect is most likely a multinational effect. We find that this conjecture
is true in general: the foreign effect is in fact a MNE productivity advantage;
multinationals are more productive than domestic plants, whether foreign
owned or not.
Our level regressions provide strong evidence of a US productivity advan-
tage. US owned establishments are consistently more productive than other
MNEs. Indeed the ranking of productivity advantage from our level regres-
sions is exactly the same as the one found by Doms and Jensen: US MNEs
are the most productive, followed by non US MNEs and establishments of
domestic non-MNEs being the least productive.
When we analyse the nature of the MNE effect in more detail using the
longitudinal dimension of our data we find a significant causal effect from
multinationality on productivity of 2 percent. We cannot find that multi-
national firms have a positive impact on plants they take over beyond that
which would indicate a Best Firm Effect. We find a large positive difference
between fixed effects of MNE and non MNE plants which suggest that MNEs
are very good at taking over the best firms or starting up the best plants
on green field. For economic policy this implies that encouraging MNEs to
takeover domesitc firms or domestic firms to become MNE would at best
lead to direct productivity gains of 2 percent.
References
[1] Robert H. Creecy Abowd, John M. and Francis Kramarz. Comput-
ing person and firm effects using linked longitudinal employer-employee
data. March 2002.
[2] Christensen L.R. Caves, D.W. and W.E. Diewert. Mulitlateral compar-
isons of output, input and productivity using superlative index numbers.
37
Economic Journal, 92:73–86, March 1982.
[3] Chiara Criscuolo and Ralf Martin. Using the annual inquiry into foreign
direct investment to create a multinational identifier in the ard. Ceriba
data guide, 2003.
[4] Richard Disney, Jonathan E. Haskel, and Ylva Heden. Exit, entry and
establishment survival in uk manufacturing. Journal of Industrial Eco-
nomics. forthcoming.
[5] A. Dixit and J. Stiglitz. Monopolistic competition and optimum product
diversity. American Economic Review, 67:297–308, June 1977.
[6] Mark E. Doms and J. Bradford Jensen. Comparing wages, skills and
productivity between domestically and foreign-owned manufacturing es-
tablishments in the united states. In R.E. Lipsey R.E. Baldwin and J.D.
Richardson, editors, Geography and Ownership as bases for economic
accounting, pages 235–258. Universtiy of Chicago, 1998.
[7] John Dorgan, Stephen J.and Dowdy and Peter Whawell. Better uk
productivity: An inside job. The McKinsey Quarterly, (4), 2001.
[8] J. H. Dunning. International Production and the Multinational Enter-
prise. George Allen and Unwin, 1981.
[9] Office for National Statistics. Foreign Direct Investment 2000 - Business
Monitor MA4. HMSO, London, 2002.
[10] Rachel Griffith, Helen Simpson, and Frank Windmeijer. Understanding
productivtiy differences between foreign and domestic firms. October
2001.
[11] Rachel Griffith and Helen Simpson. Characteristics of foreign owned
firms in british manufacturing. IFS Working Paper, March 2001.
38
[12] Rachel. Griffith. Using the ard establishment level data to look at foreign
ownership and productivity in the uk. Economic Journal, 109:F416–
F442, June 1999.
[13] Richard Harris. Efficiency in uk manufacturing 1974-1994. mimeo, 12
1999.
[14] C. Hsiao. Analysis of Panel Data. Cambridge University Press, Cam-
bridge, 1986.
[15] Stephen H. Hymer. The efficiency (contradictions) of multinational cor-
porations. The American Economic Review, 60, Papers and Proceed-
ings of the Eighty-second Annual Meeting of the American Economic
Association(2):441–448, May 1970.
[16] James R. Markusen. The boundaries of multinational enterprise and
the theory of international trade. Journal of Economic Perspektives,
9(2):169–186, Spring 1995.
[17] Ralf Martin. Building the capital stock. 2002.
[18] Barnes Matthew and Ralf Martin. Business data linking: An introduc-
tion. Economic Trends, (581), 2002.
[19] Mary O’Mahony and Willem de Boer. Britain’s relative productivity
performance: updates to 1999. Final report to dti/hm treasury/ons,
NIESR, March 2002.
[20] Nicholas Oulton. The abi respondents database: A new resource for
industrial economics research. Economic Trends, 528:46–57, November
1997.
39
A Variable Definitions
•Capital stock: capital stock was calculated using a perpetual inventory
method (PIM). For a more detail description of the method adopted
we refer to Martin [17]
•Deflators: to deflate output measures (gross output and value added)
we use producer price indices at the 4-digit SIC92 industry level. To de-
flate intermediates, we use material price deflators at the 2-digit SIC92
industry level. The base year is 1995. Capital stock is deflated us-
ing investment deflators with base year 1995; for years pre-1995 these
are implicitly derived from nominal and real sectoral ONS historical
investment series. From 1995 on we use the publicly available MM17
series.
•Foreign plants are plants owned by foreign owned enterprise groups
•Foreign owned, Headquarters in the UK (Foreign Head-UK) are foreign
owned enterprise groups that are undertaking foreign direct investment
from the UK.
•we define MNEs with affiliates in the manufacturing sector (Manu-
facturing MNEs) those MNEs that have at least one affiliate in the
manufacturing sector abroad.
•Country groups:
EUnorth includes plants owned by Austria, Belgium, Denmark, Fi-
nalnd , Luxembourg, Sweden and Republic of Ireland.
EUsouth includes plants owned by Italy, Spain and Canary Islands,
Portugal, Greece.
