POLICY RESEARCH WORKING PAPER 2598
Weightless Machines and Knowledge, lacking weight
and other physical attributes
Costless Knowledge and being intangible, is a
hidden factor of production,
An Empirical Analysis of Trade making economies grow
"weightless." But knowledge
and Technology Diffusion is also embedded in
technology. If. because of low
Giorgio Barba Navaretti productivity, poor countries
Isidro Soloaga keep buying low-technology
machines, will they remain
stuck in a low-technology,
low-growth trap?
The World Bank
Development Research Group
Trade
May 2001.
I POLICY RESEARCH WORKING PAPER 2598
Summary findings
Barba Navaretti and Soloaga examine the impact on At any point in time between 1989 and 1997, there is
productivity of technologies imported by a sample of a persistent (even increasing) gap between the unit values
developing and transition economies in Central and of the machines imported by the United States and those
Eastern Europe and the Southern Mediterranean- imported by the sample of developing countries.
economies becoming increasingly integrated with the Although developing economies buy increasingly
European Union. productive machines, the technology embodied in the
They depart from earlier studies of technology machines persistently lags behind that in the machines
diffusion by focusing on the technology embodied in the purchased by the United States-so far as unit values are
machines imported. Earlier work focused mostly on good proxies of embodied technologies.
spillovers from foreign research and development Barba Navaretti and Soloaga also find that productivity
conveyed through trade, without controlling for the growth in manufacturing depends on the types of
characteristics of the goods imported. machines imported in a given industry. So although the
The authors jointly estimate the choice of foreign optimal choice for developing countries is to buy
technology and its impact on domestic productivity for a cheaper, less sophisticated machines, given local skills
set of manufacturing sectors. They proxy the and factor prices, this choice has a cost in long-run
technological level of the machines imported by using an productivity growth. If productivity is low, countries buy
index relating the unit value of the machines imported by low- technology machines, but doing so keeps them in a
a given country to the unit value of similar machines low-technology, low-growth trap.
imported by the United States.
This paper-a product of Trade, Development Research Group-is part of a larger effort in the group to assess the role
of trade in technology diffusion. Copies of the paper are available free from the World Bank, 1818 H Street NW,
Washington, DC 20433. Please contact Lili Tabada, room MC3-333, telephone 202-473-6896, fax 202-522-1159, email
address Itabada@worldbank.org. Policy ResearchWorking Papers are also posted on the Web athttp://econ.worldbank.org.
The authors may be contacted at barba@unimi.it or isidros@iadb.org. May 2001. (29 pages)
The Policy Research Working Paper Sehes disseminates the findings of work in progress to encourage the exciange of ideas about
devmelopment issues. An obJective of the series is to get the findintgs out quickely, even if the presentations are less than fully polished. The
papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the autbors. They do not necessanily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced by the Policy Research Dissemination Center
Weightless Machines and Costless Knowledge
An Empirical Analysis of Trade and Technology Diffusion
by
Giorgio Barba Navaretti (*)
and
Isidro Soloaga (**')
(*) Universita' degli Studi di Ancona, Centro Studi Luca d'Agliano and Fondazione Eni
Enrico Mattei
(**) Poverty and Inequality Unit, Inter-American Development Bank
This paper has been prepared as part of the World Bank project on 'The Microeconomic
Foundations of Technology Diffusion' and for the World Bank-Bari University Conference:
"Whither South-Eastern European Countries' Development? A Snall and Medium-Sized
Enterprises Perspective for Integration with the European Union". The paper was written
while the authors were respectively Visiting Researcher and Long Termn Consultant at the
World Bank's DECRG-Trade Group. They wish to thank Francis Ng and Alessandro Nicita,
for invaluable help in constructing the data base, Gani Aldashev and Paolo Zagaglia for their
research assistance and Bernard Hoekman, Aart Kraay, Riccardo Lucchetti and James
Tybout for providing useful suggestions. Giorgio Barba Navaretti is also grateful to David
Tarr and Tony Venables for arranging and supporting his stay at the World Bank while
working on this project.
L Inbvductdon
Federal Reserve Chairman Alan Greenspan once noted that through the second half
of the twentieth century, the US tripled the real value of its output with no increase in the
weight of the material produced (Washington Post, 2000). Accordingly, in a recent paper
Danny Quah labels as 'weightless' an economy where knowledge products represent an
increasingly large share of national income(Quah, 1999). Knowledge, lacking physical
attributes and, to use more standard accounting definitions, being intangible, is a hidden
factor of production making economies grow 'weightless'.
Weioasnessnes has its nominal counterpart: crsd&snss. Every one familiar with a
keyboard knows well that the cost of computers in the last ten years remained more or less
stable or even declined, although computers' capacity to process information skyrocketed.
TIhus, knowledge spreads and accounts for an increasing share in income because it is
virtually costless. However, this is true only if we allow for a sufficiently long time horizon.
Top technologies do not diffuse instantly and it takes time before they become available to
everybody at a reasonable price, even when embodied in durable equipment or production
machines. Unfortunately, (fortunately for inventors) what matters for diffusion is the link
between technologies and prices at any point in time, not the time trend of a given
technology. If we enter a computer shop we find that the prices of machines grow with the
numnber of megahertz and other ragnificent embodied features. Only a few can buy a top
computer of today, even if anyone will be able to afford it sometimes in the future. The
prices of technologies at any point in time reflect their relative productivity.
This paper discusses whether developing countries' choices of imported factors of
production are influenced by the link between technological complexity and prices and, in
turn how these choices affect productivity in manufacturing.
