Exceptional Exporter Performance?

Evidence from
Chinese Manufacturing Firms
Dan Lu
University of Chicago
Job Market Paper

Nov 2010

I am grateful to Samuel Kortum, Thomas Chaney, Robert Lucas, and Nancy Stokey for their continuous
advice and extremely helpful comments. I also thank Fernando Alvarez, Costas Arkolakis, Lorenzo Caliendo,
Pedro Gete, Chang-Tai Hsieh, Priscilla Man, Asier Mariscal, Ezra Ober…eld, John Romalis, Robert Shimer,
Anna Wang, and seminar participants at Empirical Investigation in Trade, the University of Chicago Money
and Banking Workshop, Capital Theory Working Group, and International Economics Working Group for
very helpful comments. Please send comments to danlu@uchicago.edu.

1
 Abstract

This paper uses Chinese …rm-level data to document facts that run counter to the accumu-
lated evidence about exporting …rms and provides a model that reconciles these contrasting
patterns. The new facts are: (1) China’s exporters are typically less productive and sell
less in the domestic market than non-exporters, and (2) the distribution of export inten-
sity exhibits a U-shape, with more than half of China’s exporters exporting most of their
output. Previous studies of …rms in more developed countries have found that exporters
are more productive, sell more in the domestic market, and export only a small fraction of
their output. The new facts call into question the generality of recent trade theory, which
has been extremely successful in explaining the behavior of exporters in developed countries.
However, I show that the economic forces described by Melitz (2003), when properly inter-
preted, are exactly the ones needed to explain the observed patterns among Chinese …rms.
When countries di¤er in their factor endowment, sectors that are intensive in the locally
abundant factor face higher competition in the domestic market than in foreign markets.
Hence domestic rather than export markets select the most e¢ cient …rms. In the Chinese
data, both the productivity di¤erences between exporters and non-exporters as well as the
distribution of export intensity are systematically related to the labor intensity of the …rm
or its industry. This relationship is exactly what is predicted by a Melitz model augmented
to allow for factor intensity to vary by industry. Lastly, I show that the model correctly
predicts the e¤ects of trade liberalization in China following China’s integration into the
WTO in 2001.

2
1 Introduction
This paper documents stylized facts about exporting …rms in China that contrast with the
accumulated evidence in the literature. First, China’s exporters are typically less productive
and sell less in the domestic market than non-exporters. Second, the distribution of export
intensity in China exhibits a U-shape, i.e., while many exporters only export a very small
fraction of their output, another large set of exporters export the majority of their output.
However, previous studies of …rms in developed countries have found that exporters are
more productive, sell more in the domestic market than their non-exporting counterparts,
and export only a small fraction of their output.1
These facts seem to be inconsistent with existing trade theories as well. Existing theories
(e.g. Melitz (2003), Chaney (2008), Eaton, Kortum, and Kramarz (2009)) postulate that
…rms di¤er in production e¢ ciency. With …xed costs and trade costs of selling to foreign
markets, only the more productive …rms will be able to export. These models seem to be a
poor match for the behavior of Chinese …rms, which raises the question: How can we explain
the very di¤erent …rms’exporting behavior?
A useful clue comes from the fact that export patterns of Chinese …rms are systematically
related to the factor intensity of …rms/sectors. In labor-intensive sectors, exporters are
actually the majority, are less productive than non-exporters, and export a large fraction of
their output. On the other hand, in capital-intensive sectors, exporters are the minority, are
more productive than non-exporters, and export a small fraction of their output.
This brings us to an explanation based on factor intensity. Indeed, Melitz’s (2003) insight
is that only the more productive …rms are able to enter tougher markets, which we have
always assumed to be the foreign markets. This is not necessarily the case, especially if
countries di¤er in factor endowments, and sectors di¤er in factor intensities. In particular,
for a labor-abundant country like China, wages in its main export destinations are likely to
be higher than its domestic wage. For labor-intensive goods, prices are higher in the foreign
market than in the domestic market. Thus the exporting markets are less competitive
than the domestic market for Chinese …rms in the labor-intensive sectors. As a result, less
productive …rms sell only to foreign markets while the survivors in the more competitive
domestic market are the more productive …rms. For example, for a Chinese producer of
apparel (a labor-intensive sector), the Chinese market, populated by …rms with access to
cheap labor, is much tougher than the US market, populated mostly by …rms with access
to more expensive labor. The opposite is true for capital-intensive sectors. For a Chinese
1
For example, Bernard and Jensen (1995, 1999), Aw, Chung, and Roberts (1998), and Bernard and
Wagner (1996) show that across a wide range of countries and industries, employment, shipments, wages,
productivity, and capital intensity are all higher at exporters than non-exporters at any given moment.

