The Role of Trade Costs in Global Production Networks.
Evidence from China’s Processing Trade Regime
Alyson C. Ma,a Ari Van Assche,b,*
a
University of San Diego
b
HEC Montréal
*
Corresponding author. HEC Montréal, Department of International Business, 3000
Chemin de la Côte-Sainte-Catherine, Montréal (Québec), Canada H3T-2A7. Phone:
(514)340-6043. Fax: (514)340-6987. E-mail: [email protected].
ARTICLE INFO
JEL Classification: F12, F14, F23
Keywords: vertical specialization, global production networks, trade costs, China,
processing trade, export platform FDI.
Abstract
In this paper, we use data from China’s processing trade regime to analyze the role of trade costs
on trade within global production networks (GPNs). Under this regime, firms are granted duty
exemptions on imported inputs as long as they are used solely for export purposes. As a result,
the data provide information on trade between three sequential nodes of a global supply chain: the
location of input production, the location of processing (in China) and the location of further
consumption. This makes it possible to examine the role of both trade costs related to the import
of inputs (upstream trade costs) and trade costs related to the export of final goods (downstream
trade costs) on intra-GPN trade. The paper first sets up a three-country general equilibrium model
with asymmetric trade costs to develop testable hypotheses related to the impact of upstream and
downstream trade costs on intra-GPN trade. The empirical analysis provides empirical support for
the model’s predictions.
1 1. Introduction
The rise of global value chains has profoundly changed the nature of international trade.
Thanks to reductions in communication, transportation and other trade barriers,
multinational firms have sliced up their supply chains and dispersed their production
activities across multiple countries. This has led to a rapid growth in vertically
specialized trade between different nodes of the same global production network (GPN).
A large theoretical literature has recently emerged to investigate the drivers of
international production fragmentation and the implications for trade flows. A first stream
of this literature has focused on explaining the international production fragmentation
process by incorporating intermediate goods trade in otherwise standard models with
homogeneous goods, perfectly competitive markets and frictionless contracting.1 A
second stream has introduced elements of the theory of the firm into general-equilibrium
trade models to investigate the boundaries of the firm in international production.2 A third
stream has further dissected the value chain to investigate how a product’s technological
architecture affects the organization of global value chains.3
Due to data limitations, however, empirical research on the structure of trade within
global production networks has remained scant. As we explain in more detail in Section
2, firm level data and international trade data do not make clear what goods are inputs to
other goods within the same value chain, and multi-country input-output tables are at too
high a level of aggregation to capture the level of detail suggested by industry examples.
In this paper, we attempt to gain new insights into the structure of intra-GPN trade by
taking advantage of a unique data set on China’s processing trade regime for the period
from 1988 to 2008. Under this customs regime, firms are granted duty exemptions on
imported raw materials and other inputs as long as they are used solely for export
purposes. As a result, the data set provides, for each Chinese processing location, a
unique mapping of the source countries where processing inputs are imported from and
1
See for instance Jones and Kierzkowski (1990, 2000) and Venables (1999) and Grossman and RossiHansberg (2008). Antràs and Rossi-Hansberg (2009) review this literature.
This literature includes McLaren (2000), Grossman and Helpman (2002), Antràs (2003), Antràs and
Helpman (2004), Antràs and Helpman (2008) and Ornelas and Turner (2008). Spencer (2005) and Helpman
(2006) review this literature.
3
See Antrà and Chor (2011), Baldwin and Venables (2011) and Costinot, Vogel and Wang (2011).
2
2 the destination countries of processed exports. This makes it possible to examine the role
of both trade costs related to the import of inputs (upstream trade costs) and trade costs
related to the export of final goods (downstream trade costs) on intra-GPN trade. Such
mapping of GPNs cannot be conducted with regular trade data since imports are not
necessarily used solely for export purposes, but can also be consumed domestically.
In Section 3, we identify three stylized facts that suggest that both upstream and
downstream trade costs play an important role on China’s processing trade. First, China’s
processing exports heavily rely on foreign inputs, with a relatively low share of the value
made in China. According to a recent estimate by Koopman, Wang and Wei (2008), only
18% of China’s processing exports value is produced in China, while the remaining 82%
consists of the value of imported processing inputs. Second, the average distance traveled
by processing imports (import distance) is shorter than the average distance traveled by
processing exports (export distance). In 2008, 75% of China’s processing imports
originated from within the East Asian region, while 62% of the processing exports were
destined to non-Asian OECD countries. Third, this spatial pattern is not consistent across
processing locations. In a cross-section of 29 Chinese provinces, import distance is
negatively correlated to export distance for most years between 1995 and 2008. In other
words, locations in China that import their processing inputs from nearby tend to export
their processed goods far away, and vice versa.
To explain these stylized facts, we in Section 4 develop a three-country generalequilibrium trade model. In the model, the world consists of three countries: East (for
advanced East Asian countries), West (for Europe and North America), and China.
Multinational firms from the two advanced regions, East and West, sell goods in each
other’s markets. Each firm can serve the other market in one of two ways. It can produce
its goods at home and directly export them to the other market. Alternatively, it can
indirectly export its goods to the other market by processing them in the low cost country,
China. Since China is located in the vicinity of East, the model provides an explanation
for the negative correlation between export and import distance for China’s processing
trade: the inputs that China imports from nearby East are processed into final goods and
exported to the far-away West; conversely, the inputs that China imports from the faraway West are processed into final goods and exported to the nearby East.
3 If China’s processing trade is driven by indirect exports, the model predicts that China’s
processing exports to East should be more sensitive to export distance and less sensitive
to import distance than its processing exports to West. The intuition is the following. For
Eastern firms, the key distance factor that determines China’s attractiveness as a
processing location is its vicinity to Eastern input suppliers, i.e. import distance. The
larger is import distance, the less attractive China becomes as a location for processing
activities and therefore the less processed goods China exports. Conversely, for Western
firms, the critical determinant of China’s attractiveness as a processing location is its
proximity to the East Asian market, i.e. export distance. The larger is export distance, the
less attractive China becomes as a location for processing activities.
Using China’s bilateral processing trade data, we find support for this theoretical
prediction. Specifically, our empirical analysis provides evidence that processing exports
to East Asian countries are more sensitive to export distance and less sensitive to import
distance than processing exports to non-Asian OECD countries.
Beyond explaining China’s processing trade, our paper contributes to two emerging
literatures. First, it builds on a theoretical literature on export platform FDI, i.e. on
multinational firms that process their final goods in a foreign subsidiary for export to a
third-country market.4 Yeaple (2003) and Ekholm Forslid and Markusen (2007) examine
the determinants of export platform FDI by setting up a model with two similar advanced
“Northern” countries and a third Southern country in which the final good can be
assembled at lower cost. Both studies analyze how firms’ choices of using South as an
export platform depends on trade costs, factor-cost differentials, and the fixed costs
associated with foreign investment. Grossman, Helpman and Szeidl (2006) introduce
intra-industry firm heterogeneity in this type of setting and examine the role of different
types of complementarities on export platform activities. Our theoretical framework
complements these studies by using elements of Helpman, Melitz and Yeaple’s (2004)
model structure to derive a closed-form solution of an export platform’s bilateral
processing exports. Furthermore, we introduce more realistic assumptions about trade
costs by exploiting the empirical regularity that export-platform countries are generally
4
Yeaple (2003) uses the term ”complex integration strategy”.
4 located in the geographical proximity of large markets. For example, China lies in the
vicinity of developed East Asia; Mexico neighbors the United States. This allows us to
develop testable hypotheses that relate an export platform’s bilateral exports to both
export and import distance.
Second, our paper contributes to the New Economic Geography literature by theoretically
deriving and empirically validating how export platform FDI and intra-GPN trade
depends on firms’ foreign market access, foreign supplier access and the complex
interaction between both (Redding and Venables, 2004; Amiti and Javorcik, 2008;
Redding, 2010; Baldwin and Venables, 2010).