We take the most physical of all factors of production: machines. We use trade data
to construct average unit values per ton of machine' exported bythe EU to a sample of
' We can show that unit values per ton of machine are very highly correlated to unit values per number of
machines. We use the former, as the latter are available for a limited number of machines and countries only
neighboring developing and transition countries in Central-Eastern Europe and in the
Southern Mediterranean. We work with homogeneous groups of machines, by using very
disaggregated trade data. We also take the export of EU machines to the US as a top
technology counterfactual.
If we look at country averages, the evidence is perfectly consistent with our stylized
facts on computers. The unit values in nominal Euros of machines imported by the US from
the European Union between 1989 and 1997 rises just marginally, notwithstanding dramatic
increases in productivity. However, at any point in time there is a persistent gap between the
unit values of the machines imported by the US and those imported by our sample of
developing countries. We also find that this gap is inversely correlated to a broad
development indicator like income per capita. Thus, although backward economies buy
increasingly productive machines, the technology embodied in these machines persistently
lags behind the one purchased bythe US, as far as unit values are good proxies of embodied
technologies.
We then work at the industry level and we analyze jointly the choice of the
technological complexity of the machines imported and the impact of embodied
technologies on total factor productivity. We find that the choice of lower technologies is
optimal for developing countries, given local skills and factor prices. We find that imported
technologies have a favorable effect on productivity, but this effect hinges upon the type of
technologies embodied in the machines imported. In other words, an increase in the level of
complexity of the machines imported has a larger impact on TFP growth than an increase in
the share of imported machines on total investments. This implies that the persistent
technological gap also generates a persistent divergence of income levels between advanced
and industrialized economies.
Many recent contributions empirically analyze how international economic
integration creates new channels for transferring technologies and knowledge and how these
channels affect productivity2. These contributions have looked at the impact of imports as
conveyors of R&D spillovers (Coe and Helpman, 1995, Coe Helpman and Hoffmaister,
2 See Barba Navaretti and Tarr, 2000 for a review
2
1997, Keller, 2000), of foreign direct investments (Blomstrom and Kokko, 1998)3 and of
exports (Cerides, Lach and Tybout, 1998, Bernard and Jensen, 1999, Aw, Chung and
Roberts, 2000, Kraay, 1996).
None of these works, however, have looked at the role of imported machines in
transferring embodied technological progress. Take the most quoted reference in this area,
Coe and Helpman 1995. The basic idea in their paper, derived form earlier theoretical
contributions on endogenous growth in open economies, is that growth in a given country
depends both on domestic and foreign stocks of technological knowledge. Foreign
knowledge is acquired as a costless externality by importing goods from countries which are
rich in R&D. Coe and Helpman construct a trade weighted foreign R&D stock using the
share of imports from each partner country on total imports as weights and they evaluate its
impact on aggregate domestic total factor productivity (1PP). Coe, Helpmian and
Hoffmaister, 1997 carry out a similar exercise, using imports of machines as weights and
Keller, 2000, estimates the impact of foreign R&D stocks on TFP at the sector level.
That trade is a great channel to circulate ideas it is a well known and very ancient
story. Trade in goods certainly generates parallel channels for the weightless exchange of
technological knowledge. Yet, technological knowledge is also emnbodied in the goods
imported. When these goods, like machines, are used as factors of production their
technological features are likely to directly influence productivity. The contributions
discussed above, by just focussing on extemality, somehow misrepresent the process of
technology transfer. Imagine a given country importing the same total value of goods (or
capital goods) from two countries which have the same R&D capital stock The impact of
imports from both countries on domestic productivity is expected to be exactly the same in
the Coe-Helpman, Hoffmeister, Keller framework. But what if the bundle of goods
imported from the two countries is different, and in particular what if the machines
imported from one of the two countries are much more productive? Then the impact of
imports on productivity should be different in the two cases. This is preciselythe central
result of this paper.
3 Blomstrom and Kokko 1998 provide a good survey. See also Blomstrom and Persson, 1983, Hiddad and
3
In principle it should be possible to argue that the larger the R&D capital stock of a
country the more technologically advanced the machines imported from that country. But
we show in this paper that this is not the case in general. If we take the average unit value of
the manufacturing machines exported from the European Union theyvary quite widely
across importing country. Thus the bundle of machines exported from any given country (or
group of countries) may vary, independently of this country's R&D capital stock
The results of this paper are consistent with and extend our earlier work In Barba
Navaretti, Soloaga and Takacs, 2000 we explore the choice of the vintage of the machines
imported by developing countries from the US. We find that vintage, like technological
complexity is explained by factor prices and skills prevailing in the importing countries. In
Barba Navaretti, Galeotti and Mattozzi, 2000 we use a measure of technological complexity
related to the skills necessary to use the machines imported and show that this measure of
technological complexity has a positive impact on the export performance of textile products
for a sample of Eastern European and Mediterranean countries. Compared to these earlier
works, here we use a different measure of technological complexity (unit values of importd
machines), that can be computed for all machines used in production. We can consequently
work on a broad number of manufacturing sectors and focus on TFP as our performance
variable.
In the next section we discuss our data set and sample countries. We then construct
our measure of embodied technology and present some descriptive evidence on trends
imported technologies. Section four examines the impact of imported technologies on total
factor productivity and the deteminants of the choice of imported technologies for a sample
of manufacturing sectors. Section 5 concludes. In Appendix 1 we develop a simple analytcal
framework as a background to the econometric analysis of sections 4.