3
producer of pharmaceuticals (a capital-intensive sector), it is much harder to enter the US
market than the domestic market.
The relative competitiveness of markets also explains …rms’export intensity. By adopting
Arkolakis’s (2008) market access costs, we expect that the more productive …rms in markets
will access more consumers and sell more. For Chinese …rms in labor-intensive sectors, the
foreign markets are less competitive than the domestic market. These exporters are relatively
more pro…table in foreign markets and are thus able to access more consumers there. Hence,
a large fraction of their output is sold abroad. Again the opposite is true for exporters in
capital-intensive sectors. Exporters sell a small fraction of their output abroad. The model
explains the di¤erent patterns of trade, productivity, and export intensity in China as well
as in developed countries documented in the literature.
When embedded in a general equilibrium framework, the model predicts that if we lower
import tari¤s in China, labor will shift to labor-intensive sectors because of Heckscher–Ohlin
e¤ects. In addition, the productivity advantage of exporters decreases across all sectors as the
domestic markets become more competitive due to foreign competition. These predictions
are con…rmed by the Chinese data when China joined the WTO in 2001 and reduced import
tari¤s dramatically.
In the literature, the superior performance of exporting …rms relative to non-exporters is
often explained by self-selection (i.e., good …rms become exporters) rather than by learning
by exporting. This paper supports the self-selection mechanism precisely because exporters
do not necessarily exhibit superior performance.2
Introducing factor endowments and di¤erent factor intensity across sectors into Melitz
(2003) follows the model of Bernard, Redding, and Schott (2007). But in this paper, I show
why we need to bring Heckscher-Ohlin theory to the study of …rm-level trade. When …rms in
a labor-abundant country like China export to developed countries, comparative advantage
makes it easier to enter foreign markets for labor-intensive goods; hence …rms’ exporting
behavior exhibits patterns opposite to those observed in capital-abundant countries.
The rest of this paper is organized as follows: Section 2 reports the facts. Section 3
explains individual …rms’decisions in a partial equilibrium setup, and shows how the model
…ts the data. Section 4 presents the general equilibrium analysis and the e¤ects of tari¤
change. Section 5 concludes.
2
Clerides, Lach, and Tybout (1998) concludes that the well-known e¢ ciency gap between exporters and
non-exporters is due to self-selection of the more e¢ cient …rms into the export market, rather than due to
learning by exporting. Bernard and Jensen (1999) also …nds evidence that good …rms become exporters. Both
growth rates and levels of success measures are higher ex-ante for exporters, while the bene…ts of exporting
for the …rm are less clear. Whether exporting improves Chinese …rms’performance is not analyzed in the
paper.

4
2 Empirical Evidence
The data used in this paper comes from a …rm-level data set from the Annual Census of
Enterprises by the Chinese National Bureau of Statistics from 1998 to 2007.3 It includes all
the State-Owned Enterprise (henceforth SOE) and non-SOEs with sales over 5 million RMB
(about 600,000 US dollars), yielding 162,855 …rms in the 2000 survey. This number rises
to 336,768 in 2007. The data contain all information from the three accounting statements
(balance sheet, pro…t & loss, and cash ‡ow), which include more than 100 …nancial variables
for each …rm. The analyses are based on all manufacturing …rms, representing 90% of all
…rms in the sample. Their exports total about 90% of Chinese manufacturing exports from
aggregate trade data.