2. Mapping Global Production Networks
Empirical evidence on the organization of global production networks is remarkably
scant, largely because concrete evidence is difficult to obtain. To map global production
networks, one would like to know the number of countries involved in the production
process of a specific good, the value added created in each country, and the sequential
supply chain linkages between production activities. However, this information is hard to
come by. Firms are generally not willing to provide data on the cost structure of their
own supply chain activities, and often do not know the full range of value chain activities
conducted by its suppliers.5 Furthermore, national statistical agencies do not generally
track the domestic value added of goods that their countries trade, nor do they track the
use of these traded goods, that is, whether they are used for sales to final consumers,
whether they are used for further processing in a specific industry, and what share of this
industry’s output is exported.
In the field of international economics, scholars have used a variety of approaches to gain
at least partial insights into the structure of GPNs. One method has been to rely on the
highly disaggregated product codes and descriptions in international trade statistics to
5
Dedrick, Kraemer and Linden (2010) is a rare study that has been able to capture the value added created
by a lead firm and its most important component suppliers for specific electronics products. For this
purpose, they have relied on lists of components and their factory prices from industry analysts’
“teardown” reports, which capture the composition of the product at a specific point in time.
5 classify traded goods according to their main use. Yeats (2001) and Ng and Yeats (2001),
for example, categorized intermediate goods as those products whose description include
the words “parts” or “components”. Lemoine and Ünal-Kesenci (2004) and Zebregs
(2004) used the United Nation's "Broad Economic Categories" (BEC) classification to
distinguish between intermediate and final goods. While this approach has been useful to
demonstrate the large and growing role of GPNs in international trade, it faces two
important shortcomings. First, classifying goods according to their product codes is
somewhat arbitrary since product descriptions provide insufficient information to identify
a product’s main use (Hummels, Ishii and Yi, 2001). Indeed, some goods, such as
semiconductors, can be used both as a final good by consumers and as an intermediate
good by electronics manufacturers. Second, even if traded goods were correctly classified
as intermediate or final goods, international trade data do not identify in which sector
intermediate goods are used, if it is processed for domestic consumption, or if it is used
for export purposes. This makes it difficult to accurately link a trade flow with other trade
flows within the same GPN.
An alternative approach used to map activities within GPNs is to combine international
trade data with input-output (IO) table data. The advantage of IO data is that – for
domestic activities – it unambiguously defines intermediate inputs by their use, i.e., in
which industry they are put to use and what share of the industry’s output is exported.
This information on main use, however, is not available for imported goods in many
input-output
tables.
Lacking this information, researchers
have
adopted
the
proportionality assumption to approximate the main use of imports. That is, every
domestic sector is assumed to import inputs in the same proportion as its economy-wide
use of that input. For example, if an industry such as electronics relies on semiconductors
and 10% of all semiconductors are imported, it is assumed that 10% of the
semiconductors used by the electronics industry are imported. With this assumption,
scholars have been able to link the flow of imported inputs to the flow of exported goods
within the same global production network. Hummels, Ishii and Yi (2001) and Johnson
and Noguera (2009) have used this approach to quantify the import content embodied in a
country’s exports. Recent studies, then again, have questioned the accuracy of the
proportionality assumption (Feenstra et al., 2010). Winkler and Milberg (2009) have
6 shown that, in Germany, the cross-sectoral variation in the use of imported inputs differs
significantly from the cross-sectoral variation in the use of domestic inputs. Koopman,
Wang and Wei (2008) show that China’s policy preferences for processing exports has
led to a significant difference in the intensity of imported inputs in the production for
processing exports than in other productions (for domestic final sales and non-processing
exports). Puzzello (2010) demonstrate that the proportionality assumption overstates
Asian countries’ use and relative use of foreign inputs.
A third approach has been to use firm-level data on multinationals to measure the
dispersion of GPNs. Collinson and Rugman (2008), Rugman, Li and Oh (2009) and
Rugman and Oh (2009), for example, use data on the geographic distribution of assets for
large multinational firms to measure the dispersion of GPNs. Hanson, Mataloni and
Slaughter (2005) use BEA data on U.S. multinationals to estimate the drivers of trade in
intermediate inputs for further processing between parent firms and their foreign
affiliates. These studies, however, give an incomplete and potentially biased picture of
the organization of GPNs since many multinationals outsource a large portion of their
manufacturing activities to external firms (Gereffi, 1999; Gereffi, Humphrey and
Sturgeon, 2005; Swenson, 2006; Bonham, Gangnes and Van Assche, 2007; Desai, 2009).
If these outsourced activities are more dispersed geographically than the assets owned by
the multinational firms, the existing estimates on the dispersion of upstream activities
will be biased.
In this paper, we will exploit a unique data set that allows us to overcome some of the
shortcomings in the existing literature. Specifically, we will exploit a data set collected
by the General Administration of Customs of the People’s Republic of China on China’s
processing trade regime. Under this regime, firms are granted duty exemptions on
imported raw materials and other inputs as long as they are used solely for export
purposes. Since imported processing inputs may not be consumed domestically, the
processing trade data provides, for each processing location, a unique mapping of the
source countries where processing inputs are imported from and the destination countries
of processed exports. This makes it possible to directly link imported inputs to exported
products within the same GPNs, without relying on the proportionality assumption.
Furthermore, since the processing trade data incorporate both intra-firm and arm’s length
7 trade, it provides a more complete measure of trade flows within GPNs. In the next
section, we provide an overview of the processing trade data and identify three stylized
facts that relate import and export distance to processing trade patterns.
3. China’s Processing Trade Regime
China’s processing trade regime was installed in the mid-eighties in order to both attract
foreign direct investment and promote exports. Under the regime, firms were granted
duty exemptions on imported raw materials and other inputs as long as they are used
solely for export purposes. Largely ignored by many scholars, the regime was much more
far-reaching than similar systems introduced in other East Asian countries. Unlike in its
neighboring countries, China’s concessionary provisions were not geographically limited
within strictly policed export processing zones, but rather applied over its entire territory
(Naughton, 2006). As a result, China’s processing trade regime has turned into an
important part of its overall trade performance. As it is illustrated in Figure 1, between
1988 and 2008, the share of processing exports (i.e. exports conducted under the
processing regime) in China's total exports has risen from 30% to 51%, while the share of
processing imports in total imports has increased from 27% to 38%.
[Figure 1 about here]
A special characteristic of China’s processing exports is that it more heavily relies on
imported inputs than China’s non-processing exports. According to recent estimates by
Koopman, Wang and Wei (2008), only 18.1% of China’s processing export value is
produced in China, while the remaining 81.9% consists of the value of imported inputs
(see Figure 2). In comparison, the domestic content share of China’s non-processing
exports stood at a much higher 88.7%, meaning that imported inputs only represented
11.3% of the export value.
[Figure 2 about here]
In this section, we are primarily interested in the geographic characteristics of China’s
processing trade regime. To analyze the countries of origin of processing imports and the
8 destination countries of processing exports, an important data issue that needs to be
addressed is that 90% of China’s trade with its largest trading partner, Hong Kong, are reexported elsewhere (Feenstra et al., 1999; Feenstra, Hanson and Lin, 2004; Ferrantino
and Wang, 2007). This can significantly affect the analysis since it biases the true source
country of processing imports and the true destination country of processing exports that
are shipped through Hong Kong. To account for these re-exports, we follow Ma, Van
Assche and Hong (2009) by linking the processing trade data from China’s Customs
Statistics to a data set from the Hong Kong Census and Statistical Office on Hong Kong
re-exports. This allows us to estimate the country of origin of processing imports reexported through Hong Kong and the destination country of processing exports reexported through Hong Kong. A comparison of columns 1-2 and 3-4 in Table 1 illustrates
the impact of adjusting for re-exports through Hong Kong on China’s processing trade
with its major trading partners. It almost doubles the share of processing imports
originating from China’s other major trading partners and increases by a quarter the share
of processing exports destined to these same countries.