2. Data and sample countries
The aim of the empirical analysis is to study the determinants of the choice of
imported machines and the impact of embodied technologies on total factor productivity for
Hanrison, 1993, Harrison, 1996 and Djiankov and Hoekman, 2000.
4
some manufacturing sectors. The main problem concerning data is to derive
correspondences at the industry level between categories of machines imported and the
industries using them in production. We are able to do so at the three digit (ISIQ industry
level by matching data on productivity derived from industrial statistics (UNIDO) and data
on imports of technology derived from trade statistics (GOMEXT-Eurostat). The industry
matching is available for thirteen sectors, reported in Appendix 2. Before deriving an
empirical model, we discuss the choice of the sample countries and the construction of the
variables measuring imports of technology. We also run some simple descriptive statistics of
average unit values of machines imported by our sample countries.
The study focuses on six Central and Eastern European (Bulgaria, Poland and
Hungary) and Southern Mediterranean (Egypt, Israel and Turkey) countries and on their
imports of machines from the European Union. There has been a dramatic increase in the
degree of economic integration between these areas, with growing flows of trade and foreign
direct investments. For most of the sample countries the EU is by far the major source of
imported technologies: 60 to 90% of their machines are imported from the Union. We do
not miss much by not considering their trade with the rest of the world.
The sample countries differ quite much in terms of their pattern of trade
liberalization. The Central and Eastern European ones have liberalized suddenly, following
the fall of the Berlin Wall. There is overwhelming consensus that the extent of trade reforms
has been considerable. Turkey is instead an earlier liberalizer. It implemented extremely
liberal policies already in the Eighties and it recently decided to implement a Custom Union
with Europe, applying the EU common external tariff to third countries' imports. The other
Southern Mediterranean countries have mixed performances. Although some of these
countries Israel, and Egypt) have negotiated or are negotiating reciprocal free trade
agreements with Europe, trade regimes have been up to now, and often still are, quite
protectionist.
The countries in our sample also differ in terms of their level of development and
production structure. GNP per capita varies between 1,380 US $ of Bulgaria to the 16,180
US $ of IsraeL We have therefore a sufficiently differentiated picture to understand the role
5
of trade liberalization and of differences in the level of development which may affect
technological choices.
3. Measurng technological complexity
Now, how can we measure the level of technological complexity, or, rather the type
of technologies embodied in imported machines? In an earlier paper we devised a measure
of technological complexity for metalworking machines (Barba Navaretti, Soloaga and
Takacs, 2000). This measure was based on the minimum skills necessary to use a machine
and it was constructed with the help of specialized engineers. With sufficiently disaggregated
trade data, it was possible to assign the skill index to each type of machine. The same
measure was also applied to textile machines in a subsequent paper (Barba Navaretti,
Galeotti and Mattozzi, 2000). Unfortunately this measure is only restricted to these two
industries and cannot be used for the rest of manufacturing. Moreover, it cannot be used to
compare technologies embodied in the same type of machines imported by different
countries, as all machines in a given trade category would get the same ranking. For example
numerically controlled horizontal lathes have a higher index than manually controlled one.
But we would be unable to distinguish between more or less advanced numerically
controlled horizontal lathes imported by a given country.
In this paper we use a more straight forward measure, the unit values of machines.
Does this indirect measure capture differences in technological complexity? In a competitive
market we expect that differences in the price of similar machines (e.g. numerically
controlled horizontal lathes) reflect differences in productivity. As discussed in the
introduction, at any point in time price of machines differ according to productivity. Indeed,
if we correlate the unit values of the metalworking machines exported from the U S, with the
index of technological complexity discussed above, we find veryhigh correlation ratios
varying from 0.60 to 0.95, depending on the level of definition of the machines considered.
A second problem is the use of unit values to compare across different types of
machines. Different types of machines can have very different prices because they are
inherently different (a loom vs. a lathe) not because theyare more or less complex. To
control for this composition effect, we construct a unit value index by normalizing the unit
6
values of machines classified at the six digits level in trade statistics (harmonized code)
imported bya given country by the unit value of the same machines imported by the US.
More specifically, the unit value index for a six digit machine i imported by country c at time
t is given by.4
UViat = (UVia/UViLSi)
Where the denominator is the unit value of the same machine i imported by the US
at time t. To match data on unit values with data on productivity, which are measured
according to three digits ISIC classification, we must aggregate all the six digits unit value
indices corresponding to all those machines used in any three digit ISIC category. For
example, if we are interested in computing the unit value index for the textile industry to see
what is its impact on productivity in textile, we must aggregate the unit value index of all
textile machines. The correspondence between the harmonized codes of the machines and
the ISIC codes of the industries using them is reported in appendix 2. Thus, the average unit
value index of the machines used in the 3 digit ISIC industry j in country c at time t is given
by.
UVI6 = E (UvIct-~)
i=O V (1)
where n is the number of six digit categories i corresponding to the ISIC three digit
categoryj, and V.,, is the value of machines i imported byc at time t and Vj is the total
values of machines used in j imported by c at time t.5
4Note that if we were to use the skill index discussed above, we could not compare machines whithin a six
digit category.
5 Note that this index is affected by a composition effect. The index can increase with time either because
countries buy the same bundle of machines, and the value of each or some of them increases with time, or
because bundles change towards machines with a higher average unit value. To avoid this problem it is possible
to constnict Tornqvist price indices, where weights are fixed with time, normally the period average weights,
Aw and Roberts (1986). However, our unit values are already normalised across machines. Thus an increase in
the index due to a composition effect does indeed capture a process of technological upgrading that we want to
observe.
