Download Download, PDF, 321 KB

Economic Geography and Industrialization
Holger Breinlich
University of Mannheim, CEP and CEPR
Alejandro Cuñat
University of Vienna
February 25, 2011
Abstract
We draw attention to two cross-sectional facts which are di¢ cult to account for in existing
models of industrialization. First, theories that stress the role of local demand in generating
su¢ cient expenditure on manufacturing goods are not well suited to explain the strong and
negative correlation between distance to the world’s main markets and levels of manufacturing activity in the developing world. Secondly, theories that emphasize the importance of
comparative advantage and are sceptical about the pro-industrializing e¤ects of high agricultural productivity are at odds with a positive correlation between the ratio of agricultural
to manufacturing productivity and shares of manufacturing in GDP. This paper provides a
potential explanation for these puzzles by nesting the above theories in a unifying framework
and introducing features of economic geography. A multi-location model with trade costs
is constructed in which industrialization is driven by the tension between access to markets
and comparative advantage patterns. Using comparative statics exercises and a fully ‡edged
multi-country calibration, we show that the model can replicate the above stylized facts.
KEY WORDS: Industrialization, Economic Geography, International Trade
JEL CLASSIFICATION: F11, F12, F14, O14
1
1
Introduction
One of the most striking aspects of economic development is the decline of agriculture’s share in
GDP and the corresponding rise of manufacturing, and later services. Economists have proposed
a number of theories to explain this transformation. Among the most in‡uential approaches are
explanations that focus on di¤erences in the income elasticity of demand across sectors (e.g.,
Rosenstein-Rodan (1943); Murphy et al. (1989b); Kongsamut et al. (2001)). As per-capita
income increases, non-homothetic preferences lead to a shift of demand from agriculture to manufacturing goods, thus increasing manufacturing’s share in GDP. Traditionally, these approaches
have analyzed closed-economy models and stressed the role of local demand. More recently, authors such as Matsuyama (1992) and (2009) have provided extensions to open-economy settings
and have shown that some key results of the closed-economy literature, such as the positive impact
of increased agricultural productivity on industrialization, can be reversed in such models.
The present paper draws attention to two cross-sectional facts which, taken together, are not
easily explained by either closed-economy or open-economy models of demand-driven industrialization. We argue that to understand these facts we need to move beyond the closed-versus-openeconomy dichotomy prevalent in the literature, and to consider multi-country settings in which
countries interact with each other through international trade, but in which bilateral interactions
are partly hampered (to a di¤erent extent across country pairs) by the fact that trade is not
costless.
Our …rst observation is that proximity to foreign sources of demand seems to matter for
industrialization. For example, it has long been noted that Hong Kong, Singapore, and Taiwan
not only bene…ted from an outward-oriented trade policy but also close proximity to the large
Japanese market (e.g., Puga and Venables (1996)). A cursory look at the data suggests that
distance to foreign markets has a more general relevance: Figure 1 plots the manufacturing share
in GDP against the minimum distance to the European Union, Japan and the US for a crosssection of developing countries in 2000.1 The …gure shows that developing economies close to one
of these main markets of the world show proportionally higher levels of industrialization.
Whereas this …rst fact suggests that interactions between economies are important, and thus
points to the relevance of open-economy models, our second fact seems to suggest the opposite: Figure 2 plots manufacturing shares against a standard proxy for comparative advantage
in agriculture, labor productivity in agriculture relative to manufacturing, for a cross-section of
developing countries for the year 2000.2 The …tted line has a positive, albeit statistically insignificant slope. As we show in our more detailed econometric analysis in Section 2, extending the
sample to include more countries and years leaves this positive correlation intact and actually
makes it statistically signi…cant as well. This is of course puzzling for open-economy theories of
1
We use the Netherlands as the approximate geographic centre of the European Union in Figure 1. Developing
countries are de…ned as countries belonging to the income categories “low”, “lower middle” and “upper middle”
published by the World Bank (corresponding to less than 9,265 USD in 1999). The simple OLS regression underlying
the …tted line in …gure 1 yields a negative slope statistically signi…cant at the 1%.
2
Developing countries are de…ned as in footnote 1. Relative productivy in agriculture vs. manufacturing is the
proxy of choice in many studies of Ricardian comparative advantage, e.g., Golub and Hsieh (2000).
2
industrialization such as Matsuyama (1992). If countries are indeed integrated through trade,
should we not expect them to specialize according to their comparative advantages?
We argue that both facts can be understood in a standard model of industrialization in
which di¤erences in the income elasticity of demand across sectors and relative productivities
are present and active. The key di¤erence of our approach in comparison with existing closedeconomy approaches is that we allow for a setting with many countries which are integrated
through trade. But, crucially, and in contrast to open-economy models such as Matsuyama
(1992), trade is not costless and geographic position is therefore important.
In our model, developing countries closer to foreign sources of demand will experience higher
demand for the agricultural and manufacturing goods they produce than more distant countries.
This will lead to higher wages in more central countries, which, because of the non-homotheticity
in demand combined with the presence of trade costs, will shift local production towards the manufacturing sector. Trade costs for agricultural product also hamper the comparative-advantage
mechanism put forward by free-trade models. High agricultural productivity leads to higher
wages which, again because of the combination of agricultural trade costs and non-homothetic
demand, leeds countries to specialize in manufacturing (we call this the ‘income e¤ect’of agricultural productivity). The standard comparative advantage e¤ect, which would drive specialisation
patterns in the opposite direction, is also present but is overcompensated by the income e¤ect
for the levels of trade costs revealed by the data.
Our paper relates to at least three sets of contributions in the literature. Most directly, we
contribute to the literature on industrialization that relies on di¤erences in the income elasticity
of demand across sectors for explaining structural change (“demand-driven industrialization”).
We add to this literature by drawing attention to the role of geography in shaping cross-sectional
patterns of industrialization. Similar to papers such as Murphy et al. (1989a, b), Matsuyama
(1992) or Laitner (2000) we focus on the initial shift from agriculture to manufacturing which
it is the key transition for the group of countries we are interested in in this paper, i.e., low- to
middle-income countries. That is, we do not model the services sector, whose share stays roughly
constant during the initial phase of development (see Chenery et al., 1986).
Consistent with our focus on explaining cross-sectional facts, we also disregard the dynamic
aspect of the industrialization process and rely on an entirely static model. In this respect, we
are similar to the contributions by Murphy et al. (1989a, b) but di¤erent from most other papers
in the literature. Our approach, however, avoids the criticism by Ventura (1997) and Matsuyama
(2009), among others, of closed-economy dynamic approaches to issues such as industrialization; namely that explaining cross-country patterns taking place in a globalized world on the
basis of dynamic closed-economy arguments can be quite misleading. In fact, we extend this
methodological criticism to the standard two-country, free-trade way of thinking about trade and
development: once one recognizes that bilateral distances and geographic position matter, one
must extend the model to many countries and allow for di¤erences in bilateral transport costs.
While our criticisms apply most directly to theories that approach the phenomenon of industrialization from the demand side, the stylized facts we have presented are not easily explained
3
by models that focus on supply side determinants either, be it contributions from the barriers-tomodern-growth literature,3 attempts to reconcile balanced (neoclassical) growth or connvergence
with structural transformations,4 or approaches from traditional international trade theory:5 the
ageographical nature of these approaches makes them unsuited to explain phenomena that have
an inherent geographic component, such as the ones described above. Thus, while not denying the
importance of these models and theories for many aspects of industrialization, this paper draws
attention to geographical proximity as a new and potentially important factor in explaining the
dramatic di¤erences in levels of industrialization across the world.6
Methodologically, our paper is closely related to recent work in international trade and economic geography which is interested in the e¤ects of comparative advantage and relative location
on trade ‡ows and production structures (e.g., Krugman (1980), Golub and Hsieh (2000), Davis
and Weinstein (2003), or Redding and Venables (2004), to name but a few). To the best of our
knowledge, however, the insights from this literature have never been applied in the light of the
aforementioned stylized facts, nor to the process of industrialization more generally.7
The remainder of this paper is organized as follows. Section 2 shows that the two correlations
displayed in Figures 1 and 2 also survive in a broader cross-section of countries, and are robust
to the inclusion of proxies for local demand. Section 3 develops a multi-country model with trade
costs. This model is used in Section 4 to shed light on the puzzles raised in the introduction. In
Section 5 we show that a fully calibrated version of the model is able to replicate our stylized
facts. Finally, section 6 concludes.
2
Empirical Evidence
In this section, we replicate the correlations from the introduction for a larger sample of countries
and years, and check their robustness to the inclusion of control variables. Our full econometric
speci…cation will be
ltshareMlt =
+ dt +
1 RPlt
+
2 CENlt
+ APlt + P OPlt + "lt ;
(1)
where RPlt is relative productivity (of agriculture to manufacturing) and CENlt the ‘centrality’
of country l, i.e. its access to foreign markets (to be de…ned below). APlt denotes agricultural
3
E.g. Parente and Prescott (1994) and (2000), Goodfriend and McDermott (1995), McGrattan and Schmitz
(1998), and Galor and Weil (2000).