[Table 1 about here]
China’s processing trade regime heavily relies on East Asian inputs. As it is shown in
Figure 3, China heavily sources its inputs from its neighboring East Asian countries, with
75.1% of its processing imports originating from within East Asia in 2008. By contrast
the United States, EU-196 and Canada contributed relatively little to the supply of
processing inputs, together accounting for less than 19% of processing imports in 2008.
This asymmetric sourcing pattern of processing inputs has become more pronounced over
time. Between 1988 and 2008, the share of processing imports originating from China’s
most important East Asian trading partners has risen from 59.6% to 75.1%, while the
share of processing imports originating from non-Asian OECD countries has decreased
from 37.7% to 18.7% over the same period.
[Figure 3 about here]
6
The EU-19 include all European Union countries prior to the accession of the 10 candidate countries on 1
May 2004, plus the four eastern European member countries of the OECD, namely Czech Republic,
Hungary, Poland, Slovak Republic.
9 Conversely, the majority of processing exports are destined to non-Asian OECD
countries, except for an interlude between 1992 and 1997. As it is shown in Figure 4, the
share of processing exports destined to non-Asian OECD countries has risen from 54.7%
in 1997 to 59.4% in 2008. On the contrary, the share of processing exports destined
within the East Asian region has declined from 36.0% to 28.3% during the same period.
[Figure 4 about here]
This unbalanced processing trade pattern is generally attributed to the reorganization of
GPNs in East Asia (Yoshida and Ito, 2006; Gaulier, Lemoine and Ünal-Kesenci, 2007;
Haddad, 2007). With rising costs in Japan and the Newly Industrialized Economies
(NIEs) – Taiwan, Singapore, South Korea and Hong Kong – East Asian firms are
increasingly using China as a lower cost export platform. Instead of directly exporting
their final goods to the Western markets, these firms now export high value intermediate
goods to their processing plants in China and then export it on to the West after assembly.
As a result, it is argued that a triangular trade pattern has emerged in GPNs in which
China heavily relies on processing inputs from East Asia, while predominantly sending
processed goods to the West.
Data on the bilateral intensity of China’s processing trade provide further evidence of this
triangular trade structure. As it is shown in Figure 5, East Asian countries more
intensively supply China with processing inputs than countries outside of East Asia.
Except for Indonesia and Vietnam, more than 35% of China’s imports from its major
East Asian trading partners were processing imports in 2007 (see Figure 5). Almost 40%
of its imports from Japan and between 40% and 60% of its imports from the Newly
Industrialized Economies (South Korea, Taiwan and Singapore) were aimed at supplying
inputs for processing industries. This is a significantly higher share than for Western
countries. The share of processing imports in China’s total imports from the EU-19,
Canada and the United States amounted to 15.4%, 17.6% and 25.0%, respectively.
[Figure 5 about here]
10 At the same time, China more intensively supplies processed goods to developed
countries than to its East Asian neighbors. As it is shown in Figure 6, more than 50% of
the exports that China sends to the United States, the EU-19 and Japan are processing
exports. For most developing East Asian countries the number is significantly lower.
[Figure 6 about here]
The triangular trade pattern suggests that China is primarily used as an export platform by
East Asian firms that sell their goods to Western markets. However, in a cross-section of
29 Chinese provinces, the weighted average distance traveled by processing imports
(import distance) has been negatively correlated to the weighted average distance
travelled by processing exports (export distance) for most years between 1995 to 2008
(Ma, Van Assche and Hong, 2009). In other words, locations in China that import their
processing inputs from nearby tend to export their processed goods far away and vice
versa.
4. Trade Costs and Intra-GPN Trade
In this section, we first set up a three-country general equilibrium model that can
replicate the stylized facts of China’s processing trade and that can provide additional
empirical predictions (subsection 4.1). In subsections 4.2 and 4.3, we then set up the
empirical specification and present the results, respectively. In subsection 4.4, we check
the robustness of the results.
4.1. Theoretical framework7
Our model builds on a recent literature about export platform FDI (Yeaple, 2003;
Grossman, Helpman and Szeidl, 2006; Ekholm, Forslid and Markusen, 2007). Consider a
world with three countries. Two advanced countries, East and West, have high wages and
large markets for differentiated products. These countries are completely symmetric,
7
This is a generalized version of the model developed by Ma, Van Assche and Hong (2009).
11 with the exception of the size of their markets. In addition, there is a third developing
country, China, that has low wages and no local market for differentiated products.8
Households in the advanced countries consume goods from two industries. One industry
supplies homogeneous products in a perfectly competitive environment. The other
produces
differentiated
goods
under
monopolistically
competitive
conditions.
Consumers’ preferences are symmetric in East and West and are characterized by the
utility function
/
where
,0
1,
(1)
is a homogeneous good, q(v) is the vth variety in the differentiated goods sector
and n is the measure of varieties in the industry. With this utility function, the elasticity of
substitution between any pair of differentiated goods is
1.
Households in advanced country i spend an exogenous fraction of their income on
differential products so that the aggregate expenditure on the differentiated goods sector
equals
0.9 Maximizing the utility function in (1) subject to the consumer’s
,
budget constraint then generates the demand function for a variety v in advanced country
i:
,
(2)
where the demand level Ai is exogenous from the point of view of the individual firm.10
As it is generally the case in monopolistic competition models of this type, firms charge
the following price for its product:
,
where
(3)
denotes the firm’s marginal unit production cost and 1/ represents the markup
factor.
8
The assumption that China does not have a market for the industry’s output is not limiting since, by the
very nature of the processing trade regime, processed goods are not allowed to be sold on the Chinese
market.
9
Households in China lack sufficient income to consume differentiated products and instead concentrate
their purchases on the homogeneous good
10
As is well known,
available in country i and
in general equilibrium, where
is the price of variety v.
12 is the measure of varieties
Firms from advanced country i are more productive in producing the homogeneous good
than those located in China. We assume that one unit of labor is needed to produce
one unit of the homogeneous good in East and West, but that 1/w>1 units of labor are
needed to produce one unit of the good in China. We also assume that the homogeneous
good is produced in equilibrium in all three countries and take this good to be the
numéraire. This implies that
1
, where is the wage in country
i.
China is geographically closer to East than to West, while West is equidistant to both East
and China. Denote ij as the melting-iceberg trade cost of shipping goods from country i
to country j. We assume that ii = 1 and ij = ji > 1 for i ≠ j and that trade costs increase
linearly with distance so that EC = t < WC = WE =  (see Figure 7).
[Figure 7 about here]
In the remainder of the model, we focus on the differentiated goods industry. To simplify
notation, we will from now on drop the v’s. To produce a variety, we assume that a firm
needs to be headquartered in an advance country i  {E,W}. The production process
consists of two vertical value chain stages.11 First, a firm needs to produce its
intermediate good in its headquarter country j at cost
, where
equals the firm’s
labor-per-unit-output coefficient. Next, the firm can process the final good in any country
l  {E,W,C} at an extra ad valorem cost wl. We assume that these costs enter
multiplicatively.
The combination of production costs and trade costs implies that the marginal cost for a
firm of producing an intermediate good in its headquarter country j, processing it into a
final good in country l and delivering the final good to country i equals:
.
(4)
11
Since our study focuses on the geographic organization of international production and the resulting
intra-GPN trade flows, our model abstracts away from ownership decisions within global production
networks. In other words, if a “firm” in our model decides to offshore production to China, he can do so by
setting up its own subsidiary or by outsourcing to a local firm.