7
A final problem, is that unit values may capture market imperfections like market
power or trade barriers, which do have an influence on prices. However, our countries are
small and we can reasonably assume that the price of machines is given for them Moreover
prices are f.o.b prices in current Ecu at the EU frontier, thus theyshould not be distorted by
tariffs and other policies in the importing country
Given that the EU trade statistics (Eurostat-Comext) provide sufficiently
disaggregated data in both values and quantities, we focus on imports from the EU As
argued, all of our sample countries import most of their machines from the European
Union. For example, the average share of textile and clothing machines imported from the
EU on total textile machines imported between 1988 and 1996 is never lower than 66%
(Barba Navaretti, Galeotti and Mattozzi, 2000). Quantities of machines are measured in
metnic tons. For some countries, machines quantities are also measured in terms of number
of machines, but these dat are not as widely available as the former ones. Unit values
computed using the two quantityunits are very highly correlated. TMe analysis is carried out
for the period between 1989 and 1997.
It is useful to observe how these indices behave across countries and with time. To
do so we compute simple numerical averages for each year for some of the importing
countries. We start with row average unit values (not normalised), so as to compare the trend
of unit values of machines imported by the IDS from the EU with those imported by other
countries (Figure 1). We observe a stable gap between the unit values of the machines
imported by the US and by the other countries. If we exclude Hungary and Poland, we note
that for all the other countries unit values are quite stable, although for the US they tend to
rise and for the other countries slightly decline. This evidence suggests that the price of
machines is stable vith time and does not increase with technical progress. But if at any
moment in time the price of any given machine is linked to its productivity, developing
countries are on average importing less productive machines than the US.
As discussed above, the trends reported in figure 1 maybe affected by the
composition effect. In figure 2 we therefore report the average of the Unit Value Index (7),
where unit values are nonmalized by the unit values of US imports (Figure 2). We see that
the trends are very similar to the one reported in figure 1.
8
Another way to assess the persistency in the technology gap of the machines
imported is to compare unit value indices with their lags. In Figure 3 we plot the unit value
indices of machines imported at time t with the unit value indices of the same machines
imported 1 (3a), 2(3b) and 7 years earlier (3c). The thick line is the diagonal. The figures
show a striking persistence in the gap. The indices are positively correlated with their lagged
values, even with 7 years lags. Moreover, note that a large share of the dots lie above the
diagonal, and that this share increases the longer the lag. This implies that for many sectors
and countries the gap between the machines they import and the machines imported by the
US is not only persistent but even increasing.
Figure 1
Trends in Aggregate Unit Values
28-
26-
24 - _Hungary
22- -- . -..*>. ...............-. -- . ... Israel
Morocco
20
-*-Poland
- Tunisia
2 16 _-+-Turkey
14 U -USA
... _2 s,. - - ._ , --s- --w-
10
1989-91 1990-92 1991-93 1992-94 1993-5 1994-6 1995-97
Three years moving averages
Figure 2
Trends in the Aggregate Unit Value Index
130
1^20i@
_ 1 x n -~~~~~~~~ffi-@ i
110 ~~~~~~~~~~~~~~~~~~~~~~*-Hungary
Israel
100
--Morocco
go> s = _-Poland
=" 80 _ eE + Turkey
70
60
1989-91 1990-92 1991-93 1992-94 1993-5 1994-6 1995-97
Three years moving averages
9
Figure3 a,b
Persistency of the technology gap
land 2 year lags
0 llnuvit A eqlnuv
6.04197 0 A
0 0 0 0sd
o 0
0o 0 aD 00
3.41283~~~~~~ as
0 0~
o °o O .° p 0
0000 0 0 0
0~~~~
0 0
0 0 0 °
00 ~ ~ ~ ~~0 0
o 00
3.41283 - SI
3.41283 6.04197
Inuvit
o 121nuvit A eqlnuv
6.04197 -0A
0 0
0 0 0
0 0 4
0 0 -~ 0
0 00 0 0 0 00
00 3 0 0000
00~~~~~~~~~~~~
0 ~~~ ~~ 0% o00 C o0
00 S oo 0 00
0 0 00 0000 00 0
0 , 0 00
0 0
0~~~~~~~1
Figure3 c
Persistency of the technology gap
7 year lags
o eqlnuv a 171nuvit
6.04197- E
00
a 0
b w Ag~~~~~~~~~~~~~~
_ A A
AOA
3.41283 0 o
3.41283 6.04197
Inuvit
It is not clear why we observe a sudden decline in the unit values of the machines in
Eastem European countries. A possible explanation of this factual evidence, which is
consistent with earlier findings based on the skdll index (Barba Navaretti, Galeotti and
Mattozzi, 2000) runs as follows. Eastern European countries used to buy most of their
machines within the Soviet Block They would onlyimport top technology machines from
Europe. The first years observed in our data may capture this earlier distortion. Once trade
was liberalised with the EU a huge geographical re-orientation of imports took place and
later most rachines were imported from Europe. Consequently the average quality of the
machines inported now is lower th it used to be.
4. Do embod1ed imported technologies boostptvductiv*y? Econometric analysis
We have shown some descriptive evidence that developing countries import
machines embodying simpler technobgies than industrialized ones. Does this matter for
productivity and income growth? As discussed in the introduction, earlier papers found that
11
imports have a favorable impact on growth because they act as a channel for R&D
spillovers. (Coe and Helpman, 1995, Coe Helpman and Hoffmaister, 1997, Keller, 2000).