4
See Caselli and Coleman (2001) and Ngai and Pissarides (2007), among others, as important examples in this
literature.
5
For example, Leamer (1987) and Schott (2003).
6
In a recent paper, Yi and Zhang (2010) share our concern that many aspects of industrialization cannot be
analyzed neither within a closed-economy setting nor under free trade. They analyze the e¤ects of changes in
productivity and declining trade barriers on production structures within a three-sector, two-country model, but
focus on dynamic rather than cross-sectional aspects of industrialization.
7
One exception are the papers by Puga and Venables (1996, 1999) who use a New Economic Geography model
to shed light on industrialisation in South-East Asia. Their focus, however, is on the implications of intermediate
goods usage for the forming of agglomerations and the sequential spread of industries across countries, rather than
on explaining cross-sectional facts related to industrialization. Also, the presence of multiple equilibria (absent
from the model presented below) makes a thorough empirical test of their theory much harder.
4
productivity, P OPlt the population size of country l, and dt is a full set of year …xed e¤ects.
The dependent variable is the logistic transformation of a country’s share of manufacturing value
added in GDP. We use a logistic transformation to account for the fact the manufacturing share is
limited to a range between 0 and 1.8 Concerning the regressors, we discuss the choice of suitable
empirical proxies in turn. Additional details on the data and their sources, as well as a list of
countries used in the regressions below are contained in Appendix B.
Keeping in line with existing studies on Ricardian comparative advantage (e.g., Golub and
Hsieh, 2000), we use labor productivity as a proxy for productivity. In contrast to total factor
productivity, this has the advantage of considerably increasing the number of available observations. We measure country l’s centrality (CENlt ) as the sum of all other countries’ GNP,
weighted by the inverse of bilateral distances, which are taken to proxy for transportation costs
between locations:
CENl =
X
GN Pj
distjl1 :
(2)
j6=l
This speci…cation re‡ects the basic intuition of our discussion. What matters is centrality in
an economic geography sense, that is proximity to markets for domestic products. Of course,
the above centrality index is nothing else but the measure of market potential …rst proposed by
Harris (1954), frequently used in both geography and - more recently - in economics. A number of
studies have demonstrated that this approach has strong explanatory power and yields results very
similar to more complex approaches that estimate trade costs from trade ‡ow gravity equations
(see, for example, Head and Mayer (2005), or Breinlich (2006)).9
As additional control variables, we also include agricultural productiviy (AP ) to account for
the pro-industrialising income e¤ect discussed above, and population size (P OP ) as an additional
proxy for the extent of the domestic market. We have data for all the required variables for up
to 120 countries for the years 1980, 1990, and 2000, yielding a total of 253 observations. Keeping
in line with the focus of this paper on the industrialization of developing countries, however,
we exclude high-income countries from our regression sample (although of course all available
countries are used to calculate the centrality measure).10
In columns 1-3 of Table 1, we separately regress the logistic transformation of manufacturing
shares on our proxies for comparative advantage (relative productivity, RP ) and centrality. Column 1 con…rms the positive correlation between comparative advantage in agriculture and levels
of industrialization, as measured by manufacturing’s share in GDP, which we demonstrated for
the year 2000 in Figure 2. Note that the coe¢ cient on RP is now signi…cant at the 5% level.
In column 2, we regress manufacturing shares on the log of minimum distance to the European
Union, Japan and the USA for comparison with Figure 1. Again, the slight change in speci…cation
and the use of a larger sample leave the negative correlation observed there intact and further
8
Using untransformed manufacturing shares instead does not change any of the qualitative results reported
below.
9
Using a nonstructural measure also seems to be better in line with the more explorative character of this
section.
10
We use the World Bank’s income classi…cation and exclude all countries with gross national income per capita
in excess of 9,265 USD in 1999 (“high income countries”).
5
increase its signi…cance. In column 3, we use our more sophisticated measure of centrality (2).
Note that we would now expect to …nd a positive and signi…cant sign, which is indeed what we
do.
In columns 4-6, we gradually build up our results to the full speci…cation (1). Three main
insights arise from these regressions. First, centrality has a positive and signi…cant in‡uence
throughout. Second, comparative advantage in agriculture either has a positive and signi…cant
e¤ect on industrialization or an insigni…cant one. In no case do we observe a negative and
signi…cant correlation. Third, comparative advantage is always insigni…cant once we control for
absolute agricultural productivity. This suggests that relative productivity might be picking up
the in‡uence of absolute productivity levels.
In column 7, we replace population by local GDP as an alternative proxy for the extent of
the local market. In column 8, we further drop agricultural productivity and replace it with
per-capita GDP. Per-capita GDP controls for the purchasing power of the local population, skill
levels, and other potentially confounding factors. Note, however, that it is very highly correlated
with agricultural productivity so that in practice both variables are likely to pick up the in‡uence
of similar omitted variables. The high correlation also makes the inclusion of both variables in the
same regression impossible.11 In any case, using GDP and GDP per capita instead of population
size and agricultural productivity does not a¤ect our qualitative conclusions.
Limited data availability for relative and absolute agricultural productivity prevents us from
estimating speci…cation (1) for a yet larger sample. In columns 9 and 10, we exclude these
variables, which increases the sample size more than tenfold since we can now use observations
for every year from 1980 to 2005. This allows us to provide some further results on the importance
of centrality for industrialization by running the following speci…cation:
ltshareMlt =
+ dt + dl +
1 CENlt
+
2 P CGDPlt
+
3 P OPlt
+ "lt ;
(3)
where P CGDPlt denotes per-capita GDP and dt and dl are a full set of time and country …xed
e¤ects. Column 9 of Table 1 reports results for an OLS regression pooled over the period 19802005 with year dummies only. Column 10 estimates the full speci…cation (3) by including country
…xed e¤ects, thus controlling for potential problems arising from unobserved time-invariant heterogeneity across countries. Both regressions give a similar picture as the results for the smaller
sample: both the size of the domestic market and access to foreign markets are strongly positively correlated with levels of industrialization. If anything, controlling for country speci…c
e¤ects implies an even stronger role for centrality.12
11
The correlations of the variables in logs is 87% in our sample
The downside of omitting relative and absolute agricultural productivity is of course that their exclusion is
likely to lead to omitted variable bias. To verify the likely magnitude of this bias, we estimated both (1) and (3)
on the same sample used for Table 1. Comparing the coe¢ cient on CEN in these regressions does indeed suggest
that omitting AP and RP leads to an upward bias, albeit a small one.
12
6
3
The Model
Consider a world with countries j = 1; :::; R, each with Lj consumers, each of which supplies
one unit of labor inelastically. There are two sectors, agriculture and manufacturing; we assume
perfect labor mobility between sectors, and no international labor mobility.
3.1
Demand Side
Preferences are identical across countries. Country-j individuals maximize a Stone-Geary utility
function over consumption of an agricultural and a manufactured composite good:
Uj =
ln(Mj ) + (1
) ln(Aj
A);
¯
(4)
where
Mj
Aj
=
=
"
"
R
X
(
1)=
mlj M
l=1
R
X
(
1)=
alj A
l=1
A
M
#
#
A =( A
A =( A
1)
;
(5)
1)
:
(6)
Both Aj and Mj are Armington aggregators of country-speci…c varieties: every country is assumed
to produce one di¤erentiated variety.13 Aj is consumption of the agricultural composite and alj
is the amount of the variety produced in l that is consumed by an individual consumer in j.
Similarly, Mj is consumption of the manufacturing composite and mlj is the amount of the variety
produced in l that is consumed by an individual consumer in j. The elasticities of substitution
between varieties are assumed constant at
A;
> 1. A denotes minimal consumption of
¯
agricultural goods, i.e. the subsistence level. These preferences guarantee that above the level
M
A the expenditure share of agricultural goods declines with rising per capita income. This is the
¯
so-called Engel’s law, which has strong empirical foundations (see Crafts (1980)). We assume
A< Al for all l, where Al is agricultural productivity in country l (to be de…ned more precisely
¯
below). This assumption guarantees that per capita income in each country is su¢ cient to reach
the subsistence level. Thus, at least some expenditure will be devoted to manufacturing products.
Below we impose enough structure so that labor is the only source of income. The individual’s
budget constraint in country j is therefore given by PAj Aj + PM j Mj = wj , with wj denoting
the wage in j, equal across sectors. PM j and PAj are price indices for the manufacturing and
agricultural composite goods. Prices paid for the di¤erent products in the importing location
j, pM lj and pAlj , consist of the mill price charged in country l plus industry-speci…c bilateral
transportation costs TljA ; TljM
A = T M = 1 for all j.) These transportation costs are of the
1. (Tjj
jj
iceberg-type form: for every unit of a good that is shipped from l to j, only 1=Tlj arrive while
13
The Armington assumption ensures that all countries consume all varieties provided transport costs and elasticities of substitution are …nite. This implies that all countries have diversi…ed production structures in equilibrium,
which renders the model relatively tractable below.
7
the rest “melts” en route.