13 In the spirit of Melitz (2003), firms are heterogeneous in their labor unit requirements.
Entry requires a firm to bear a fixed fee Fe, measured in labor units. With this fee, the
entrant acquires the design for a differentiated product and draws a labor-per-unit-output
coefficient of
from a cumulative Pareto distribution G(a) with shape parameter
1. Upon observing this draw, the firm decides either to exit the industry or to
start producing. If it decides to produce, it bears an additional fixed cost fD>0 of initiating
production operations. There are no other fixed costs when the firm sells only for the
domestic market. If the firm chooses to export to the foreign market, however, it bears an
additional fixed cost fX>0 of forming a distribution and servicing network in the foreign
country. Finally, if it sets up a processing plant abroad, it bears an additional fixed cost
fO>0.12
Our set up implies that maximum four types of firms can sell their final goods in
advanced country ∈

,
:
Domestics (D) are local firms that are headquartered in i and that process their
final goods in country i.

Offshorers (O) are local firms that are headquartered in i and that process their
final good in China.

Exporters (X) are foreign firms that are headquartered in
and that process
their final goods in j.

Networkers (N) are foreign firms that are headquartered in
and that process
their final goods in China.
To make our model tractable, we set two additional assumptions that affect the types of
organizational forms that operate in the model. First, we assume that
1
.
(5)
As we show in appendix A, this implies that the marginal cost of processing final goods
in China is sufficiently high so that not all firms locate final good processing to China.
Specifically, there will be at least one domestic and one exporter selling in market i.
12
O stands for offshoring.
14 Second we assume that
1
.
(6)
This assumption means that China only has a local comparative advantage in final goods
processing. While China’s proximity to East makes it feasible for some Eastern
offshorers to operate, the large distance between China and West makes this unfeasible
for Western offshorers, thus ruling them out.13
Using equations (1)-(4), and taking into account assumptions (5) and (6), the operating
profits of the various firm types that sell in country i are:
Table 2: Profit functions
Market East
Market West
Domestic
Offshorer
Exporter
Networker
where Bi = (1 − )Ai/1−ε.
In Figure 8, we depict the profit functions of the four firm-types that sell in East. The
figure looks similar for West, with the exclusion of the profit function for offshorers. In
the figure, a1- ε is represented on the horizontal axis. Since ε > 1, this variable increases
monotonically with labor productivity 1/a, and can be used as a productivity index.14
[Figure 8 about here]
The figure shows that firms can be ranked according to their productivity. Consider first
the local firms headquartered in East. For a given productivity level, domestics face a
lower fixed cost but a higher marginal cost than offshorers. This implies that local firms
with a productivity level below
expect negative operating profits and exit the
13
14
While this result is not key to the model’s results, it significantly simplifies our analysis. This graphical approach of presenting the results is adopted from Helpman, Melitz and Yeaple (2004).
15 industry, firms with productivity levels between
and firms with productivity levels above
and
become domestics,
become offshorers. In other words, the
most productive local firms offshore their production to China (offshorers), while the less
productive firms produce their final goods locally (domestics).
Foreign firms headquartered in West face a similar pattern when selling in East. For a
given productivity level, exporters face a lower fixed cost but a higher marginal cost than
do not
networkers. This means that foreign firms with productivity levels below
and
sell their products in market i; foreign firms with productivity between
become exporters; while those with a productivity higher than
become
networkers. Once again, the most productive foreign firms offshore their production to
China (networkers), while the less productive foreign firms produce their final goods in
their home country (exporters).
Using the profit functions in Table 2, it is straightforward to derive that the cutoff
coefficients (aik)1-ε between firm types solve the following equations:
Table 3: Cut-off conditions
Market East
Market West
Domestic
Offshorer
1
Exporter
Networker
1
1
We can now proceed to calculate the industry sales of the various organizational forms
that sell in market i. The industry sales of type-k firms in country i amounts to the joint
revenue of all type k firms selling in country i. Using equations (2) and (3), the revenue
of a firm of type k in country i equals
, where c is given by equation (4). By
taking the integral of all firms of type k operating in country i, the industry sales of each
firm type k in country i totals:
16 Table 4: Industry sales
Domestic
Offshorer
Exporter
Networker
Market East
Market West
Ω
Ω
Ω
1
1
Ω
Ω
1
Ω
1
1
Ω
1
1
.
where
In the appendix, we derive a closed-form solution for Ω by taking the following three
steps. First, we use the market equilibrium condition that aggregate industry demand in a
country should equal aggregate industry supply (
. Next, we solve for
for every
Ω
Ω
Ω
Ω ) to solve for
by use the property of the Pareto distribution that
and
in the support of the distribution of a.15 Finally, we insert
the cutoff conditions of Table 3 into Table 4. This yields the following closed-form
solution:
Table 5: Closed-form solution for industry sales
Market East
Market West
Domestic
1
Ω
Offshorer
Ω
∆
∆
1
Ω
1
Ω
∆
See Helpman, Melitz and Yeaple (2004) for more details. 17 ∆
1
Exporter
15
Ω
∆
Networker
1
Ω
∆
where ∆
and x
1
Ω
1
, ∆
∆
∆
1
0.
The equations in Table 5 allow us to conduct comparative statics. We will focus on East
since there are more organizational forms active in that market, but the results are
identical for West. It is straightforward to derive that an increase in
leads to a
proportional expansion in the sales of all organizational forms, leaving the market share
of each organizational form unchanged.16
A rise in t and w decreases the sales of offshorers and networkers, but increases the sales
of domestics and exporters. This result is intuitive. A rise in t and w increases the cost of
processing goods in China, therefore increasing the marginal production cost for
offshorers and networkers. On the margin, this induces some of these firms to switch into
domestics and exporters, respectively (switching effect). For the firms that do not switch,
it raises their prices, therefore pressing consumers to substitute for products of domestics
and exporters (substitution effect). Both the switching effect and substitution effect lead
to a reduction in the sales of offshorers and networkers, and to an increase in the sales of
domestics and exporters.
A rise in
increases the sales of domestics and offshorers, but decreases the sales of
exporters and networkers. This is because a rise in
increases the trade cost of sending
products from China and East to West therefore increasing the marginal costs of
exporters and networkers. This, on its turn, increases the price of products sold by
exporters and networkers and reduces their sales. The price increase leads to a
substitution effect towards products made by domestics and offshorers, therefore
increasing their sales.
Finally, an increase in
and a decrease in
result into a rise in the sales of domestics
and exporters, and a decline in the sales of offshorers and networkers. Conversely, an
16
The derivation of these comparative statistics are available upon request. 18 increase in
leads to a reduction in the sales of exporters, while proportionately
expanding the sales of the three other organizational forms.
4.2. Drivers of China’s processing exports
The results in Table 5 also allow us to derive a closed-form solution for China’s
processing exports to East and West. As it is shown in Figure 9, processing exports from
China to East in our model equal the aggregate sales of offshorers and networkers in the
market East, i.e. the sum of Ω
Ω . Conversely, processing exports from China to
West equals the aggregate sales of networkers in West, i.e. Ω .
[Figure 9 about here]
Using these definitions, we can investigate if our model can replicate the stylized facts in
China’s processing trade documented in section 3. If
then the results of Table 5 imply that Ω
Ω
is sufficiently larger than
,
Ω . In that case, China’s processing
exports is dominated by networkers that sell in the West (see Figure 9). These firms
import their components from East into China and then export their final goods from
China to West, therefore replicating the triangular trade pattern in China’s processing
trade.
If t is high and
is low, the model also replicates the negative correlation between import
distance and export distance witnessed in the processing trade data. A higher t negatively
affects Ω more than Ω , since it increases the distance-related trade costs of offshorers
on both the import and export side. Conversely, a lower
increases
Ω , while
decreasing Ω . If as a result of this Ω >Ω , then China’s exports to both East and West
are dominated by networkers. In that case, China’s exports to West have a shorter import
distance than export distance. Simultaneously, China’s exports to East have a longer
import distance than export distance. Together, this implies a negative correlation
between import distance and export distance for firms using China as a processing
location.
19 If China’s processing trade is dominated by networkers, the model allows us to make an
additional prediction. In that case, the model predicts that China’s processing exports to
East should be (i) more sensitive to export distance and (ii) less sensitive to import
distance than its processing exports to the West (see the appendix for the derivation).