2.1 Export Participation and Productivity

This section describes the overall patterns of Chinese manufacturing …rms’exporting behav-
iors. By comparing the statistics with those from US …rms,4 I demonstrate the systematic
di¤erences between Chinese and US …rms.

Table 1: Exporting Participation

Percentage of Manufacturing Firms
Export Status Chinaa USb Germanyc Franced Taiwane
Some Exports 29.6 21 44 17.4 46

Export Value / Gross Production

China US
22 14
a
The statistics are calculated from the Annual Census of Enterprises in China,
for all manufacturing …rms in 2005.
b
Bernard, Eaton, Jensen, and Kortum (2003), 1992 US Census of Manufactures.
c
Bernard and Wagner (1997), 1978 and 1992 Annual Survey of Establishments,
Lower Saxony State, Germany.
d
Eaton, Kortum, and Kramarz (2004), 1986 Customs and BRN-SUSE data
source.
e
Aw and Hwang (1994), 1986 Taiwanese electronics industry.

The …rst natural question would be: How many …rms export? Table 1 reports the overall
export participation for manufacturers. The fraction of exporting …rms in China is higher
than that in the US, but exporters are still in the minority.
3
I thank Liutang Gong in Beijing University for providing this data.
4
Statistics from other countries, if available, are also shown. Note that they are similar to the US …gures.

5
 For …rms that export, how much do they sell abroad? Table 2 reports the distribution
of export intensity –export values as the percentage of total output –for exporters. Notice
that the Chinese export intensity distribution is U-shaped: fewer than 20% of exporters sell
less than 10% of their output abroad, while about 40% of them export more than 90% of
their output.

Table 2: Export Intensity of Exporters

Percentage of Exporters
Export Intensity (%) Chinaa USb Germanyc
0 to 10 18.15 66 50
10 to 20 8.60 16
20 to 30 4.98 7.7
30 to 40 4.52 4.4
40 to 50 4.26 2.4
50 to 60 4.13 1.5 12.6
60 to 70 4.35 1.0
70 to 80 5.08 0.6
80 to 90 6.46 0.5
90 to 100 39.49 0.7
a
The statistics are calculated from the Annual Census of Enterprises in China,
for all manufacturing …rms in 2005.
b
Bernard, Eaton, Jensen, and Kortum (2003).
c
Bernard and Wagner (1997).

This is signi…cantly di¤erent from the US distribution. The second column of Table 2,
taken from Bernard et. al. (2003), indicates that two-thirds of US exporters sell less than
10% of their output abroad, and fewer than 5% of them export more than 50% of their
output.5
Are exporting …rms systematically di¤erent from non-exporting …rms in China? The
answer is yes, as demonstrated by Table 3.6 Chinese exporting …rms are larger than non-
exporters, though the di¤erence is not as pronounced as for their US counterparts. The
more striking …nding is that exporting …rms have lower labor productivity (about 10%) than
non-exporters. Again these patterns are di¤erent for US plants. US exporters are more
productive than their non-exporting counterparts with higher (more than 30%) sales per
worker and value added per worker.7
5
The third column for Germany should indicate that 50% of exporters export less than 15% of their
output; overall, 12.6% of exporters export more than 50% of their output.
6
In this table, I de…ne exporters relative to non-exporters as the arithmetic mean value of exporters
divided by the arithmetic mean of non-exporters to facilitate a direct comparison with the US …gures in
Bernard and Jensen (1995). However, since the data distributions are close to log-normal, a geometric mean
is probably more appropriate and is used in the rest of this paper.
7
One may argue that, since Chinese exporters use labor more intensively, di¤erences in value added per

6
 Table 3: Firms Characteristics
Exporters Relative to Non-Exportersa
China USb Germany France Taiwan
Total sales 2.92 6.48 5.87 28 15.41
Employment 2.59 4.38 3.84 8.17
Value added 2.74 13.35

Sales per worker 0.91 1.36 1.01 1.56

Value added per worker 0.86 1.39 0.99 1.22
a
Mean value of exporters divided by mean value of non-exporters.
b
Bernard and Jensen (1995), 1987 US Census of Manufactures.