Utility maximisation yields country-j individual’s demand for manufactured and agricultural
goods produced in l. Aggregating across individuals and countries, total demands (inclusive of
transport costs) for country-l goods are
mD
= pM l M
l
R
X
1
TljM
M
PMMj
1
EM j ;
(7)
j=1
aD
= pAl A
l
R
X
1
TljA
A
PAjA
1
EAj ;
(8)
j=1
where
PM j
=
R
X
p1M lj M
l=1
PAj
=
R
X
p1Alj
l=1
EM j =
A
!1
!1
1
M
=
1
A
=
"
"
R
X
pM l TljM
1
M
l=1
R
X
1
pAl TljA
A
l=1
#1
#1
1
M
;
(9)
1
A
:
(10)
PAj A) Lj and EAj = [(1
)wj + PAj A] Lj denote total expenditures on man¯
¯
ufacturing and agricultural goods in country j, respectively.
3.2
(wj
Production
Each country produces a di¤erentiated variety of the agricultural and the manufacturing good.
Sectors are perfectly competitive, operate under constant returns to scale, and use labor as the
only input. The amount of labor employed in agriculture in country l is denoted by LAl , and
supply of the local variety is al =
Al LAl .
The amount of labor employed in manufacturing
in country l is denoted by LM l , and supply of the local variety is ml =
M l LM l ,
where
Ml
denotes productivity in manufacturing in country l. Productivity levels are allowed to vary
across countries and sectors. Positive production implies f.o.b. prices equal the cost of producing
one unit of output: pAl = wl =
3.3
Al
and pM l = wl =
M l.
Equilibrium
D
Equilibrium in the agricultural and manufacturing goods markets requires al = aD
l and ml = ml .
This yields
al =
A
Al
wl
2
R
X
A 4
TljA
1
A
PAjA
1
j=1
ml =
M
M l wl
M
2
R
X
4
TljM
j=1
8
1
M
PMMj
3
EAj 5 ;
1
(11)
3
EM j 5 :
(12)
Labor demands in agriculture and manufacturing are, respectively,
LAl =
A
1
Al
wl
2
R
X
A 4
TljA
1
A
PAjA
1
j=1
LM l =
M
1
Ml
wl
M
2
R
X
4
TljM
1
M
3
EAj 5 ;
PMMj
1
j=1
(13)
3
EM j 5 :
(14)
Notice these are functions of the vector of wages of all countries. Full employment requires
LM l + LAl = Ll , which can be rewritten as
M
Ml
1
wl
M
2
R
X
4
TljM
j=1
1
M
PMMj
1
3
EM j 5 +
A
Al
1
wl
2
R
X
A 4
TljA
j=1
1
A
PAjA
1
3
EAj 5 = Ll ; (15)
These are R non-linear equations in the R wage rates, determining the vector of wages and
subsequently all other equilibrium variables of the model.14
To summarize, the crucial features of this model that will drive the results in the remaining sections are as follows. First, there are varying levels of agricultural productivity across
locations, which together with non-homothetic preferences will drive industrialization and deindustrialization through comparative advantage and Engel’s law. Second, positive transport
costs render relative geographical positions important, both by conferring a market size advantage to more central regions and by softening the impact of comparative advantage across space.
As the focus of this paper is on industrialization, we also introduce the share of manufacturing in
GDP as a further variable which in the following is also referred to as the level of industrialization
of a location:
M Sharel =
4
LM l
pM l ml
=
:
pM l ml + pAl al
Ll
(MS)
Analysis
This section analyses the properties of the model just developed and uses it to shed light on the
puzzles raised in the introduction. As will become clear, the present model nests many of the
existing approaches in the literature on demand-driven industrialization as the two special cases
of in…nite and zero transport costs (“closed economy” and “free trade”). We will demonstrate
that in the case with positive transport costs new results arise that help resolve our two puzzles.
14
By Walras’s Law, one of this R equations is redundant.
9
4.1
Closed Economy
With in…nitely high levels of transport costs, i.e. under autarky, it is easy to show that the
expression for the share of manufacturing in GDP simpli…es to
M Sharel =
1
A
¯
:
(M SAU T )
Al
As is apparent from equation (M SAU T ), the manufacturing share in GDP increases with agricultural productivity. Non-homothetic preferences (due to the positive subsistence consumption
level in agriculture, A> 0) are crucial for this result. Intuitively, the increases in per capita
¯
income resulting from higher values of Al lead to a decline in the share of subsistence consumption in total expenditure. As every unit of income above the subsistence level is spent in …xed
proportions on agricultural and manufacturing varieties, the expenditure share of the latter rises.
In a closed economy, this leads in turn to a shift of labor into manufacturing and an increase
in M Sharel . In the following, we will refer to this positive impact of agricultural productivity
on industrialization as the “income e¤ect” of agricultural productivity shocks.15 Very similar
e¤ects are obtained in the existing literature (e.g. Matsuyama (1992) or Murphy et al. (1989b)).
Needless to say, the autarky assumption renders any cross-country di¤erences in centrality or
comparative advantage completely irrelevant.
4.2
Free Trade
Under free trade, Ricardian comparative advantage emerges as an additional factor for the determination of the level of industrialization. With costless trade, it is easy to show that
LM l =LAl
=
LM l0 =LAl0
Since
M Sharel =
lower ratios
M l = Al
M l = Al
M l0 = Al0
M
1
:
LM l
1
=
Ll
1 + (LM l =LAl )
(16)
1;
(17)
imply a stronger bias towards agriculture in the production structures of
countries. As in standard Ricardian models of international trade, being relatively productive in
agriculture biases a country’s production structure towards agriculture, thus reducing the share
of manufacturing in GDP as the location specializes accordingly. We will refer to this as the
“comparative advantage e¤ect”. Notice that free trade eliminates any independent in‡uence of
the productivity level
Al
on industrialization via the non-homothetic preferences channel we
discussed above.
15
Obviously, the change in agricultural productivity also brings about a substitution e¤ect with it, as prices and
wages change and agricultural goods become relatively cheaper. As this e¤ect is overcompensated in the closed
economy version of the model, we do not stress it here. However, it plays a crucial role in the open economy, where
it determines changes in comparative advantage.
10
4.3
Costly Trade
In the presence of positive transport costs that are di¤erent across country pairs, the model
becomes much less tractable. We therefore simplify the model in a number of examples that
illustrate the new types of results our model can yield in this new environment.16
It is a long standing theoretical result in international trade theory that the size of the
home market matters for industrial structure (Krugman (1980), Krugman and Helpman (1985)).
Recently, Davis and Weinstein (1998) found empirical support for home market e¤ects in a
study on OECD countries. However, their …nding depended crucially on taking into account
demand linkages across locations, indicating the importance of foreign demand.17 In models of
industrialization, however, the role of access to foreign markets has been ignored so far, even
though its inclusion seems to be a logical extension of the existing literature. In a world with
positive transport costs, central locations have e¤ectively a larger market size as they are closer
to sources of demand, ceteris paribus. Note that this holds in addition to any size advantage the
domestic economy may have and depends on its position relative to other locations.
There are several theoretical reasons why one would expect central locations with better access
to foreign markets to have a higher manufacturing share than peripheral ones. First, being more
central raises demand for both agricultural and manufacturing goods and raises wages.18 With
non-homothetic preferences, this leads to an expansion of domestic manufacturing expenditure
which, with positive transport costs, will translate into a domestic manufacturing share higher
than in other countries, as the resulting increase in manufacturing expenditure will have a stronger
e¤ect on the domestic manufacturing good than on those produced by other countries.
Second, higher wages lead to higher prices of both types of goods. If agricultural goods
are more homogeneous than manufacturing goods (this would correspond to
A
>
M
in our
model), central locations will specialize in manufacturing, ceteris paribus. This is since demand
for locally produced manufacturing varieties will be less sensitive to higher prices than demand
for the location’s agricultural variety. The following particular case of our model illustrates this
mechanism: more central countries can bene…t from their position to industrialize even in the
absence of any technological or size advantage, simply because being more central raises demand
for the central country’s manufacturing good.
Consider a three-country world: R = 3. Assume all parameters are identical across countries
(except for the bilateral transport costs) and, in particular, that
and
M
Aj
=
Mj
= Lj = 1,
A
= 1,
positive but …nite. Consider a geographic structure such that country 1 takes a “central”
position while countries 2 and 3 are in the “periphery”: we model this by assuming that country
1 can trade freely with both 2 and 3 (T12 = T21 = T13 = T31 = Tjj = 1) and that countries 2 and
3 cannot trade with one another (T23 = T32 = 1).19 Transport costs are equal across sectors
16
In Section 5 we relax many of the drastic assumptions we make to produce these examples, and solve the model
numerically.
17
Indeed, in an earlier version of the same paper, Davis and Weinstein (1996) interpreted local demand as purely
domestic and ignored linkages across borders, and were unable to detect home market e¤ects.
18
See Redding and Venables (2004) for empirical evidence on the positive e¤ect of centrality on income levels.
19
We rule out the possibility that countries 2 and 3 can trade via country 1.