This result follow from the comparative statics results presented above, which states that
networkers’ sales (and therefore exports) are more sensitive to an increase in t than to an
increase in . Importantly, t and  have opposing roles for networkers selling to East and
West. For networkers selling to East, t reflects the distance-related trade costs of
importing components from West to China and  reflects the distance-related trade costs
of exporting the final good from China to East. Conversely, for networkers selling to
West, t reflects the distance-related trade costs of exporting final goods from China to
West and  reflects the distance-related trade costs of importing components from East to
China. Taken together, This implies that networkers’ processing exports to East are (i)
more sensitive to export distance and (ii) less sensitive to import distance than its
processing exports to the West.
4.3.Empirical Specification
To test the theoretical prediction, we estimate the following equation on Chinese
provinces’ processing exports with their foreign trading partners (adjusted for Hong
Kong re-exports) for the period 1988-2008:
ln
ln
ln
∗ ln
∗ ln
(6)
,
where the natural log of processing exports from a Chinese province i to a destination
country j in year t, lnXijt, is the dependent variable; XDij is export distance between
province i and country j; MDit is the weighted import distance for province i in period t;
refers to a standard vector of time-varying province-specific control variables, Eastj
is a dummy variable that equals 1 if the destination country is an East Asian country and
20 is 0 otherwise,
are province fixed effects,
are time fixed effects,
are country-time
is a normally distributed error term.17
fixed effects, and
Our model includes two independent distance variables to capture the role of distancerelated trade costs. To measure export distance (XDij), we use the arc distance between
the Chinese port closest to province i and the destination country j. To measure import
distance (MDit), we need to take into account that multiple inputs from various countries
are used in the production of a specific export good. As a consequence, we measure
import distance using the following formula:
∑
∑
.
,
(7)
where Mijt is province i’s imports from country j in period t; and XDij is the arc distance
between the Chinese port closest to province i and the source country j.
To analyze if China’s processing exports to East Asian countries are more sensitive to
export distance and less sensitive to import distance than its processing exports to
Western countries, we introduce a dummy variable, Eastj , that equals 1 if the country of
destination is an East Asian country and 0 if the destination market is a non-Asian OECD
country. We then introduce interaction terms between Eastj and our two distance
variables lnXDij and lnMDit as independent variables in our model. Chinese processing
exports are dominated by networkers if (i) the coefficient on the interaction term between
Eastj and lnXDij is significantly negative and (ii) the coefficient on the interaction term
between Eastj and lnMDit is significantly positive.
We estimate the effect of distance on processing exports using a set of standard controls.
Specifically, we use data from China’s Statistical Yearbook to control for the timevarying variables GDP per capita, population size and wages for Chinese provinces and
destination markets. To construct a specification that captures multilateral resistance, we
follow Rose and van Wincoop (2001) and Feenstra (2004) by adding both province-fixed
17
Our data consists of China’s East Asia neighbors (Indonesia, Malaysia, Thailand, Japan, Singapore,
South Korea, Philippines, Taiwan, and Hong Kong) and the Non-Asian OECD countries (Australia,
Canada, United States, and EU-19) since as shown in Table 1 the share of its processing imports and
exports from these trading partners accounts for most of its trade with 95.4% and 87.8%, respectively.
21 effects and country-time fixed effects.18 Finally, we use time fixed effects to account for
economic shocks common to all country pairs.19
4.4. Regression Results
Table 5 presents our OLS estimation results of equation (6). Column 1 includes the
independent variables that are generally used in gravity equations. Column 2 adds import
distance lnMDit as an independent variables. Column 3 includes the dummy variable
Eastj and the interaction terms.
[Table 5 about here]
The results provide support for the theoretical prediction. Specifically, we find that in
column 3 the coefficient on EastjlnXDij is negative and statistically significant, while the
coefficient on EastjlnMDit is positive and statistically significant. This suggests that
processing exports destined to East Asian countries are more sensitive to export distance
and less sensitive to import distance than processing exports destined to non-Asian
OECD countries.
4.4. Robustness Tests
The OLS results do not take into account the potential endogeneity of import distance and
GDP per capita. To account for this, we follow Sissoko (2004) and Carrère (2006) in
using the Hausman and Taylor (1981) estimation model to select the appropriate
instruments. The corresponding Hausman test leads us to reject the null hypothesis at the
0.01 level of significance and conclude that the model with the internal instruments
provides most efficient estimates. The results are presented in Table 6. Column 1
replicates the OLS regression results in column 3 of Table 5. Columns 2-4 provide the
Hausman-Taylor estimates for different combinations of endogenous variables. The
18
Ideally the province-fixed effects should also be time-varying since the multilateral resistance term may
vary over time (Egger, 2008). The inclusion of province x time fixed effects, however, would mean that no
time-varying parameters specific to an exporting province such as import distance can be estimated. 19
Other studies have argued for the inclusion of pair fixed effects to control for unobserved characteristics
of pairs of countries, but this would mean that no time-invariant parameters such as export distance
elasticity can be estimated (Baldwin and Taglioni, 2006).
22 results continue to provide supporting evidence for the important role of networkers in
China’s processing trade.
[Table 6 about here]
We also check the robustness of our results if we appropriately address zero trade flows.
Recent studies have highlighted that the standard OLS procedure to estimate gravity
models throws away important information contained in zero trade flows (Santos-Silva
and Tenreyro, 2006; Helpman, Melitz and Rubinstein, 2008). Zero trade flows become
undefined when converted in logarithms for estimation, thus creating a sample selection
bias. To address this issue, we follow Helpman, Melitz and Rubinstein (2008), henceforth
HMR, who propose estimating a two-stage model where a Probit equation is first
estimated to predict whether or not a province i exports to a country j in year t. In a
second stage, the gravity equation is then estimated in a non-linear specification. As in
HMR we use the unidirectional trade value with province and country fixed effect. To
ensure that we do not need to rely on the normality assumption for the unobserved trade
costs, we include the dummy variable coast only in the selection equation. The dummy
variable coast equals to 1 if both the originating province and the destination country
have a coastline, and 0 otherwise. An OLS estimation of equation (1) with the inclusion
of the dummy variable coast found that it does not have a significant impact on the
volume of trade.
The results using the HMR correction for zero trade flows are provided in columns 1-5 in
Table 7. The specifications in columns 1 and 2 include both province-year and countryyear fixed effects to control for unobserved heterogeneity across province and over time.
In general, the HMR specification explains at least 84% of the variation in bilateral
processing exports. While the inclusion of country*year fixed effects implies that we can
no longer estimate the coefficient on import distance, we can still estimate the
coefficients on the interaction effects between EastjlnXDij and EastjlnMDit, which as
shown in column 2 are negative and significant and positive and significant, respectively.
In columns 3 to 5 we replicate the benchmark specification found in columns 1 to 3 in
23 table 5 using the HMR correction for zero trade. The results are very similar to those in
the OLS estimation.
Additionally, we also used the Poisson Pseudo-Maximum Likelihood (PPML) to estimate
the gravity model as suggested by Santos -Silva and Tenreyro (2006) to account for the
presence of heteroskedasticity in trade data. The results of the PPML are similar to those
of the HMR and OLS estimations. In particular, the coefficient on EastjlnXDij is
negative and significant, while the coefficient on EastjlnMDit is positive and significant.
[Table 7 about here]
There is an important robustness check that we cannot conduct due to the nature of the
available data. Ideally we should control for structural differences between provinces by
disaggregating our analysis at the industry level. However, data limitations do not allow
us to do so. Specifically, it is not possible to disaggregate the analysis at the industry
level because we do not have the necessary input-output information regarding the
combination of inputs that are used to produce specific exports. For example,
semiconductors imported into a Chinese processing location can be used to produce both
cars and computers. The processing trade data identifies the value of semiconductors that
are imported by a certain location but not in which industry they are put to use.