The larger the R&DI) stock in the exporting country, the larger the spillovers induced by
imports and the larger the effect on productivity.
The descriptive evidence above shows that even if we consider one homogeneous
exporter, the EU, the quality of the machines exported varywith the importing country.
Thus, the same total expenditure on machines imported from the same exporter (the Eq
may comprise very different bundles of technologies. In this paper, we estimate the impact
of the technologies embodied in imported machines on total factor productivity. To single
out the effect of embodied technologies, we also control for the total expenditure on
imported machines.
Given that machines are sector specific, we will examine their impact on total factor
productivity in the industries using them as factors of production. Appendix 2 reports
industry matching. Appendix 3 describes the methodology we have folloved to construct
TFP. Our estimations are therefore carried out at the industry level (j), assuming that
industry values represent the behavior of the average firm in the industry. We have a panel
comprising 13 industries, six countries and 8 years (1989 to 1996).
As discussed in appendix 1, we assume that total factor productivity at time t
depends on lagged productivity and on a vector of technologies imported in the past. The
empirical version of the productivity function (A3), discussed in appendix 1, for sector j,
country c at time t can be represented as follows:
Ln(yrj,)=al +a2 (E Ln yr-r,) + a3LnIMPji,-l + a4LnUVI 0,-l + a5LnGDP, + a6Dj+a7Dc+a 8Dt + Cq,
r=l
(2)
where yj, is total factor productivity for industryj in the importing country c at time
t, the second term on the RHS is lagged TFP, the third term, IMP is the share of imported
machines on total investments in industry j, UVI is the unit value index as defined in (1)
which proxies the complexity of the machines imported, GDP measures the overall level of
12
development of country c at time t and D are industry country and time dummies
respectively.
The theoretical model assumes that firms are only facing the choice between two
alternative machines. Given that the unit value index is a continuous variable, in the
empirical analysis we assume that firms are facing an infinite number of options for
technologies embodied in a given machine. However, as discussed in the appendix,
productivity and imported technologies are simultaneously determined. On the one hand,
the use of more advanced technologies is expected to have a positive effect on productivity.
On the other hand, firms will buy advanced technologies if the expected effect on
productivity is positive. We consequently also need to analyze the choice of the embodied
technology, as a function, among other things, of expected productivity. In the appendix we
develop a model for the choice of technology. We leamn from the theory that this choice is
affected by past choices, relative factor prices, and the ability of importers to use high tech
technologies efficiently. This latter terms is composed of the level of productivity firms
manage to achieve immediately and the level of productivity they manage to achieve in the
future by learning to use a given technology. From equation (A8) in the appendix we can
derive an empirically implementable equation of the choice of technology.
LnUVIjt = 60 + 13 (E nUVI j_r) + 21Ln CJ+t + 63Lnycjt + f4Ln oCjAv +
2: n UI,,l-, ) fi21Pusit (1 + rct +)
+I35GDP,t +,/6Dj+/37DC+,I8Dt+vci (3)
where UTVI is the unit value index for the machines imported by sector j in country c
at time t. The first RHSE term is the lagged unit value index which captures the hysteresis
effect of lagged machines choices. The second term is a wage rental ratio which captures the
effect of relative prices: w is the average sectomal wage, r is the real inters rate 3 a fixed yearly
depreciation rate of 10% and PL+ is the price of the machines imported by the US in sector j
at time t (a proxi of the price of the top tech machines). The third term is TFP at time t for
industry j in country c. The fourth term is average yearlyproductivity growth (the learning
ability of the firm) which can be measured either directly or indirectly by looking at factors
that may affect the learning ability, like foreign investments. The fifth variable is GDP per
capita of country c at tirne t, which is a broad indicator of the level of development of
country c. We then have sector, country or timne dummies.
13
Equation (2) and equation (3) jointly define the system of equations to be empirically
estimated to analyze the link between imported technologies and productivity. However, we
face several econometric problems.
First, our results may be driven by spurious correlation, in that there may be
unobserved time-invariant factors affecting both productivity and the choice of technology.
One factor could be the share of foreign investors in the industry, or the degree of export
orientation. Although we may control for some of these variables, others may remain
unobservable. Second, as discussed in the theory there is persistence over time in both
productivity and choice of technology, which is not necessarily related to the learning
process associated to high tech machines. Third, there is an endogeneity problem arising
from this simultaneity between productivity and the choke of technology.
To eliminate the effect of time-invariant unobservable factors we carry out our
estimations in first differences. To isolate the impact of technological choices on
productivity from trend effects, we estimate both productivity and the choice of technology
on their lagged values. As for the endogeneity problem, we could in principle sort it out by
nurning a system of simultaneous equations, where productivity and technology are jointly
determined. However, given that we work in first differences, endogeneity also applies to
lagged variables and we would be left with a non sufficient number of exogenous variables
to run the system. A simpler altemative is to run two independent regressions, using the
appropriate lagged variables as instruments. The GMM-Instruxental Variable - GMM-IV -
mnethod developed by Arellano and Bond (1991) for dynamic panels is the right alternative to
the estimation of a simultaneous system of equations by three-stage least squares. The
endogeneity problem can be sorted out by using the appropriate lagged variables as
instruments.