11
here. We take the agricultural good as the numéraire. Under incomplete specialization for all
countries, the labor market equilibrium conditions comprise equations
1
(1
3
LM 1 =
A) +
¯
1
=
(1
3
LM 2 = LM 3
(1
A) ;
¯
(18)
A) + (1
¯
2
A) :
¯
(19)
It is easy to show that in this case country 1’s manufacturing share is larger than that of countries
2 and 3, since LM 1 > LM 2 = LM 3 . If parameter values in this incomplete specialization scenario
yielded LM 1 > 1, then country 1 would specialize completely in manufacturing.20 In this case,
the labor market equilibrium conditions comprise equations
LM 1 = w1
M
2
(w1
4
LM 2 = LM 3 =
2+
w11
(w1
2 + w11
A)
+
¯
M
2 (1
1+
A)
+
¯
M
w11
(1
1 + w11
3
A) 5
= 1;
¯
M
A)
¯
(20)
< 1;
(21)
M
which imply w1 > w2 = w3 = 1.
A similar result obtains if the manufacturing good is subject to lower transport costs than
the agricultural good. Assume, as above, R = 3 and
now
A
=
M = TM =
T23
32
M =
TljA =
LM 1
LA1
LM 2
LA2
positive but …nite,
M
T12
=
M
T21
=
Aj
M
T13
=
=
Mj
M
T31
= Lj = 1. We assume
M = T A = 1, and
= Tjj
jj
1 for all j 6= l. In this case, the labor market equilibrium conditions imply
= [(1
= [(1
) + A]
¯
1
) + A]
¯
1
"
"
(w1 A)
¯
1
2w2 + w11
2 (w2 A)
+
¯
w11 + w21
(w1 A)
¯
1
2w2 + w11
(w2 A)
+
¯
1
w1 + w21
#
#
;
(22)
:
(23)
It is easy to see that LM 1 =LA1 > LM 2 =LA2 .
A third mechanism generating the home market e¤ect is based on the manufacturing industry
using increasing returns to scale (IRS) production techniques. In this case, central locations will
be the …rst, ceteris paribus, to reach the critical level of demand that makes IRS production
pro…table. This mechanism is absent from our model, as we assume constant returns to scale
across sectors.
5
A Calibrated Multi-Country Model of Industrialization
The discussion in Section 4 suggests that a thorough understanding of the e¤ects of distance on
industrialization patterns requires the modeling of bilateral transport costs in a many-country
20
Under the assumption
agricultural good.
A
= 1, there is no need for every country to produce its own “variety” of the
12
setup. In this section we endeavor to achieve this goal by presenting a fully calibrated multicountry version of the model of Section 3. We use this model to generate a synthetic dataset on
manufacturing shares and centrality, and use the simulated data to try to replicate the correlations
from Section 2.
5.1
Parameter Values
For a calibrated version of our model, we need values for the parameters governing substitution
elasticities (
A,
M ),
productivity levels (
Al ,
M l ),
trade costs (TljA , TljM ), the manufacturing
expenditure share ( ), and subsistence consumption (A). Table 2 provides parameter estimates
¯
and a brief description of the calibration procedure and data sources used. In the following, we
describe the calibration in more detail.
We use values for
M
and
A
based on existing estimates in the literature. A …rst set of
relevant results are those by Broda, Green…eld and Weinstein (2006) who estimate elasticities
of substitution between varieties of goods produced in each of approximately 200 sectors. They
report estimates for 73 countries, yielding well over 10,000 individual estimates. The median
across these estimates for the 60 countries also present in our data is 3.4. Given the much higher
degree of aggregation in our data (two instead of 200 sectors), we take
M
=
A
= 3:4 as an
absolute upper bound. Closer to our level of aggregation is the estimate by Eaton, Kortum
and Kramarz (2008) who use French …rm-level data to estimate an elasiticity of substitution
between individual manufacturing varieties of
M
=
A
M
= 1:7. In the light of these estimates, we set
= 2 in our baseline calibration but also report results for elasticities of substitution of
1:5, 2:5,and 3 in our robustness checks. Note that throughout, we assume
to bias our results in favor of …nding a home market
M
=
A
in order not
e¤ect.21
Since labor is the only factor of production in our model, we proxy
Al
and
Ml
by labor pro-
ductivity in agriculture and manufacturing, respectively. However, the cross-sectional variation
in labor productivity across countries which we observe in the data is driven by both di¤erences
in technological e¢ ciency and di¤erences in prices. That is, lp:l = V A:l =L:l = p:l x:l =L:l in terms
of our model because we abstract from intermediate inputs. Since we are only interested in
:l
= x:l =L:l , we use data on purchasing power parities for agriculture and manufacturing goods
consumption from the International Comparison Program (ICP) to construct proxies for p:l and
strip out price variation from the data (see Appendix A for details).22
Estimates of the trade cost matrices can be obtained via gravity equation regressions using
manufacturing and agricultural trade data. To see this, note that exports in the model are:
XljA = p1Al
XljM
1
= pM
l
A
M
TljA
TljM
21
1
A
1
PAjA
M
1
PMMj
EAj ;
1
EM j :
(24)
(25)
We have also used the estimates by Broda, Green…eld and Weinstein (2006) to separate HS sectors belonging to
the agricultural sector (HS010 to HS0150) from those in the manufacturing sectors (HS0160 to HS0961). Computing
median elasticities separately for both sectors yields A = 3:7 and M = 3:4, so that our assumption A = M is
a conservative one.
22
Echevarria (1997, section 5.5) uses a similar approach based on US data.
13
The only bilateral variable on the right-hand side in the above expressions is trade cost Tlj . We
proxy for these costs by Tlj = distlj1 e
countries l and j, and
1
2 dint;lj
, where distlj denotes the bilateral distance between
denotes the elasticity of trade cost with respect to distance. The dummy
variable dint;lj indicates if a trade ‡ow crosses national borders (i.e., dint;lj = 1 if l 6= j and 0
if l = j). This is a parameterisation of trade cost which is common in the international trade
literature (e.g., Wei (1997)). Proxying all other variables by importer and exporter …xed e¤ects
and adding an error term, we can rewrite bilateral exports as
XljA = dexp;A
dimp;A
XljM
dimp;M
= dexp;M
(1
distlj
A ) A1
(1
distlj
e(1
M ) M1
A ) A2 dint;A
e(1
"lj;A ;
M ) M 2 dint;M
"lj;M :
(26)
(27)
We estimate (26) in its original multiplicative form via Poisson QMLE, using data from the
sources listed in Table 2 and following Wei (1997) in proxying internal trade ‡ows as domestic
production minus exports. As has been pointed out by Santos-Silva and Tenreyro (2006), Poisson
QMLE can accomodate zero trade ‡ows, which are common in our data, and leads to consistent
parameter estimates even in the presence of heteroskedasticity in "lj . In robustness checks below,
we will also use estimates of Tlj obtained via estimating a log-linearized version of (26) via OLS.
Appendix B contains details of the estimation results. The distance coe¢ cients in the above
equations provide estimates for (1
estimates for
)
which, together with our assumed values of
and thus for Tlj = distlj1 e
2 dint;lj
.
We again use price and expenditure data from the ICP to estimate
manufacturing and agricultural expenditure shares are
SEM j
=
SEAj
=
, yield
EM j
=
wj Lj
EAj
= (1
wj Lj
and A. From the model,
¯
PAj
A
;
¯ wj
(28)
PAj
)+ A
:
¯ wj
(29)
The ICP provides cross-country data on prices and nominal expenditure on food and manufacturing goods which can be used as proxies for PAj , SEM j , and SEAj . We further approximate
wj by GDP per worker, which is consistent with our theoretical model. Estimates for
and A
¯
can than be obtained by regressing either expenditure share on the corresponding right-hand side
variables in the above equations. (Both equations yield identical results because we normalize
expenditure data such that SEM j + SEAj = 1 in accordance with our model.) Alternatively,
note that as wj ! 1, SEM j !
; and as wj !ĀPAj , EAj = (PAj Lj ) !Ā. So as an addi-
tional robustness check in Section 5.2., we use the normalized manufacturing expenditure share
(SEM j = (SEM j + SEAj )) of the richest country in our data as a proxy for . Likewise, we use the
real expenditure per worker of the poorest country as a proxy for A.
¯
14
5.2
Results
We now present results for the same regressions as in Table 1, but this time we use simulated
rather than actual data for the year 2000. That is, we use the calibrated model to generate
arti…cial data on manufacturing shares for the 74 countries in our simulation sample. We then
regress these generated manufacturing shares on the same variables as in Table 1. Intuitively, if
the true data generating process for manufacturing shares is similar to the one postulated by our
model, we should expect to …nd comparable results from both sets of regressions.23
Since the data availability for international price data limits our simulation analysis to the
year 2000, we start by replicating the regressions from Section 2 for the smaller sample of 74
countries used in the simulations. The results are presented in Table 3, and show that the
qualitative patterns present in Table 1 carry over to the smaller regression sample. In particular,
centrality is positive and signi…cant throughout, and comparative advantage never has a negative
and signi…cant sign.