These …ndings suggest that …rm exporting behavior in China exhibits a pattern di¤erent
from the United States and other countries. Why is this the case?

2.2 Determinants of Market Selection: Labor-Intensity of Firms

or Their Sectors
In this section I consider the explanatory power of the capital-labor ratio. I examine the
three questions in Section 2.1 by the labor-intensity of …rms or their sectors: How many …rms
export? How much do they export? What is the productivity di¤erence between exporters
and non-exporters?
Export Participation
To do so, I …rst rank sectors by their capital-labor ratio. Figure 1 plots the share of
exporting …rms in 2-digit sectors (on the y-axis) against the median capital-labor ratio in
the sector. There is a clear negative relationship between export participation and the
capital-labor ratio: more …rms in the sector export if the sector uses labor more intensively.
Distributions of Export Intensity
I then plot the histogram of export intensity in sectors grouped by their factor inten-
sity. Figures 2, 3, and 4 show the distribution of export intensity of exporters in sectors
with low, medium, and high capital-labor ratio, respectively. A common pattern is that
exporters in labor-intensive sectors export a large fraction of their output, while exporters
in capital-intensive sectors export only a small fraction of their output. The distributions
of export intensity shifts to the left as sectors’relative capital usage increases. In general,
the distribution of foreign-owned …rms (including Hong Kong, Macau, and Taiwan …rms,
worker in China cannot be directly compared to the same ratio in US, as it may misrepresent the di¤erences
in TFP. However, Chinese exporters on average are still less productive than non-exporters after controlling
for factor usage. See Section 2.2 for details.

7
 Figure 1: Share of Exporters vs K/L Across Sectors.

.8 Toys

Crafts
.6

Furs
Clothes

Computer
Fraction Exporting

Furniture
.4

Instruments
Textile
Rubber
Electric
Metal Plastics
Timber Tobacco
SpecialFoods
Machinery
Transportation Medicals
Foods Processing
RawChemicals
.2