While we need to take into account this limitation, our evidence suggests that China’s
processing exports not only depend on downstream trade costs (export distance), but also
on upstream trade costs (import distance). Specifically, we find that processing exports
are negatively affected by both import and export distance. Furthermore, processing
exports destined to East Asian countries are more sensitive to export distance and less
sensitive to import distance than processing exports destined to non-Asian OECD
countries.
5. Conclusion
We have tackled the question how trade costs affect intra-GPN trade by using a unique
data set on China’s processing trade regime. Under this customs regime, firms are
24 granted duty exemptions on imported raw materials and other inputs as long as they are
used solely for export purposes. As a result, the data set provides information on trade
between three sequential nodes of a global supply chain: the location of input production,
the location of processing (in China) and the location of further consumption. This makes
it possible to examine the role of both trade costs related to the import of inputs
(upstream trade costs) and trade costs related to the export of final goods (downstream
trade costs) on intra-GPN trade.
In a first step to evaluate the role of trade costs on China’s processing trade, we have
developed a three-country industry-equilibrium model in which heterogeneous firms
from two advanced countries, East and West, sell their products in each other’s markets.
Each firm can use two modes to serve the foreign market. It can directly export its
products from its home country. Alternatively, it can indirectly export to the foreign
market by assembling its product in a third low-cost country, China, which is assumed to
be located closer to East than to West. Our model illustrates that China’s processing
exports should not only depend on downstream trade costs (export distance), but also on
upstream trade costs (import distance), and the interaction of both. Using China’s
bilateral processing trade data, we have found empirical support for this complex impact
of these trade costs on intra-GPN trade.
25 Appendix A
We illustrate here that at least one exporter and domestic sells in country i if equation (5)
holds. First, we derive that there is at least one exporter selling in country i. Using the
cut-off conditions presented in Table 3, it is straightforward to calculate that
∗
.
(A-1)
Inserting equation (5) into (A-1) then implies that
1. This inequality suggests
that a least one exporter sells in country i.
Next, we derive that there is at least one domestic selling in country i. For the Western
market, this is always the case since equation (6) rules out that offshorers sell in West.
Using the cut-off conditions in Table 3, we can calculate that
∗
Since
∗
.
>
(A-2)
∗
, this implies that
is at least one domestic selling in each market.
26 1. As a result there
Appendix B
This appendix derives a closed-form solution for Ω . In a first step, we solve for
by
using the market equilibrium condition that aggregate sales equals aggregate expenditures
in market i:
Ω
Ω
Ω
Ω ,
(B-1)
If we insert the results of Table 4 into (B-1) and rearrange, we obtain that:
2
1
1
1
1
Notice that in our model V a
1
.
1
(B-2)
0.
Inserting equation (B-2) into Table 4 then gives:
Table B-1
Market East
Domestic
Market West
1
Ω
Ω
W
E
Offshorer
Ω
E
Exporter
Ω
Ω
E
W
Networker
Ω
where
Ω
E
and
1
W
1
1
.
Next, we can use the assumption that firms randomly draw a labor-per-unit-output
coefficient of
from a cumulative Pareto distribution
with shape parameter z. In
that case, Helpman, Melitz and Yeaple (2004) show that
shape parameter z − (ε − 1). The Pareto distribution implies that
27 , is also Pareto with the
(B-3)
for every
and
in the support of the distribution of a. Inserting the cutoff conditions
in Table 3 into equation (B-3) then yields:
Table B-2
Market East
Market West
1
1
1
1
1
1
1
Inserting the results of Table B-2 into Table B-1 then gives us the closed-form solution in
Table 5.
28 Appendix C
In this section we demonstrate that the following result holds when China’s processing
trade is dominated by networkers: China’s processing exports to East should be (i)
more sensitive to export distance and (ii) less sensitive to import distance than its
processing exports to the West.
1 so that there are no offshorers
To show this, we first consider the case where
selling to East. In that case, our model predicts that China’s processing exports are solely
conducted by networkers. China’s processing exports to East then equals Ω , where
amounts to trade costs on imports and t amounts to trade costs on exports. Conversely,
China’s processing exports to West equals Ω , where amounts to trade costs on imports
and amounts to trade costs on exports.
From Table 5, it is easy to see that the sales of the various organizational forms will be
the same (except for income). We therefore conduct the analysis on country i to
demonstrate that China’s processing exports to East is more sensitive to export distance
than China’s processing exports to West:
Ω
Ω
Ω
.
Ω
(C-1)
We start off by investigating the elasticity of Ω to . For this purpose, we log-linearize
the expression for Ω in Table 5:
ln Ω
ln
1
ln
ln
ln
1
ln ∆ .
(C-2)
Taking the derivative of ln Ω on then yields:
Ω
.
∆
Ω
(C-3)
into equation (C-2) and taking the absolute value then
Inserting the equality
gives us the elasticity of Ω
Ω
∆
Ω
1
to :
.
(C-4)
Next, we investigate the elasticity of Ω
expression for Ω in Table 5:
Ω
Ω
1 1
to t. For this purpose, we log-linearize the
1
∆
29 .
(C-5)
Denote
Inserting
Ω
∆
1
∆
Using ∆
1
1
1
1
1
.
∆
Ω
1
1
1
(C-6)
, we can rearrange (C-6) to:
1
Ω
.
1
.
1
If we take into account the parameter restrictions that
(C-7)
, that
and that
(from equation (5)), a comparison of equations (C-4) and (C-7)
shows that Ω is more sensitive to t than to :
30 1,
into equation (D-5) then yields:
Ω
1 and
. Due to our parameter restriction that
Ω
Ω
Ω
Ω
.
Bibliography
Amiti, M.; Javorcik, B. (2008). Trade costs and location of foreign firms in China.
Journal of Development Economics 85(1), 129-149.
Anderson, J. & van Wincoop, E. (2003). Gravity with gravitas: a solution to the border.
American Economic Review 93(1), 170-192.
Anderson, J. & van Wincoop, E. (2004). Trade costs. Journal of Economic Literature 42,
691-751.
Antràs, P. (2003). Firms, contracts, and trade structure. Quarterly Journal of Economics
118(4), 1374-1418.
Antràs, P. & and Chor, D. (2011). Organizing the global value chain. Mimeo.
Antràs, P. & Helpman, E. (2004). Global sourcing. Journal of Political Economy 112(3),
552-580.
Antràs, P. & Helpman, E. (2008). Contractual frictions and global sourcing. In Helpman,
E.; Marin, D. & Verdier, T. (Eds.). The Organization of Firms in a Global Economy.
Cambridge, MA, Harvard University Press, 9-54.
Baldwin, A. & Taglioni, N. (2006). Gravity for dummies and dummies for gravity
equations. NBER Working Paper 12516.
Baldwin, A. & Venables, A. (2006). Relocating the value chain: offshoring and
agglomeration in the global economy. NBER Working Paper 16611.
Bonham, C., Gangnes, B. & Van Assche, A. (2007). Fragmentation and East Asia’s
information technology trade. Applied Economics 39(2), 215-228.
Brun, J., Carrère, C., Guillaumont, P., & de Melo, J. (2005). Has distance died? Evidence
from a panel gravity model. World Bank Economic Review 19 (1), 99-120.
Carrère, C. (2006). Revisiting the effects of regional trading agreements on trade flows
with proper specification of the gravity model. European Economic Review 50(2), 223247.
Coe, D., Subramanian, A., & Tamirisa, N. (2007). The missing globalization puzzle :
evidence of the declining importance of distance. IMF Staff Papers, 54(1), 34-58.
Collinson, S. & Rugman, A. (2008). The regional nature of Japanese multinational
business. Journal of International Business Studies 39, 215-230.
Costinot, A., Vogel, J. and Wang, S. (2011). An elementary theory of global supply
chains. Mimeo.
Dedrick, J., Kraemer, K., & Linden, G. (2010). Who profits from innovation in global
value chains? A study of the iPod and notebook PCs. Industrial and Corporate Change
19(1), 81-116.