In Table 4, we use our generated data to run the same regressions as in Tables 1 and 3. The
pattern of results is qualitatively similar across these tables. The coe¢ cient on market potential
is positive and signi…cant in all speci…cations in Table 4. Likewise, relative productivity is never
signi…cantly negative. Similar to Table 1, it has a positive impact on industrialization in columns
1,3 and 7, but loses its signi…cance as soon as we control for agricultural productivity. Thus, we
replicate the basic …ndings highlighted in the introduction and in Section 2.
Tables 5 and 6 report a number of robustness checks. In Table 5 we let the elasticities of
substitution vary between
M
=
A
= 1:5 and
M
=
A
= 3. For each scenario, we present
results for the key regressions from Table 4 (…rst, including both relative productivity and market
potential; and second, including agricultural productivity in addition). Results for the other
columns are similar to before, and omitted here for the sake of brevity.
The pattern that emerges from Table 5 is that the impact of relative productivity becomes
more signi…cant as the elasticity of substitution increases. Intuitively, a higher
implies that the
relative demand for agriculture versus manufacturing goods becomes more sensitive to changes
in relative prices, which in turn are determined by relative productivities. Once we reach
A
M
=
= 3, our initial results from Table 4 are reversed, in the sense that comparative advantage is
now signi…cantly negative. Note, however, that an elasticity of substitution of three is already
very close to the value which we consider as an absolute upper bound (
M
=
A
= 3:4). For
empirically more plausible estimates, our model does indeed replicate the key features of our
earlier result. An interesting and initially not obvious implication of our results is that changes
in tastes leading to higher elasticities of substitution will help countries industrialize.
In Table 6, we report a number of additional robustness checks. The …rst two columns use
23
Our model also generates data for per-capita GDP (equal to wages in our model), GDP (wages times population
size) and market potential (calculated according to (2), using the same distance data but replacing GNP with
model generated GDP). We use these generated data in the regressions below consistent with the notion that we
would like to evaluate whether our model is comparable to the data generating process in the real world. Note
that population size and productivity data are parameters of the model, and identical to the data used in the
regressions from Section 2.
15
ordinary least squares to estimate equation (26), leading to the alternative estimates for
1M ,
2A ,
and
2M
.24
1A ,
In columns 3-4, we use the manufacturing expenditure share of the richest
country in our data as a proxy for , and real expenditure per worker of the poorest country as
a proxy for A, as described above. Finally, in columns 5-6 we use producer prices rather than
¯
consumer prices to de‡ate relative productivities (see Appendix A). As shown, none of these
changes alters the basic message from Table 4. Market potential is positive throughout and
signi…cant with only one exception, and relative productivity is either positive and signi…cant or
insigni…cant; it is always insigni…cant when controlling for absolute agricultural productivity.
In Table 7 we compare our preferred calibration with positive but …nite trade cost to the cases
of free trade and autarky. As already noted, most of the existing models in the literature are
based on either one or the other of these two polar scenarios. Our comparison uses two criteria.
First, can the model replicate the qualitative correlations found in the data between comparative
advantage and market potential, on the one hand, and manufacturing shares, on the other hand?
Second, how well do all three parameterisations do in terms of replicating actual manufacturing
shares? To evaluate this second criterion, we regress actual on simulated manufacturing shares,
and look at the sign and signi…cance of the corresponding regression coe¢ cient, as well as at the
associated R2 .
Looking at free trade …rst, we see that the model’s performance in this case is dismal with
respect to both criteria (see columns 3-4; columns 1-2 replicate our baseline results for convenience). The coe¢ cient on comparative advantage is negative and strongly signi…cant, whereas
the one on market potential is slightly negative and insigni…cant. The regression coe¢ cient from
the regression of actual on simulated manufacturing shares is positive but insigni…cant, and the
corresponding R2 close to zero (see the last two lines of the table). The model’s performance
with in…nitely high trade costs is somewhat better (column 5-6). As our preferred parameterisation, the model under autarky can replicate the facts related to relative productivity. However,
market potential either takes on a negative sign or is insigni…cant. Also, while the correlation
between actual and predicted manufacturing shares is positive and highly signi…cant, and the
R2 substantially higher than with free trade, both measures are lower than the ones generated
by our baseline parameterisation (see column 1-2). We conclude that allowing for positive but
…nite trade cost is necessary to replicate the stylised facts discussed in the introduction, and also
improves the …t of actual and predicted levels of industrialization.
6
Conclusion
In this paper, we have drawn attention to two cross-sectional facts which, taken together, are
not easily explained by existing models of models of industrialization. First, proximity to foreign
sources of demand seems to matter for levels of industrialization. That is, there is a positive
correlation between manufacturing shares and the ‘centrality’ of a country, i.e., its closeness to
24
This alternative procedure yields
of the estimation results.
1A
= 4:68,
1M
= 2:61,
16
2A
= 1:29,
2M
= 1:45. See Appendix B for details
foreign markets for its products. Second, a standard proxy for Ricardian comparative advantage
(labor productivity in agriculture relative to manufacturing) is positively or not at all correlated
with manufacturing shares.
We have argued that to understand these facts, we need to move beyond the closed-versusopen-economy dichotomy prevalent in the literature, and to consider multi-country settings in
which countries interact with each other through international trade, but in which interaction
is partly hampered by the fact that trade is not costless. We constructed a simple model along
these lines and used comparative statics exercises and a fully ‡edged multi-country calibration
to show that it can replicate our stylized facts.
References
[1] Abegaz, B. (2002): “Structural Convergence in Manufacturing Industries between Leaders
and Latecomers”, Journal of Development Studies, 69-99.
[2] Behrens, Kristian, Lamorgese, Andrea R., Ottaviano, Gianmarco I.P., and Takatochi
Tabuchi (2004): “Testing the Home-market E¤ect in a Multi-country World: the Theory”,
mimeo, CORE, Université Catholique de Louvain.
[3] Breinlich, H. (2006), “The Spatial Income Structure in the European Union – What Role
for Economic Geography?”, Journal of Economic Geography, Nov. 2006, Vol. 6, No. 5.
[4] Caselli, F. and Coleman, W.J. (2001): “The U.S. Structural Transformation and Regional
Convergence: A Reinterpretation,” Journal of Political Economy, 109, 584-616.
[5] Chenery, H., Robinson, and M. Syrquin (1986): “Industrialization and Growth”, World
Bank.
[6] Ciccone, A. (2002): “Input Chains and Industrialization”, Review of Economic Studies, 69,
565-587.
[7] Combes, P.P. and M. Lafourcade (2003): “Core-periphery Patterns of Generalized Transport
Costs: France, 1978-1998”, CEPR discussion paper no. 3958.
[8] Crafts, N. F. R. (1980): “Income Elasticities of Demand and the Release of Labor by Agriculture during the British Industrial Revolution: A further appraisal”, Journal of European
Economic History, 9, 153-168; reprinted in J. Mokyr (Ed.), “The Economics of the Industrial
Revolution”, Rowman & Allanheld, Totawa, NJ, 1985.
[9] Davis, Donald R. (1998): “The Home Market, Trade, and Industrial Structure”, American
Economic Review, 88(5), 1264-1276.
17
[10] Davis, Donald R. and David E. Weinstein (1996, 1998): “Market Access, Economic Geography and Comparative Advantage: An Empirical Assessment”, NBER Working Papers 5706
and 6787.
[11] Davis, Donald R. and David E. Weinstein (2003), “Market Access, Economic Geography, and
Comparative Advantage: An Empirical Assessment”, Journal of International Economics,
59(1), pp. 1-23.
[12] Deardor¤, Alan V. (2004): “Local Comparative Advantage: Trade Costs and the Pattern of
Trade”, Discussion Paper No. 500, University of Michigan.
[13] Echevarria, Christina (1995): “Agricultural Development vs. Industrialization: E¤ects of
Trade”, Canadian Journal of Economics, 28(3), 631-647.
[14] Evans, C. L. (2003): “The Economic Signi…cance of National Border E¤ects”, American
Economic Review, 93(4), 1291-1312.
[15] Fafchamps, Marcel and B. Helms (1997): “Local Demand, Investment Multipliers, and Industrialization: Theory and Application to the Guatemalan Highlands”, Journal of Development
Economics, 49, 61-92.
[16] Feenstra, R., R. Lipsey and H. Bowen (1997): “World Trade Flows, 1970-1992, With Production and Tari¤ Data”, NBER Working Paper No. 5910.
[17] Feenstra, R. (2000): “World Trade Flows, 1980-97”, University of California, Davis, mimeograph.
[18] Fleming, J. Marcus (1955): “External Economies and the Doctrine of Balanced Growth”,
Economic Journal, 65, 241-56.
[19] Fujita, M., P. Krugman and A.J. Venables (1999): “The Spatial Economy: Cities, Regions
and International Trade”, MIT Press, Cambridge MA.
[20] Gallup, J.L. (1998): “Agriculture Productivity and Geography”, mimeo, Harvard Institute
for International Development.
[21] Gallup, J.L. and J.D. Sachs, with A. Mellinger (1999): “Geography and Economic Development”, CID Working Paper No.1, Harvard University.