Nonmetal Chemicals
Non-ferrous
Papermaking Beverage
Priting
Smelting
Petroleum
0

1 1.5 2 2.5
K/L in Sectors

Figure 2: Histogram of Exporters Export Intensity : Labor-Intensive Sectors

Clothes Furs
60

60
40

40
20

20
0

0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1

FDI or HK Firms Domestic FDI or HK Firms Domestic

Toy s Craf ts
60

60
40

40
20

20
0

0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1

FDI or HK Firms Domestic FDI or HK Firms Domestic

8
Figure 3: Histogram of Exporters Export Intensity: Medium Capital-Intensity Sectors
Instruments Textiles
40

40
30

30
20

20
10

10
0

0
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1

FDI or HK Firms Domestic FDI or HK Firms Domestic

Computer Plastics
40

40
30

30
20

20
10

10
0

0 .2 .4 .6 .8 1 0 0 .2 .4 .6 .8 1

FDI or HK Firms Domestic FDI or HK Firms Domestic

Figure 4: Histogram of Exporters Export Intensity: Capital-Intensive Sectors

Pharmaceuticals Chemicals
5 10 15 20 25

30
20
10
0

0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1

FDI or HK Firms Domestic FDI or HK Firms Domestic

Tobacco Petroleum
0 20 40 60 80 100

40
30
20
10
0

0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1

FDI or HK Firms Domestic FDI or HK Firms Domestic

9
and …rms in tari¤ free zones) are more concentrated to the right (which means they sell
relatively more abroad than domestic …rms). Yet they exhibit a similar left-shifting pattern
as we move from labor-intensive to capital-intensive sectors.
Productivity Di¤erences Between Exporters and Non-Exporters
I also examine average labor productivity (in terms of value added per worker) for ex-
porters and non-exporters across di¤erent sectors. The x-axes in Figures 5 and 6 are the
median capital-labor ratio of sectors. Figure 5 plots the value-added per worker for exporting
and non-exporting …rms in each sector. Notice that in labor-intensive sectors, non-exporters
have higher productivity than exporters, whereas the reverse is true in capital-intensive
sectors. Figure 6 plots the di¤erences in value added per worker between exporters and
non-exporters for each sector. The di¤erence between exporters and non-exporters increases
with capital intensity. The di¤erences are negative for labor-intensive sectors, meaning that
exporters have lower labor productivity in these sectors.8

Figure 5: Productivity of Exporters and Non-Exporters Across Sectors

3 2.5
VA per worker
2

Exporters
1.5

Nonexporters

1 1.5 2 2.5
K/L in Sectors

It is possible that a sector median is not a good classi…cation for a …rm’s technology
and capital-labor usage. To further explore the relationship between factor intensity and
productivity di¤erences, I rank …rms by their capital-labor ratio and group them into 100
bins. Within each group, …rms are choosing a very similar capital-labor ratio. In this case,
di¤erences in value added per worker represent di¤erences in …rm TFP.
8
The pattern is similar if we exclude SOEs. Among non-SOEs, exporters are less productive in labor-
intensive sectors, and the productivity di¤erence between exporters and non-exporters has a signi…cant
positive relation with the capital-labor usage. Therefore, SOEs are not the explanation of this relationship.

10
Figure 6: Productivity Di¤erences Between Exporters and Non-Exporters Across Sectors

.6
Exporters Advantage in VA per worker
-.2 0 .2 .4

1 1.5 2 2.5
K/L in Sectors

Figure 7: Productivity of Exporters and Non-Exporters Across Di¤erent K/L Bins

2 2.5
VA per worker
1.5

Exporters
Nonexporters
1

0 20 40 60 80 100
K/L

11
Figure 8: Productivity Di¤erences Between Exporters and Non-Exporters and the 95% con-
…dence intervals Across Di¤erent K/L Bins

.3 .2
Exporters Advantage in VA per worker
-.2 -.1 0 .1

0 20 40 60 80 100
K/L

Figure 7 plots the mean level of log value added per worker of exporters and non-exporters
across di¤erent capital-labor-ratio bins. Figure 8 shows the di¤erences between the value
of exporters and non-exporters and the 95% con…dence intervals. Again, the productivity
di¤erence between exporters and non-exporters has a signi…cant positive relation with the
capital-labor ratio. Exporters are more productive among capital-intensive goods, but not
among labor-intensive goods. Figure 9 reports the results from the same exercise excluding
SOEs. The pattern is similar, suggesting that SOEs are not generating the signi…cant positive
relationship observed.
Another possible productivity measure is sales. Under monopolistic competition and
CES preferences, more e¢ cient …rms have larger sales. Figure 10 plots the total sales of
exporters and non-exporters across di¤erent capital-labor-ratio bins. It is not surprising
that exporters have higher sales than non-exporters, as they enter into more markets. If
we consider sales in a single market, the domestic market only, we get Figure 11 (I also
plot the percentage of exporter selling in the domestic market for the readers’information).
Domestic sales of exporters are smaller than non-exporters for labor-intensive goods, but
larger for capital-intensive goods. Both sales and domestic sales …gures exhibit a systematic
relationship between the exporter-nonexporter di¤erence and capital-labor usage as seen
earlier.
The evidence suggests a systematic relationship between the capital-labor ratio and ex-
port patterns. Any good explanation of the overall pattern of export behavior should also