We consequently transform equations (2) as follows:
ALn(ycjt) = aIALn(ycj.-1)+a2ALnUVIC - + a3ALnIMPc,_1 + a4ALnGDP,-. +a5Dt +AAsyt
(4)
and equation (3):
14
ALnUVICt =b1ALnUVIj1-1 +b2 ALn[ (q i -]+b3ALnrjy +b4ALnOPTcj, +b5ALnGDPct +
+b6Dt+Av1,j (5)
Besides for transforming (2) and (3) in first differences, we have only included one
lag for both productivity and UVI. In the technology choice equation we also substitute the
variable measunng the learning capacity of industryj with a variable measuring the
involvement of foreign firms in the industry, under the assumption that foreign investors
indirectly speed up the learning process with new technologies. As we do not have
consistent sector specific data on FDI, we measure them indirectly by looling at the share of
exports of sector j from country c which is classified as outward processing trade (OPI).
OPT captures flows of temporary trade between subcontractors and between parent
companies and subsidiaries.
Table 1: Determinants of TFP Growth
Dependent variable: DiffLn Total Factor Productivity
Rl R2
Lag Diff Ln Total Factor Productivity -0.32*** -0.45***
(-3.64) (-4.53)
Lag Diff Ln Unit Value Index 0.77*** 0.80***
(5.44) (4.99)
Lag Diff Ln Import Shares 0.06** 0.06***
(2.19) (1.74)
Lag Diff Ln Gross Domestic Product 1.74
(1.68)
N observations 90 90
Wald fioint) 29.98*** 33.05***
Wald (dummy) 76.85*** 23.95***
Sargan test: 16.13 15.42
AR (1) test -2.32** -2.46**
AR (2) test 0.91 0.74
The table includes only results from the second step of two-stage GMM-Instrunental Variables
estinates.
"Diff" indicates first-order differencing.
Time dummies are included in all the equations.
t values into brackets. ***99% significance, **95% significance, *90% significance
Transformed instruments are the second lags of all the explanatory variables. Level instnrments
include time dummies and the dependent variable at time 0 and time t- 1.
15
The results of our estimations are reported in table 1 for productivity and 2 for the
choice of technology. A detailed description of the variables used and their sources is
found in appendix 4. In general terms all regression perform well. The non significant
Sargan tests tell us that there is no over-identification in the instrument matrix used.
Disturbances are serially uncorrelated given that we find evidence of first order-auto-
correlation and no second order auto-correlation.
Table 2: Choice of Technolog
Dependent variable: DifLn Unit Value Index
R3 R4
Lag Diff Ln Unit Value Index -0.44**** -0.45****
(-13.3) (-14.1)
Diff Ln Wage Rental 1.76**** 0.77*
(5.47) (1.82)
Squared Diff Ln Wage Rental -0.17**** -0.08**
(-5.52) (-2.03)
Diff Ln Outward Processing Trade -6.89)
______________________________________ ~~~~(-9.89)
Squared Diff Ln Outward Processing Trade 6.27****
Diff Lu Total Factor Productivity (.60) (6.84)
Diff Ln Gross Domestic Product 2.501** 1.47
(3.00) (1.48)
N observations 333 333
Wald Goint) 364 4*** 462.8***
Wald (dunmm~ 308.5*** 295.6***
Wald (dummi ~~~* *
Sargan test: 33.2 30.4
AR (1) test -2.01** -1.78*
AR (2) test -0.18 -1.05
The table includes only results from the second step of two-stage GMM-Instrumental Variables
estimates.
"Diff" indicates first-order differencing.
Time dummies are included in all the equations.
t values into brackets.
****100% significance, ***99% significance, **95% significance, *90% significance
Transformed instruments are the first lags of all the explanatory variables. Level instruments include
time dummies and the dependent variable from lag 1 up to all the possible lags.
We first focus on the determinants of TFP growth. Regression 2 includes gross
domestic product as an explanatory variable. We find that the share of imported machines
on total investments in the industry has a positive impact on TFP. This result is in line
16
with findings in Coe and Helpman, 1995, Coe Helpman and Hoffmaister, 1997, Keller,
2000. However, we find that embodied technologies, as measured by the unit value index
of the machines imported have a larger effect on TFP and this variable is more robust to
different specifications of the model. If we look at regression 1 and we interpret the
coefficients as elasticities, we have that an increase in 10% in the rate of growth of the
unit value index generates an increase of 7.7% in the rate of TFP growth; in contrast the
same increase in the rate of growth of the import share generates an increase of 0.6% in
the rate of TFP growth. We also find, as expected, that GiDP growth has a positive effect
on sectoral TFP.
We now move to the choice of technology. We find that TFP has a positive and
significant effect on the Unit Value index, confirming our hypothesis of simultaneity
between the two variables. Firms buy high tech machines if they have enough skills to
use them in a sufficiently productive way. Factor prices have the expected effects,
although we find some non-linearity. The negative sign on the coefficient of the squared
variable, shows that growing wages induce firm to move to more capital intensive
machines, embodying more advanced technologies only when wages are relatively low.
When wages are high, any further increase has little effect on the decision to change the
technology of the machines used.
In regression 4 we include the variable OPT, capturing the involvement of foreign
firms. Also in this case we have a non-linear relationship. The share of high tech
machines increases only if the involvement of foreign firms in the industry is relatively
large. If foreign involvement is low, an increase in the role of foreign firrns has a negative
impact on the level of the technologies purchased. The reason for this result could be that
foreign firms may have a double effect in the choice of the technology. On the one hand
they may help locals to choose appropriate technologies. If foreign firms are only
marginally involved, this means that they also have limited requirements for high quality.