Tax includes plants owned by British Virgin Islands, Channel Islands,
Isle of Man, Liechtenstein, Antigua and Barbuda, Cyprus, US
Virgin Islands.
40
otherEurope includes plants owned by Norway and Switzerland.
otherOECD includes plants owned by Australia, Iceland, Poland,
Mexico, Turkey, Czech Republic and South Korea.
other is a residual category that include plants owned by the rest of the
world and plants which are foreign owned but whose nationality
is unknown.
•We calculate TFP relative to the 4 digit industry median using the
differential TFP formula of Caves et al. [2]; i.e. we calculate TFP as
lnT F P it =lnYit −ln ¯
YIt
−¯αK(lnKit −ln ¯
KIt )
−¯αL(lnLit −l n ¯
LIt )
−¯αM(lnMit −ln ¯
MIt )
where ln ¯
YIt denotes the 4 digit industry median and the factor shares
are the mean of the plant factor share and the median industry factor
share ¯αK=alphaK it+¯
alphaKI t
2.
•Weights are calculated using the register employment information on
the basis of 4 digit sector, region and employment cells. For each cell
ithe weight is caclulated as Number of plants in register in cell i
Number of selected plants cell i .
41
B Equilibrium in the MNE model
This section shows that an equilibrium exists in the Dixit-Stiglitz style econ-
omy described earlier. Recall how the equilibrium is determined in the stan-
dard Dixit-Stiglitz Model. There, unit (c)and fixed costs (f),and conse-
quently prices (p) are the same across all firms. The total number of firms
that an industry supports (N∗) is then found by the zero profit condition
which reduces to
0 = N−1κ−f(24)
where κ=c−1(1 −ρ)R.
This condition is well defined only for σ > 1 because the first term on the
right hand side will be positive and declining in N. In our case the problem
is more complex because unit costs and fixed costs vary across entrepreneurs
that are active. Matters can be solved in a very similar fashion however once
we realize that - subject to the cost distribution 1
2Λ(·) being invertible - we
can write costs as a function of the number of active entrepreneurs. If we
normalise the total mass of entrepreneurs in each country to 1 we can write
the mass of active entrepreneurs as
F(µ, µF) =
2 (1 −Ξ(c)) if µ≥µF
2−Ξ(µF−Ξ(µ) otherwise
(25)
If Ξ(·) is invertible we can invert F(·)29. The result is:
µ(N, ¯
NF) =
Ξ−1¡1−1
2N¢if N≤¯
NF
Ξ−1¡1−N+1
2¯
NF¢otherwise
(26)
where ¯
NFis the mass of firms in the market beyond which foreign multi-
nationals do not enter. µ(·) is decreasing in Nbut non-decreasing in ¯
NF.
For a given mass of firms, ˜
N > ¯
NF, increasing the mass of firms from abroad
29All that is required for that is a positive density of the productivity distribution
42
allows to fill up the mass with more higher productivity firms because we
can draw from both the home and foreign pool of firms.
The market equilibrium can now be stated in terms of ¯
NFand ¯
N- the
total mass of active firms. It is characterized as a situation in which the
least productive foreign firm and the least productive domestic firm make
zero profits:
P(¯
N, ¯
NF)σ−1µw
Ξ−1(1 −¯
NF)¶1−σ
˜κ−fF= 0 (27)
P(¯
N, ¯
NF)σ−1µw
Ξ−1(1 + 1
2¯
NF−¯
N)¶1−σ
˜κ−fH= 0 (28)
where ˜κ= (1 −ρ)ρσ−1Rand
P(¯
N, ¯
NF) = ÃZN
0µw
ρµ(n, ¯
NF)¶1−σ
dn!1
1−σ
(29)
Note that P(·) is decreasing in both, ¯
Nand ¯
NF. The intuition for this is as
follows: If ¯
NFincreases while ¯
Nstays constant we have the same mass of
firms in the market but because this mass is now selected for a larger interval
from foreign as well there will be more higher productivity firms than before.
Because the lower costs are partly passed through to consumers the overall
price index declines. Increasing ¯
Non the other hand increases the total
number of products produced and therefore competition among producers.
Because consumers have now more products to substitute to they are forced
to reduce prices.
To proceed divide the 2 conditions. This yields
Ξ−1¡1−¯
NF¢µfF
fH¶(1
1−σ)
= Ξ−1µ1 + 1
2¯
NF−¯
N¶(30)
The equilibrium can now be characterized by 30 and 27. Equation (30)
establishes ¯
Nas an increasing function of ¯
NF:
d¯
N
d¯
NF
>0 (31)
43
Because the left hand side of 27 is decreasing partially in ¯
Nand ¯
NFit is thus
decreasing in ¯
NFtotally. Profits are therefore always lower than fixed costs
and no production takes place or we can always find a mass ¯
NFand in turn
¯
Nsuch that profits of the least productive firms become zero.
44
C Robustness checks
Table 11: Robustness checks
!"
#""
$!
%"
&'%
()&'%
'()&'%
*!#
!*!#
!#
+,
,!!
-.
*!#
!!" # $% &!! #' !! ('
)* #+( ,- #+( ,- #+( ,
45
Table 12: Further Robustness checks
!
"
#"
$
$
$%
%
!
&'(
)
%"
%
%*#+
,
*%
*"-"
."/
!"# !"# !"
$ ! !%!& !&'&()*
+''),(-''&''!&
46
Table 13: Dependent Variable: log real value added
!
! "
#$
%&'
()'
!"# !"# !"
$%%&%%
47