Desai, M. (2009). The decentering of the global firm. World Economy 32(9), 1271-1290.
31 Disdier, A., & Head, K. (2008). The puzzling persistence of the distance effect on
bilateral trade. The Review of Economics and Statistics 90(1), 37-48.
Egger, P. (2008). On the role of distance for bilateral trade. The World Economy 31(5),
653-662.
Ekholm, K., Forslid, R., & Markusen, J. (2007). Export-platform foreign direct
investment. Journal of the European Economic Association, 5(4), 776-795.
Feenstra, R. (2004), Advanced International Trade: Theory and Evidence (Princeton, NJ:
Princeton University Press).
Feenstra, R., Deng, H., Ma, A., Yao, S. (2004). Chinese and Hong Kong international
trade data. Mimeo.
Feenstra, R., Hai, W., Woo, W. & Yao, S. (1999). Discrepancies in international trade: an
application to China-Hong Kong entrepôt trade, American Economic Review 89(2), 338343.
Feenstra, R., Hanson, G., & Lin, S. (2004). The value of information in international
trade: gains to outsourcing through Hong Kong. Advances in Economic Analysis &
Policy, 4(1), article 7.
Feenstra, R.; Lipsey, R.; Branstetter, L; Foley, F.; Harrigan, J.; Jensen, B.; Kletzer, L.;
Mann, C.; Schott, P. Wright, G. (2010). Report on the state of available data for the study
of international trade and foreign direct investment. NBER Working Paper No. 16254.
Ferrantino, M. & Wang, Z. (2007). Accounting for discrepancy in international trade: the
case of China, Hong Kong and the United States. USITC Office of Economics Working
Paper 2007-04-A.
Gaulier, G., Lemoine, F. & Ünal-Kesenci, D. (2007). China’s integration in East Asia:
Production sharing, FDI & high-technology trade. Economic Change and Restructuring
40, 27-63.
Grossman, G. & Helpman, E. (2002). Integration versus outsourcing in industry
equilibrium. Quarterly Journal of Economics 117(1): 85-120.
Grossman, G., Helpman, E., & Szeidl, A. (2006). Optimal integration strategies for the
multinational firm. Journal of International Economics, 70, 216-238.
Grossman, G. & Rossi-Hansberg, E. (2008). Trading tasks: A simple theory of
offshoring. American Economic Review 98(5), 1978-1997.
Gereffi, Gary (1999). International trade and industrial upgrading in the apparel
commodity chain. Journal of International Economics 48(1), 37-70.
Gereffi, G., Humphrey, J. & Sturgeon, T. (2005). The governance of the global value
chains. Review of International Political Economy 12(1), 78–104.
Haddad, M. (2007). Trade integration in East Asia: the role of China and production
networks. World Bank Policy Research Working Paper 4160.
32 Hanson, G., Mataloni, R., & Slaughter, M. (2005). Vertical production networks in
multinational firms. Review of Economics and Statistics, 87(4), 664-678.
Hausman J. & Taylor W. (1981). Panel data and unobservable individual effects.
Econometrica 49, 1377-1398.
Helpman, E., 2006. Trade, FDI, and the organization of firms. Journal of Economic
Literature. 44(3), 589–630.
Helpman, E., Melitz, M., & Yeaple, S. (2004). Export versus FDI with heterogeneous
firms. American Economic Review, 94(1), 300-316.
Hummels, D., Ishii, I., & Yi, K.-M. (2001). The nature and growth of vertical
specialization in world trade. Journal of International Economics 54(1), 75-96.
Johnson, R. & Noguera, G. (2011). Accounting for intermediates: production sharing and
trade in value added. Mimeo.
Jones, R. & Kierzkowski, H. (1990). The role of services in production and international
trade: a theoretical framework. In Jones, R. & Krueger, A. (Eds.), The Political Economy
of International Trade: Essays in Honor of Robert E. Baldwin, Oxford, Blackwell: 31-48.
Jones, R.; Kierzkowski, H. (2000). A framework for fragmentation. Tinbergen Institute
Discussion Paper 056/2.
Koopman, R., Wang, Z., & Wei, S.-J. (2008). How much of Chinese exports is really
made in China? Assessing domestic value-added when processing trade is pervasive.
NBER Working Paper 14109.
Lemoine, F.; Ünal-Kesenci, D. (2004). Assembly trade and technology transfer. World
Development 32(5), 829-850.
Ma, A. & Van Assche, A. (2010). China’s role in global production networks. Mimeo.
Ma, A., Van Assche, A. & Hong, C. (2009). Global production networks and China’s
processing trade. Journal of Asian Economics, 20(6), 640-654.
McLaren, J. (2000). Globalization and vertical structure. American Economic Review,
90(5), 1239-54.
Melitz, M. (2003). The impact of trade on intra-industry reallocations and aggregate
industry productivity. Econometrica 71, 1695–1725.
Naughton, B. (2006). China’s emergence and prospects as a trading nation. Brookings
Papers on Economic Activity 2, 273-313.
Ng, F. & Yeats, A. (2001). Production sharing in East Asia: who does what for whom,
and why? In L. Cheng & H. Kierzkowski (Eds.), Global production and trade in East
Asia. Boston: Kluwer Academic Publishers, pp. 63-109.
33 Ornelas, E. & Turner, J. (2008). Trade liberalization, outsourcing, and the hold-up
problem. Journal of International Economics 74(1), 225-241.
Puzzello, L. (2011). A proportionality assumption and measurement biases in the factor
content of trade. Mimeo.
Redding (2010). The empirics of new economic geography. Regional Science 50(1), 297311.
Redding, S. & Venables, A. (2004). Economic geography and international inequality.
Journal of International Economics 62(1), 53-82.
Rose, A. & van Wincoop, E. (2001). National money as a barrier to international trade:
the real case for currency union’, American Economic Review, Papers and Proceedings,
91(2), 386–90.
Rugman, A., Li, J., & Oh, C. (2009). Are supply chains global or regional?” International
Marketing Review 26(4), 384-395.
Rugman, A., & Oh, C. (2009). Geography and regional multinationals. Mimeo.
Sissoko A. (2004). Measuring trade intensity within the European zone. Mimeo.
Spencer, B. (2005). International outsourcing and incomplete contracts. Canadian
Journal of Economics 38, 1107–1135.
Swenson, D. (2006). Country competition and US overseas assembly. World Economy
29(7), 917-937.
Venables, A. (1999). Fragmentation and multinational production. European Economic
Review 43(4-6), 935-945.
Winkler, D. & Milberg, W. (2009). Errors from the “proportionality assumption” in the
measurement of offshoring: application to german labor demand. Schwartz Center for
Economic Policy Analysis Working Paper 2009-12.
Yeaple, S. (2003). The complex integration strategies of multinationals and cross country
dependencies in the structure of foreign direct investment. Journal of International
Economics, 60, 293-314.
Yeats, A. J. (2001) Just how big is production sharing? In S. Arndt and H. Kierzkowski
(eds.), Fragmentation: new production patterns in the world economy, Oxford: Oxford
University Press, 108-143.
Yi, K.-M. (2003). Can vertical specialization explain the growth of world trade? Journal
of Political Economy 111, 52-102.
Yoshida, Y. & Ito, H. (2006). How do the Asian economies compete with Japan in the
US market? Is China exceptional? A triangular trade approach. Asia Pacific Business
Review 12(3), 285-307.
34 Zebregs, H. (2004). Intra-regional trade in emerging Asia. IMF Policy Discussion Paper
PDP04/01.
35 Figure 1: Proportion of processing trade in China’s total trade, 1988-2008
60
Percentage
50
40
30
20
10
0
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
Year
Export
Import
Source: Authors’ calculations using China’s Customs Statistics.
36 Figure 2: Domestic and foreign content share of China’s processing and nonprocessing exports
Non‐Processing Exports
Processing Exports
Foreign content share:
11%
Chinese content share:
18%
Foreign content share:
82%
Chinese content share:
89%
Source: Koopman, Wang and Wei (2008).