[22] Galor, O. and D. Weil (2000): “Population, Technology, and Growth: From Malthusian
Stagnation to the Demographic Transition and Beyond”, American Economic Review, 90,
806-828.
[23] Gollin, Douglas, Parente, Stephen L. and Richard Rogerson (2000): “Farm Work, Home
Work, and International Productivity Di¤erences”, mimeo, University of Pennsylvania.
18
[24] Golub, Stephen S. and Hsieh, Chang-Tai (2000): “Classical Ricardian Theory of Comparative
Advantage Revisited”, Review of International Economics, 8(2), 221-234.
[25] Goodfriend, M. and J. McDermott (1995): “Early Development”, American Economic Review 85, 116-133.
[26] Harris, J. (1954): “The market as a factor in the localization of industry in the United
States”, Annals of the Association of American Geographers, 64, 315-348.
[27] Head, K. and T. Mayer (2005): “Regional Wage and Employment Responses to Market
Potential in the EU”, CEPR Discussion Paper 4908.
[28] Kongsamut, P., S. Rebelo and D. Xie (2001): “Beyond Balanced Growth,” Review of Economic Studies, 68, 869-882.
[29] Krugman, Paul R. (1980): “Scale Economies, Product Di¤erentiation, and the Pattern of
Trade”, American Economic Review, 70, 950-959.
[30] Krugman, Paul R. and E. Helpman (1985): “Market Structure and Foreign Trade”, MIT
Press, Cambridge, MA.
[31] Laitner, John (2000), “Structural Change and Economic Growth”, Review of Economic
Studies, 67, 545-561.
[32] Leamer, Edward E. (1987): “Paths of Development in the Three-factor, n-good General
Equilibrium Model”, Journal of Political Economy, 95(5), 961-999.
[33] Matsuyama, K. (1992): “Agricultural Productivity, Comparative Advantage, and Economic
Growth”, Journal of Economic Theory, 58, 317-334.
[34] Matsuyama, K. (2009): “Structural Change in an Interdependent World: A Global View
of Manufacturing Decline”, Journal of the European Economic Association, Papers and
Proceedings.
[35] McGrattan, E. and J. Schmitz (1998): “Explaining Cross-country Income Di¤erences”, Sta¤
Report 250, Federal Reserve Bank of Minneapolis.
[36] Murphy, Kevin M., A. Shleifer and R.V. Vishny (1989a): “Industrialization and the Big
Push”, Journal of Political Economy, 97(5), 1003-1026.
[37] Murphy, Kevin M., A. Shleifer and R.V. Vishny (1989b): “Income Distribution, Market Size,
and Industrialization”, Quarterly Journal of Economics, 104(3), 537-564.
[38] Ngai, L.R. and C.A. Pissarides (2007): “Structural Change in a Multisector Model of
Growth,” American Economic Review, 97, 429-443.
[39] Parente, Stephen L. and Edward C. Prescott (1994): “Barriers to Technology Adoption and
Development”, Journal of Political Economy, 102, 298-321.
19
[40] Parente, Stephen L. and Edward C. Prescott (2000): “Barriers to Riches”, MIT Press,
Cambridge, MA.
[41] Puga, D. and A.J. Venables (1996): “The spread of industry: spatial agglomeration in
economic development”, Journal of the Japanese and International Economy, 10, 440-464.
[42] Puga, D. and A.J. Venables (1999): “Agglomeration and Economic Development: Import
Substitution vs. Trade Liberalisation”, Economic Journal, 109, 292-311.
[43] Rauch, J.E. (1999): “Networks versus Markets in International Trade”, Journal of International Economics, 48(1), 7-35.
[44] Redding, S. and A.J. Venables (2004): “Economic Geography and International Inequality”,
Journal of International Economics, 62, 53-82.
[45] Rosenstein-Rodan, Paul N. (1943): “Problems of Industrialization of Eastern and SouthEastern Europe”, Economic Journal, 53, 202-211.
[46] Schott, P. (2003): “One Size Fits All? Heckscher-Ohlin Specialization in Global Production”,
American Economic Review, 686-708.
[47] Skott, P. and J. Ros (1997): “The ‘Big Push’in an Open Economy with Nontradable Inputs”,
Journal of Post Keynesian Economics, 20(1), 149-162.
[48] Syrquin, M. and H. Chenery (1989): “Patterns of Development, 1950 to 1983”, World Bank
Discussion Paper.
[49] Temple, J. and H.J. Voth (1998): “Human Capital, Equipment Investment, and Industrialization”, European Economic Review, 42, 1343-1362.
[50] Ventura, J. (1997): “Growth and interdependence,” Quarterly Journal of Economics, 112,
57-84.
[51] Yi, K.M. and J. Zhang (2010): “Structural Change in an Open Economy,” mimeo, Federal
Reserve Bank of Philadelphia.
20
A
Appendix A: Using ICP Data to Proxy for Prices
In Section 5 we use data from the International Comparison Project (ICP) to strip out price
variation from measured productivity. To understand this approach, note that the ICP provides
data on a number of expenditure categories in both current US dollars and so-called international
dollars ($I). One 1$I is the amount of goods and services one US dollar would purchase in the USA
in the base period (2005 in our case as no data were available for 2000). Converting expenditure
from current US dollars into $I thus removes any price di¤erences across countries and basically
converts expenditures into quantities using suitable aggregators. By comparing local expenditures
in US dollars and international dollars, one can derive country-speci…c PPP exchange rates which
capture price di¤erences across country. For example, per-capita expenditure on food in current
US dollars was $2,040 in 2005 in the United Kingdom but only 1,586 $I, yielding an implicit price
of 1.29 (the price in the USA is normalized to 1). Dividing measured productivity levels by this
price as converts them into quantities per unit of labor used (i.e., proxies for xl and zl , rather
than pM l xl and pAl zl ). We note that Echevarria (1997, section 5.5) uses a similar procedure,
calculating proxies for agricultural and manufacturing prices by dividing expenditures in U.S.
dollars by expenditures in international dollars.
One problem with the above approach is we are implicitly using consumer prices rather than
producer prices to de‡ate production. In terms of our model, ICP prices are proxies for PAl and
PM l , not pAl and pM l . As a robustness check in section 5, we therefore use the de…nition of PAl
and PM l to extract information on pAl and pM l in a model-consistent way. In our model,
PM j
=
" R
X
pM l TljM
1
M
l=1
PAj
=
"
R
X
1
pAl TljA
l=1
A
#1
#1
1
M
(30)
1
A
;
(31)
Together with data on the elasticities of substitution and trade costs which we have obtained
independently as part of our calibration strategy, we can solve the above system of equations
for pAl and pM l . In practice, consumer and implied producer prices are almost identical, with a
correlation coe¢ cient of above 99% and a level di¤erence of on average less than 4% for manufacturing and less than 1% for agriculture.
21
Figures and Tables
.4
Figure 1: GDP manufacturing shares and minimum distance to main markets (2000)
PRI
THA TJK
Manufacturing share (%)
.1
.2
.3
BLR CHN
KOR
CZE
MYS
IDM
DOM
0
CRI
SVK
SLV
ARM
HUN
TURHND
PHL
CIV
CMR
MEX
HRV MKD
VEN
KGZ
CHL
EGY
LTU UKR
SYC
ZAF
VNM
POL BGR
TUN
EST
KAZ
ARG
MAR NIC
BRA
LAO
KHM
LKA
MDA
ZWE
JOR COL BFABGD PER
IND
PRY
BOL
WSM
PAK
SEN
ROM
URY
FJI
LVA
IRN
GEO
LBN
MWINAM MDG
MOZ
LSO
KEN
ALB
ZMB
ERI
BLZ
JAM
TKM GNB
BIH
KNA PAN
SAU
LBR
UZB
NPL
CPV
MRT
SUR
GHA
TCD
DMA
BDI
BTNBEN
TGO
GUY SDN
SLB
UGA TZA
PNG
DZA
TTO
MMR
CAF
RWA
NER
SYR
GRD
BRB
VCT
AZE
ETH
GMB
OMN
YEM
LCA
TON
KIR
ZAR
MNG
COMBWA
VUT
MLI GIN
SLE GAB COGAGO
DJI
ATG
PLW IRQ
GNQ
0
2000
4000
6000
Minimum distance
8000
10000
Notes: Figure plots manufacturing shares in GDP (in %) against the minimum distance (in km) to either of the
U.S., the European Union (Netherlands), or Japan. All data are for 2000. See appendix for data sources and
country codes.
.4
Figure 2: GDP manufacturing shares and relative productivities (2000)
Manufacturing share (%)
.1
.2
.3
MYS
THA
CHN
IDM
KOR
MEXCMR
CHL
VNM
SVK
ARM
MKD
HRV
VEN
EGY KGZ LTU
UKR
ZAF
POL
ARG EST NIC
MARKAZ
KHM BRA
LKA
MDA
ZWE PER
JOR
IND
PRY
COL
BOL
BGD
PAKURY
ROM
LVA
IRN
GEO
NAM
KENZMB LSO
ALB
BLZ
JAM
SAU PAN SUR
NPL
UZB
GHA
SDN
GUY
PNG
DZA
TZA
TTO
NER
SYR
BRB
VCT
AZE
ETHOMN YEM
LCA
MNG
BWA
BGR
0
ERI
CZE
DOM
CRI
SLV
HUN
HND
TUR
PHL
-4
-3
-2
-1
0
Labour prod. agriculture/manufacturing (logs)
1
Notes: Figure plots manufacturing shares in GDP (in %) against the ratio of labor productivity in agriculture and
manufacturing. All data are for 2000. See appendix for data sources and country codes. Section 2 provides
further discussion on the productivity measures.