So, simpler machines could be more appropriate. When foreigners get more involved,
their demand for high quality increases and they consequently help locals in using more
complex machines. In any case to reach firm conclusions on the role of foreign investors,
we would need better data than OPT on FDI.
17
Finally, GDP growth has a positive impact on the choice of technology,
confirming the descriptive evidence reported in the previous section.
S. Conclusions
In this paper we explore the impact of imported technologies on productivity in
manufacturing sectors for a sample of developing and transition countries in Central and
Eastem Europe and in the Southem Mediterranean, which have recently deeply integrated
their economies with the European Union.
This paper departs from earlier studies of intemational technology diffusion as it
focuses on the technology embodied in the machines imported. The technological level of
the machines imported is proxied by an index relating the unit value of the machines
imported by a given country to the unit value of the same machines imported by the US. We
find very strong regularities in the pattem of imported machines. Unit values are generally
stable across time, except for countries facing dramatic shocks in the period observed, like
the Eastern European ones. Moreover, there is a constant and even increasing gap between
the unit value of the machines imported by the US and the machines imported by our
sample of developing countries. This reflects two inherent characteristics of technological
progress in the last decade. On the one hand, the price of machines is stable, independently
of technological progress. On the other hand, at anypoint in time, the prices of machines
differ according to productivity. Therefore, although with time developing countries import
machines which are increasingly more advanced, the gap us a us the technological leaders is
constant. We show that this gap is significantly persistent, and that it is higher the lower the
level of GDP of the importing country.
We also show that although the choice of developing countries to buy
cheaper and less sophisticated machines is optimal, given relative factor prices and their
endowmnents of technology, this choice has a negative effect on TFP growth at the industry
level. In other words productivity growth in nanufacturing depends positively on the type of
machines imported in a given industry. The effect of embodied technologies on growth is
18
found to be much more important than the effect of the share of imported machines on
total investments in the industry.
Our results partly confirm earlier findings that importing machinery is a fundamental
channel for productivity growth. Fundamental, but not sufficient. If, as the evidence clearly
shows, developing countries keep buying low tech machines, they get captured in a poor
technology low growth trap. If their productivity is low, they buy low tech nachines. But if
they buy low tech machines, they grow less. Thus the persistency of the technological gap is
worrying, even when the share of imported machines on total investments grows.
19
Appendix 1. The Analytical Framework
We develop a simple framework to analyze the choice of the imported technology at
the firm level and to estimate the impact of imported technologies on total factor
productivity which provides a useful background to the empirical analysis.
Dome titfactorpkiaizit
We assume the following Cobb-Douglas production function with constant elasticity
to scale for a generic firm i at time t:
Iar
Q;, = *i't Lt (Al)
Total factor productivity is therefore given by
)'- = K' (A2)
i,t i,t
We assume that total factor productivity at time t depends on a vector of
technologies imported at t and in the past Mit=(m,n, ...... mt-,), on exogenous shocks x,
and on lagged productivity6:
fi = (Yi,t l, -Yi,1-",Mit'Xit) (A3)
Mit includes all different activities that allow i to acquire foreign knowledge, like
importing foreign machines, exporting and developing relationships with foreign partners as
buyers, suppliers, subcontractors or shareholders. As argued in various recent studies
(Hoekman and Tybout, 1999, Djankov and Hoekman, 2000, Clerdies Lach and Tybout,
1998, Kraay, 1996), the main problem with the empirical specification of this productivity
function is that the choice of the foreign activity is endogenous, in that it depends on
expected productivity. We therefore need to jointly analyze and explicitly model the choice
of the foreign activities. In this paper we focus on the choice of imported machines, under
the assumption that the other foreign channels of learning affect productivity only indirectly,
by reducing the cost of acquiring more complex foreign machines.
Madin
There are two types of machines: low technology (L) and high technology (F ones.
When machines are used at full capacity, productivity is higher for high tech than for low
tech machines, i.e. YH > YL.
Full capacity output, q, is invariant with the type of the machine. TFP is higher when
high tech rachines are used as the amount of labour employed per unit of output q and
6 This general specification of productivity is derived from Hoekman and Tybout, 1998
20
capital k is lower, i.e. L,1 there is learning by doing, in that productivity increases at a time constant
rate +j. Thus learning takes place at a faster rate if ji is large. The ability of i to absorb new
technologies (absorptive capacity may depend on a series of factors, like the average
education of the work force, or i's access to other sources of foreign knowledge, like linkages
to foreign firms, exports etc.
TFP in high tech rachines reaches a maximum (full capacity YH,i,t = YH,*, (which is the same
across all times and finrs) after a number of periods ni.
n = 'rH 1
rYH,i,0 Oi
thus the number of periods required to achieve full capacity TFP depends on the
ratio between full capacity TNP and initial TIP when i first decides to use H and on the firm
specific learning rate j
The dynamics of TFP for high and low tech machines is reported in Figure 1.
21
Figure 1
Time path of total factor productivity for high and low tech machines
y
YH i
YH _____i______
y1
t
When i chooses a high tech machine, productivity could initially be even lower than for low
tech machines. It then gradually increase with time at a rate that will depend on the firrn's
ability to absorb the new technology. The larger the absorptive capacity (#i) the steeper the
YH,i,t line and the faster i reaches full productivity capacity.
7Te dxio jt &
Each firm i will choose machine(s) of type L or H so as to ensure profit
maximization. Given that productivity does not affect output per machine but just the
labour requirement per unit of machine-output and given constant returns to scale, profit
maximization is equivalent to minimizing production costs per unit of capital (machine).