37 Figure 3: Share of Processing Imports, by region of origin, 1988-2008
Share of Processing Imports
90
80
70
60
50
40
30
20
10
0
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Year
East Asia
Non-Asian OECD
Source: authors’ calculations, using China’s Customs Statistics
38 Rest of the World
Figure 4: Share of Processing Exports, by region of destination, 1988-2008
Share of Processing Exports
70
60
50
40
30
20
10
0
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Year
East Asia
Non‐Asian OECD
Rest of the World
Source: authors’ calculations, using China’s Customs Statistics
39 Figure 5: Processing imports as a share of China’s total imports, by country
Share of Processing Imports in Total Imports
of origin, 2007 (%)
70
60
50
40
30
20
10
0
Countries
Source: authors’ calculations, using China’s Customs Statistics
40 Figure 6: Processing exports as a share of China’s total exports, by
Share of Processing Exports in Total Exports
destination country, 2007 (%)
70
60
50
40
30
20
10
0
Countries
Source: authors’ calculations, using China’s Customs Statistics
41 Figure 7: Geographical assumptions
ASIA
East

t
West

China
42 Figure 8: Profit functions for the four firm-types selling in country i
 ki
 Oi
 Di
 Ni
 Xi
0
 fD
a 
i 1
D
a 
i 1
O
a 
i 1
X
  f X  fD 
  fO  f D 
 ( fO  f X  f D )
43 a 
i 1
N
a 
i 1
k
Figure 9: China’s processing trade, by firm type
PROCESSING EXPORTS TO EAST
PROCESSING EXPORTS TO WEST
East
t
China

West
Offshorer
Networker
44 Networker
Table 1
The origin and destination of China’s processing import and export, 2008
Share of processing imports originating from East Asia Hong Kong Japan South Korea Singapore Taiwan Malaysia Thailand Philippines Vietnam Indonesia Macau Unadjusted for HK re‐exports 85.0 45.3 11.3 11.3 2.9 8.6 1.8 1.4 1.5 0.1 0.5 0.3 Non‐Asian OECD EU‐19 United States Canada Australia Other 10.4 4.8 4.3 0.5 0.3 0.5 ROW 4.6 Adjusted for HK re‐exports 72.8 ‐ 20.4 17.3 4.6 15.8 6.0 4.5 2.8 0.2 0.9 0.4 19.4 9.1 7.4 0.7 0.7 1.6 8.8 Share of processing exports destined to Unadjusted for HK re‐exports 49.3 29.3 8.2 4.2 2.3 1.6 1.4 0.7 0.5 0.3 0.5 0.2 Adjusted for HK re‐exports 28.2 ‐ 11.6 6.0 3.3 2.2 1.9 1.0 0.7 0.5 0.8 0.3 42.1 19.4 18.2 1.3 1.2 2.0 59.6 27.5 25.7 1.8 1.7 2.8 8.6 12.2 Authors’ calculations using China’s Customs Statistics Data
45 Tables 2, 3 and 4 are included in the text.
46 Table 5
Regression results, 1988-2008
Dependent variable: log of bilateral processing exports, by province
OLS
GDP per capita (province)
Population (province)
Wage (province)
Export Distance
(1)
(2)
(3)
1.655***
[0.132]
3.968***
[0.215]
-0.541***
[0.112]
-0.505***
[0.043]
1.601***
[0.133]
3.884***
[0.216]
-0.616***
[0.114]
-0.504***
[0.043]
-0.170***
[0.058]
Yes
Yes
Yes
Yes
Yes
Yes
1.609***
[0.133]
3.872***
[0.213]
-0.621***
[0.114]
0.087
[0.223]
-0.342***
[0.061]
-0.570***
[0.233]
0.576***
[0.070]
Yes
Yes
Yes
Import Distance
East*Export Distance
East *Import Distance
Year Dummy
Province Dummy
Country-Year Dummy
Observations
17306
17299
17299
R2
0.835
0.836
0.836
Notes: Robust standard errors are in parentheses. * means significant at 10%; ** means significant at 5%;
*** means significant at 1%. Constant not reported. 47 Table 6
Regression results, 1988-2008
Dependent variable: log of bilateral processing exports, by province
OLS
GDP per capita (province)
Population (province)
Wage (province)
Export Distance
Import Distance
East*Export Distance
East *Import Distance
Year Dummy
Province Dummy
Country-year Dummy
Hausman Taylor Estimator
(1)
(2)a
(3)
(4)
1.609***
[0.133]
3.850***
[0.213]
-0.612***
[0.114]
0.101
[0.223]
-0.336***
[0.061]
-0.620***
[0.233]
0.588***
[0.069]
Yes
Yes
Yes
1.7137***
[0.102]
3.7072***
[0.196]
-0.4870***
[0.104]
-0.1498
[0.215]
-0.3255***
[0.057]
-0.3650
[0.224]
0.6246***
[0.070]
Yes
Yes
Yes
1.5666***
[0.101]
3.3027***
[0.187]
-0.3337***
[0.102]
-0.3401
[0.212]
-0.3377***
[0.057]
-0.1662
[0.221]
0.6412***
[0.070]
Yes
Yes
Yes
1.7733***
[0.103]
4.0353***
[0.204]
-0.5698***
[0.105]
-0.0270
[0.217]
-0.3158***
[0.057]
-0.4919**
[0.226]
0.6299***
[0.070]
Yes
Yes
Yes
Observations
17285
17285
17285
17285
R2
0.837
Hausman test HT vs. GLS (χ2)
1791.36
10354.41
6004.33
Notes: Robust standard errors are in parentheses. * means significant at 10%; ** means significant at 5%;
*** means significant at 1%.
Column 2: endogenous variables = GDP per capita (province), GDP per capita (country), import distance.
Column 3: endogenous variables = import distance.
Column 4: endogenous variables = GDP per capita (province), GDP per capita (country). 48 Table 7
Regression results, 1988-2008
Dependent variable: log of bilateral processing exports, by province
HMR
HMR
HMR
HMR
HMR
PPML
(1)
(2)
(3)
(4)
(5)
(6)
1.484***
1.448***
1.445***
1.096***
[0.131]
[0.132]
[0.132]
[0.241]
***
***
***
GDP per capita
(province)
Population (province)
3.727
[0.218]
Wage (province)
Export Distance
-0.693
***
3.672
[0.218]
-0.745
***
3.630
[0.218]
-0.756
***
-1.799***
[0.422]
1.515***
[0.111]
[0.113]
[0.113]
[0.254]
-0.501***
0.170
-0.640***
-0.637***
0.073
-0.161
[0.059]
[0.203]
[0.064]
[0.064]
[0.225]
[0.145]
Import Distance
-0.120
**
[0.057]
East*Export Distance
-0.6768
***
0.5753
[0.060]
-0.708
[0.210]
East *Import Distance
-0.319
***
***
-0.291*
[0.159]
-0.679***
[0.232]
[0.135]
***
0.280**
[0.069]
[0.133]
0.677
[0.063]
***
-0.269**
Coast
[0.117]
zhat
2.471
***
[0.320]
2
zhat
-0.682
3
1.292
[0.319]
-0.676
[0.352]
***
-0.447
***
1.307
***
[0.352]
-0.449
***
1.340
***
[0.349]
-0.454***
[0.084]
[0.092]
[0.092]
[0.092]
***
***
***
***
0.045***
0.063
0.043
0.043
[0.009]
[0.009]
[0.009]
[0.009]
[0.009]
1.546***
1.540***
1.754***
1.743***
1.742***
[0.154]
[0.153]
[0.173]
[0.172]
[0.171]
Province-year dummy
Yes
Yes
Country-year dummy
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
eta_hat
Province
Observations
17306
17299
17306
17299
17299
22845
R2
0.868
0.869
0.839
0.839
0.840
0.924
49 ***
[0.084]
0.063
zhat
***
2.453
***