30
Table 1: Empirical Results
Regressor
log(RELPR)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
0.0520
(0.0506)
-0.0434
(0.0553)
0.0119
(0.0525)
0.0664
(0.0562)
0.0797*
(0.0459)
0.461***
(0.159)
0.337**
(0.153)
0.187***
(0.0661)
0.289**
(0.122)
0.214***
(0.0557)
0.181***
(0.0241)
0.297**
(0.121)
0.0296
(0.0607)
0.307**
(0.123)
0.374***
(0.128)
2.712***
(0.772)
0.184***
(0.0218)
0.897***
(0.263)
0.188***
(0.0240)
0.0105
(0.0560)
0.213***
(0.0435)
0.231***
(0.0728)
0.105**
(0.0459)
log(mindist)
-0.408***
(0.0941)
log(CEN)
0.523***
(0.138)
log(AP)
log(POP)
log(GDP)
0.184***
(0.0239)
log(PCGDP)
Fixed Effects
Year
Year
Year
Year
Year
Year
Year
Year
Year
Observations
R-squared
253
0.038
253
0.115
253
0.083
253
0.090
253
0.158
253
0.382
253
0.390
253
0.389
2,950
0.364
Year,
Country
2,950
0.881
Notes: Table displays coefficients and robust standard errors (clustered on countries) for OLS estimations. The dependent variable is the logistic transformation of
a country’s share of manufacturing in GDP. RELPR is the quotient of a country’s agricultural labor productivity and its labor productivity in manufacturing.
Mindist is the minimum distance (in km) of a country to either of Japan, the European Union (Netherlands) or the USA. CEN is a country’s centrality measure
(defined in Section 2). POP is a country’s population size, AP its labor productivity in agriculture and GDP and PCGDP its GDP and per-capita GDP,
respectively. All regressors are in logs. Results on the included constant are suppressed. For data sources see Appendix B. +, *, and ** signify statistical
significance at the 10%, 5% and 1% levels.
30
Table 2: Calibrated Parameter Values (Baseline)
Parameter
Value
σA, σM
2
θAl, θMl
country- and sector-specific
Tlj=eδ1*d(int)distljδ2
δ1A=3.34, δ1M=2.24,
δ2A=0.94, δ2M=0.73
A, α
A=208$I/year, α=0.72
Outline of Calibration Procedure
Based on Eaton, Kortum and Kramarz (2008) and Broda, Greenfield
and Weinstein (2006).
Labor productivity in manufacturing and agriculture, corrected for
sector specific price differences.
Estimated as coefficients on bilateral distance and internal trade flow
dummies in gravity equation estimations. See Section 5 for details.
Derived from cross-country regressions of manufacturing and
agricultural expenditure shares on proxies for agricultural prices and per
capita income (see Section 5 for details). $I denotes international dollars.
30
Data sources
-WDI, ICP
NBER, FAO
ICP
Table 3: Empirical Results for the Set of Countries Used in the Simulations
Regressor
log(RELPR)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
0.306**
(0.142)
-0.0825
(0.0934)
0.412**
(0.169)
-0.0910
(0.0972)
0.397**
(0.176)
0.0207
(0.0751)
-0.00121
(0.0891)
0.326**
(0.145)
0.0943
(0.0672)
0.195***
(0.0282)
0.139
(0.0936)
0.277*
(0.141)
-0.128*
(0.0658)
0.0658
(0.0824)
0.278*
(0.148)
0.208***
(0.0283)
0.204***
(0.0288)
-0.0882
(0.0629)
74
0.362
74
0.354
0.00168
(0.0844)
log(CEN)
log(AP)
log(POP)
log(GDP)
log(PCGDP)
Observations
R-squared
74
0.000
74
0.041
74
0.054
74
0.055
74
0.336
Notes: Table displays coefficients and robust standard errors (clustered on countries) for OLS estimations. The dependent variable is the logistic
transformation of a country’s share of manufacturing in GDP. RELPR is the quotient of a country’s agricultural labor productivity and its labor
productivity in manufacturing. Mindist is the minimum distance (in km) of a country to Japan, the European Union (Netherlands) and the USA.
CEN is a country’s centrality measure (defined in Section 2). POP is a country’s population size, AP its labor productivity in agriculture GDP and
PCGDP its GDP and per-capita GDP, respectively. All regressors are in logs. Results on the included constant are suppressed. For data sources see
Appendix B. +, *, and ** signify statistical significance at the 10%, 5% and 1% levels.
30
Table 4: Results for Generated Data (Baseline)
Regressor
log(RELPR)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
ltshareMGDP
0.594***
(0.175)
0.182***
(0.0482)
0.329*
(0.175)
-0.0194
(0.0434)
0.193*
(0.109)
0.399***
(0.0588)
-0.0137
(0.0443)
0.184*
(0.105)
0.405***
(0.0652)
0.0105
(0.0197)
0.00313
(0.0482)
0.181*
(0.105)
0.395***
(0.0571)
0.389***
(0.0592)
0.238**
(0.104)
0.0249
(0.0297)
0.261***
(0.0529)
0.492***
(0.0679)
74
0.674
74
0.719
0.237***
(0.0507)
log(CEN)
log(AP)
log(POP)
log(GDP)
log(PCGDP)
Observations
R-squared
74
0.209
74
0.151
74
0.244
74
0.671
74
0.672
Notes: Table displays coefficients and robust standard errors (clustered on countries) for OLS estimations using generated data (see Section 5 for
details). The dependent variable is the logistic transformation of a country’s share of manufacturing in GDP. RELPR is the quotient of a country’s
agricultural labor productivity and its labor productivity in manufacturing. CEN is a country’s centrality measure (defined in Section 2). POP is a
country’s population size, AP its labor productivity in agriculture and GDP and PCGDP its GDP and per-capita GDP, respectively. All regressors
are in logs. Results on the included constant are suppressed. +, *, and ** signify statistical significance at the 10%, 5% and 1% levels.
31
Table 5: Results for Simulated Sample (Robustness I)
(1)
Regressor
log(RELPR)
log(CEN)
R-squared
σA= σM
(3)
ltshareMGDP ltshareMGDP ltshareMGDP
(4)
(5)
(6)
ltshareMGDP ltshareMGDP ltshareMGDP
0.204***
(0.0501)
0.264*
(0.149)
0.000141
(0.0428)
0.138
(0.0954)
0.403***
(0.0611)
0.143***
(0.0478)
0.365**
(0.183)
-0.0556
(0.0440)
0.222*
(0.115)
0.395***
(0.0570)
0.0738
(0.0492)
0.397**
(0.183)
-0.121**
(0.0461)
0.249**
(0.119)
0.390***
(0.0560)
74
74
74
74
74
74
0.249
0.670
0.211
0.659
0.147
0.626
1.5
1.5
2.5
2.5
3
3
log(AP)
Observations
(2)
Notes: Table displays coefficients and robust standard errors (clustered on countries) for OLS
estimations using generated data (see Section 5 for details). The dependent variable is the logistic
transformation of a country’s share of manufacturing in GDP. RELPR is the quotient of a country’s
agricultural labor productivity and its labor productivity in manufacturing. CEN is a country’s
centrality measure (defined in Section 2). POP is a country’s population size, AP its labor
productivity in agriculture and GDP and PCGDP its GDP and per-capita GDP, respectively. All
regressors are in logs. Results on the included constant are suppressed. +, *, and ** signify statistical
significance at the 10%, 5% and 1% levels.
Table 6: Results for Simulated Sample (Robustness II)
(1)
Regressor
log(RELPR)
log(CEN)
R-squared
σA= σM
Trade Cost Matrix
A
α
Price Deflators
(3)
(4)
(5)
(6)
ltshareMGDP ltshareMGDP ltshareMGDP ltshareMGDP ltshareMGDP ltshareMGDP
0.192***
(0.0482)
0.342**
(0.160)
log(AP)
Observations
(2)
-0.00132
(0.0430)
0.174*
(0.0984)
0.402***
(0.0603)
0.210***
(0.0502)
0.319*
(0.180)
-0.00485
(0.0426)
0.171
(0.105)
0.427***
(0.0535)
0.182***
(0.0483)
0.333*
(0.178)
-0.0196
(0.0435)
0.196*
(0.110)
0.400***
(0.0588)
74
74
74
74
74
74
0.272
0.672
0.275
0.736
0.244
0.670
2
Poisson
208 I$
0.72
Producer
2
Poisson
208 I$
0.72
Producer
2
OLS
208 I$
0.72
Consumer
2
OLS
208 I$
0.72
Consumer
2
Poisson
170 I$
0.84
Consumer
2
Poisson
170 I$
0.84
Consumer
Notes: Table displays coefficients and robust standard errors (clustered on countries) for OLS
estimations using generated data (see Section 5 for details). The dependent variable is the logistic
transformation of a country’s share of manufacturing in GDP. RELPR is the quotient of a country’s
agricultural labor productivity and its labor productivity in manufacturing. CEN is a country’s
centrality measure (defined in Section 2). POP is a country’s population size, AP its labor
productivity in agriculture and GDP and PCGDP its GDP and per-capita GDP, respectively. All
regressors are in logs. Results on the included constant are suppressed. +, *, and ** signify statistical
significance at the 10%, 5% and 1% levels.