End of period production costs with machines L are given by:
CL = PL(l + r) + w( ) (A6)
YL
'Under the assumption that machines are paid at the beginning of the period and wages at the end of it
22
where, LL = ( ), from (2),
YL
End of period production costs with machine(s) H is instead given by)
CH =PH(1+r)+w( q )Y /- JGTsFPi,+, (A7)
yRi,t rlI
where, L"H; = ( )., from (2), and the third term in 6 represents the permanent future
reductions in costs (increases in productivity) due to the learning by doing taking place in t.
In other words, using machines H at time t generates a learning process that permanently
increases productivity and reduces production costs. Explicitly, future costs reduction
following the choice of H in t are given by:
w 1~ 1_Y. /1 1 Y.
tS GTFPi,t+ =-wjq( 1 1 )]a = w[q - (p -1)] Y
r=l ' r YH,i,t-l YH,i,t r NHoi,t
Thus, H's cost function can be rewritten as:
CH =PH(1+r)+w( q --wqq-(1 i -1)] Y. (A7T)
The first two RI-IS terms in six represent the cost per unit of machine of producing with
machine H at t. It is lower the higher the level of TFP achieved by i at t. Future costs
reductions are larger the larger is the absorptive capacity of firm i (Q) and the lower the level
of productivity attainable at time t. They tend to zero as productivity approaches its full
capacity level and learning is completed. At this point production costs at t with H are at
their minimum level.
Note that these cost functions may generate a hysteresis process for two reasons.
First, because present production costs (the second RH-iS term in 6 and 6') keep declining
with time if firms keep using the same type of machines, until full capacity productivity is
reached. Second, because future benefits of choosing H at t (the third RHS term in 6 and 6')
can only be accounted for if i keeps buying high tech machines in the future. Therefore,
once a firm chooses high tech machines it does never move back to low tech ones.
To explicitly analyze the choice between L and H we derive the indifference price of
high tech machines (i.e. the price at which i is indifferent between machines of type L or of
type H) by equating (5) and (6'):
=PL + q ),_ q)Ya ] + w q (¢, -1)]o (A8)
1+ I+r 2L1 No r(1 + r) yH,i,t (A82
23
which can be re-written as:
Pi,*, =P +- [L, -L,] + (I q(oi- )] (A8')
i, +r r(1I +r) yH,j,t
Firm i will buy machines of type H whenever Pi*, > PH . The probability of buying a
high tech rachine will therefore be larger the higher the indifference price of high tech
machines.
The indifference price of H will be equal to the price of L, plus the short term
savings in labour cost of using H (second RI-I term of (7) and (7')) plus the the long term
benefits of using H (third RHS term of (7) and (7')). It will be higher if at t i has already
achieved a high productivity level (large short term benefits) or if its high absorptive capacity
generates a fast learning process (high long term benefits). Given that H is a labour saving
technology with a higher cost per unit of capital (machine) than L, higher wages and lower
interest rates make short term savings more valuable. Lower interests also increase the
discount factor and make future benefits more valuable. Consequently, the wage/interest
ratio has a positive effect on the indifference price.
24
Appendix 2
Table 1. Matching between machines and products
Machines Products
Harm. SITC/3 Description Nace ISIC rev.2 Description
8437/38(e 727 Food machinery, non domestic 411-423 311 Food
xcluding
84384)/79
84384/842 727 Food machinery, non domestic 424-28 313 Beverages
121/84212
2/8435
847810/90 72843 Tobacco working machines 429 314 Tobacco
8444-51 7244/5/6/7 Textile machinery 431-9 321 Textile
8452 7243 Sewing machines 453-6 322 Clothing
8453 7248 Skin, leather working 441-2/ 323+324 Shoes and leather
machines 451-2
84793/846 72812/72819/ Machine tools for working 461-7 331+332 Wood and wood funiture
5/6 72844 woods and wood treating
machines
8439/41 725 Paper etc mill machinery 471/2 341 Paper and Pap. Prods.
8440/2/3 726 Printing and binding machry 473 342 Printing
8456- 731/3/5 Machine tools for metal 312-9/321-8/ 381+382+3 Metal products and Machines
8463/8466 351-3/361-5 84 (incl transport excl electrical)
8454/5/84 737 Metalworking machinery 221-3 371 Iron and steel
68/8515 311
8475/8464 72841 Glass working machinery 247 362 Glass
2019
8477 72842 Rubber and plastic working 481-3 355/356 Rubber and plastic
machines
25
Appen&x 3: EmpiticalDeivation of TotalFactorProductivity
Measuring changes in total factor productivity.
The estimation procedures used are very straightforward. We assume that sectoral GDP (YI)
is produced using two factors, physical capital (4 and labor (L), using a Cobb-Douglas
production function:
(1) Yj, = Aj, (O)eAii (Kj,4tLjt'-'4'
where i indicates sector, Ai(0) represents initial conditions, Xi is the rate of technological
progress in sector i, ai measures the importance of physical capital in output, and 1- i the
importance of labor. After taking logs and differentiating with respect to time, we have:
(2) d In(Yjt) = Aj +ajd ln(Kj,)+(1 - aj)d ln(Ljt)
We estimated (2) by sectorj and timne t. We pooled data for all c countries in our sample,
added a time trend dunmmy (Dt) a country dummy (Dc), and, by country, a dummy for
periods of recession in the economic activity(DRci) which takes value 1 whenever Yqt