30
Table 7: Results for Simulated Sample (Baseline, Autarky, and Free Trade)
Baseline
(1)
Regressor
log(RELPR)
log(CEN)
R-squared
Coeff. (SE) actual on
simulated data
R2 actual on
simulated data
(2)
(3)
Autarky
(4)
(5)
(6)
ltshareMGDP ltshareMGDP ltshareMGDP ltshareMGDP ltshareMGDP ltshareMGDP
0.182***
(0.048)
0.329*
(0.175)
-0.019
(0.043)
0.193*
(0.109)
0.399***
(0.059)
-1.017***
(0.008)
-0.027
(0.020)
-1.022***
(0.014)
-0.031
(0.019)
0.011
(0.016)
0.260***
(0.051)
-0.026
(0.110)
0.009
(0.047)
0.129
(0.101)
0.425***
(0.072)
74
74
74
74
74
74
0.244
0.671
0.996
0.996
0.232
0.673
0.574
(0.199)***
0.574
(0.199)***
0.213
(0.135)
0.213
(0.135)
0.533
(0.195)***
0.533
(0.195)***
0.091
0.091
0.020
0.020
0.082
0.082
log(AP)
Observations
Free Trade
Notes: Table displays coefficients and robust standard errors (clustered on countries) for OLS
estimations using generated data (see Section 5 for details). The dependent variable is the logistic
transformation of a country’s share of manufacturing in GDP. RELPR is the quotient of a country’s
agricultural labor productivity and its labor productivity in manufacturing. CEN is a country’s
centrality measure (defined in Section 2). POP is a country’s population size, AP its labor
productivity in agriculture and GDP and PCGDP its GDP and per-capita GDP, respectively. All
regressors are in logs. Results on the included constant are suppressed. +, *, and ** signify statistical
significance at the 10%, 5% and 1% levels.
31
Appendix Tables
Table B.1: Data Sources for Regressions in Section 2
Variable
Source
Share of manufacturing value added in GDP
World Development Indicators
Value added per worker in agriculture and industry (in World Development Indicators
2000 USD)
Value added per worker in manufacturing (2000 USD) United Nations Industrial Statistics
Database (UNIDO)
GDP, GNP and per-capita GDP in 2000 USD
World Development Indicators
Population size
World Development Indicators
Bilateral distances between countries
NBER World Trade Database (Feenstra et
al., 1997, Feenstra, 2000)
32
Table B.2: List of country codes and names used
Country
Code
ALB
DZA
AGO
ATG
ARG
ARM
AUS
AUT
BHR
BGD
BRB
BLR
BEL
BLZ
BEN
BTN
BOL
BWA
BRA
BRN
BFA
BDI
KHM
CMR
CAN
CPV
CAF
TCD
CHL
CHN
COL
COM
ZAR
COG
CRI
CIV
HRV
CYP
DKF
DJI
DMA
DOM
ECU
EGY
SLV
GNQ
ERI
EST
ETH
FJI
FIN
GAB
GMB
DEU
GHA
GRC
GRD
GTM
GIN
GNB
GUY
HTI
HND
HKG
HUN
IND
IDM
IRN
ITA
JAM
JPN
JOR
KAZ
Country Name
Albania
Algeria
Angola
Antigua and Barbuda
Argentina
Armenia
Australia
Austria
Bahrain
Bangladesh
Barbados
Belarus
Belgium
Belize
Benin
Bhutan
Bolivia
Botswana
Brazil
Brunei
Burkina Faso
Burundi
Cambodia
Cameroon
Canada
Cape Verde
Central African Republic
Chad
Chile
China
Colombia
Comoros
Congo, Dem. Rep.
Congo, Rep.
Costa Rica
Cote d'Ivoire
Croatia
Cyprus
Denmark
Djibouti
Dominica
Dominican Republic
Ecuador
Egypt, Arab Rep.
El Salvador
Equatorial Guinea
Eritrea
Estonia
Ethiopia
Fiji
Finland
Gabon
Gambia, The
Germany
Ghana
Greece
Grenada
Guatemala
Guinea
Guinea-Bissau
Guyana
Haiti
Honduras
Hong Kong, China
Hungary
India
Indonesia
Iran, Islamic Rep.
Italy
Jamaica
Japan
Jordan
Kazakhstan
World Bank income
classification*
LM
LM
L
UM
UM
L
HOECD
HOECD
UM
L
UM
LM
HOECD
LM
L
L
LM
UM
UM
HNO
L
L
L
L
HOECD
LM
L
L
UM
LM
LM
L
L
L
LM
L
UM
HNO
HOECD
LM
UM
LM
LM
LM
LM
LM
L
UM
L
LM
HOECD
UM
L
HOECD
L
HOECD
UM
LM
L
L
LM
L
LM
HNO
UM
L
L
LM
HOECD
LM
HOECD
LM
LM
Country
Code
KEN
KIR
KOR
KWT
KGZ
LAO
LVA
LTU
MDG
MWI
MYS
MDV
MLI
MLT
MHL
MRT
MUS
MEX
MAR
MOZ
MMR
NAM
NPL
NLD
NCL
NZL
NIC
NER
NGA
OMN
PAK
PLW
PAN
PNG
PRY
PER
PHL
PRI
RWA
WSM
STP
SAU
SEN
SYC
SLE
SGP
SVN
SOM
ZAF
LKA
KNA
LCA
VCT
SUR
SWZ
TZA
THA
TGO
TON
TTO
TUN
TUR
UGA
UKR
ARE
URY
UZB
VUT
VEN
VNM
YEM
ZMB
ZWE
Country Name
Kenya
Kiribati
Korea, Rep.
Kuwait
Kyrgyz Republic
Lao PDR
Latvia
Lithuania
Madagascar
Malawi
Malaysia
Maldives
Mali
Malta
Marshall Islands
Mauritania
Mauritius
Mexico
Morocco
Mozambique
Myanmar
Namibia
Nepal
Netherlands
New Caledonia
New Zealand
Nicaragua
Niger
Nigeria
Oman
Pakistan
Palau
Panama
Papua New Guinea
Paraguay
Peru
Philippines
Puerto Rico
Rwanda
Samoa
Sao Tome and Principe
Saudi Arabia
Senegal
Seychelles
Sierra Leone
Singapore
Slovenia
Somalia
South Africa
Sri Lanka
St. Kitts and Nevis
St. Lucia
St. Vincent and the Grenadines
Suriname
Swaziland
Tanzania
Thailand
Togo
Tonga
Trinidad and Tobago
Tunisia
Turkey
Uganda
Ukraine
United Arab Emirates
Uruguay
Uzbekistan
Vanuatu
Venezuela, RB
Vietnam
Yemen, Rep.
Zambia
Zimbabwe
World Bank income
classification*
L
LM
UM
HNO
L
L
LM
LM
L
L
UM
LM
L
UM
LM
L
UM
UM
LM
L
L
LM
L
HOECD
HNO
HOECD
L
L
L
UM
L
UM
UM
LM
LM
LM
LM
UM
L
LM
L
UM
L
UM
L
HNO
HNO
L
UM
LM
UM
UM
LM
LM
LM
L
LM
L
LM
UM
LM
LM
L
L
HNO
UM
L
LM
UM
L
L
L
L
* L: low income, LM: lower middle income, UM: upper middle income, HOECD: high income, OECD; HNO:
high income, non-OECD.
33
Table B.3: Gravity Equation Estimates for 2000 (Poisson and OLS)
Regressor
dint
log(distance)
Observations
R-squared
Estimation method
Manufacturing
(1)
(2)
Exports
log(Exports)
Agriculture
(3)
Exports
(4)
log(Exports)
-2.238***
(0.100)
-0.733***
(0.038)
-2.607***
(0.370)
-1.454***
(0.044)
-3.335***
(0.155)
-0.939***
(0.050)
-4.676***
(0.576)
-1.288***
(0.061)
6,084
-Poisson
6,084
0.813
OLS
4,563
-Poisson
4,563
0.708
OLS
Notes: Table displays coefficients and robust standard errors (clustered by exporter) for OLS and
Poisson QML estimations (see Section 5 for details). The dependent variable are bilateral exports. The
regressors are dummy variable (dint), which takes the value one if a trade flow crosses national borders,
and the log of bilateral distance. Also included are a full set of exporter and importer fixed effects.
Results on the included constant are suppressed. See Table 2 for data sources. +, *, and ** signify
statistical significance at the 10%, 5% and 1% levels.
34