12
Figure 9: Productivity Di¤erences Between Exporters and Non-Exporters and the 95% con-
…dence intervals Across Di¤erent K/L Bins, Non-SOE Firms

.2
Exporters Advantage in VA per worker
-.2 -.1 0 .1

0 20 40 60 80 100
K/L

Figure 10: Total Sales of Exporters and Non-Exporters Across Di¤erent K/L Bins
5.5
5
Sales
4.5
4

0 20 40 60 80 100
K/L

Exporters Nonexporters

13
 Figure 11: Domestic Sales of Exporters and Non-Exporters Across Di¤erent K/L Bins

.9
.8
4.5
Domestic Sales

Percentage
.6 .7
4

.5
3.5

.4
0 20 40 60 80 100
K/L

Exporters Nonexporters
Percentage of Exporters Selling in Domestic Market

be consistent with this relationship.9

3 Model
To recap, Chinese exporters are less e¢ cient than non-exporters on average. This fact seems
to contradict the existing theories of export participation (Melitz (2003)), which predict
only the more e¢ cient …rms will export to the tougher foreign market. At the same time,
the systematic relationship between exporters non-exporters productivity di¤erences and
capital-labor ratio hints at an explanation driven by factor intensity, a mechanism which
is absent in the existing models. Indeed, when relative factor endowment in the domestic
and foreign markets di¤er, whether the foreign market is "tougher" depends on the factor
intensity.
In Section 3.1, I introduce endowment-driven comparative advantage into Melitz (2003)
and Chaney (2008) in a manner similar to Bernard, Redding, and Schott (2007). The
stripped-down model is used to give the intuition for why such a model can explain the
trade patterns in China and in developed countries.
The full model which adds market access costs (Arkolakis, 2008) and …rm-speci…c shocks
in entry costs (Eaton, Kortum, and Kramarz, 2009) is presented in Section 3.2. I show that
9
In Appendix A, I consider and reject several explanations of di¤erences between US and Chinese ex-
porters. Overall, SOE or foreign-owned …rms’behavior, tari¤ free zones, and processing trade explain little
of the opposite patterns observed in China.

14
the observations documented in Section 2, which seem to contradict the existing models,
actually …t the natural hybrid model, both qualitatively and quantitatively.

3.1 A Baseline Model

There are 2 countries, m; n 2 fCN; U Sg, that produce goods using capital and labor. China
is relatively labor abundant: K CN
LCN
<K US
LU S
. In this section, I restrict the analysis to a partial
wCN wU S
equilibrium model by assuming rCN < rU S , i.e., China has relatively cheap labor. A general
equilibrium analysis consistent with this assumption is deferred to Section 4.
To focus on di¤erences in factor costs, countries are assumed to have the same preferences
and technology. There are I sectors in each country.
Preferences
In both countries, utility is a Cobb-Douglas function of the I sector goods, with i 2 (0; 1)
being the share of spending on sector i.
Y
U= Ci i
.
i2I

Sector i consumption is a composite of individual varieties sold in the market,

Z 1
Z 1
1
1
i i i i 1
C = q (z) d (z) ; P = p (z) d (z) ,
i i

where is the elasticity of substitution across varieties. Assume is the same across countries
and sectors.
Technology
Firms within each sector have heterogeneous productivity. The measure of potential …rms
in country m and sector i who have productivity at least z is:

i
m (z) = Tm z .

Assume is the same across countries and sectors.10

The production function for a …rm with productivity z in sector i is:

y i (z) = zl i k 1 i ,

where i is the labor intensity of sector i. Sector i is more labor intensive than sector i0 if
10
Assumption of Pareto distribution makes the model analytically simple. Results hold for general pro-
ductivity distributions.

15
 i > i0 .
Trade Barriers
Selling goods in a foreign country incurs transportation cost d > 1 and tari¤ charge
> 0. Transportation costs are iceberg costs: for 1 unit of goods to arrive in n, dnm units
need to be shipped from m. On the other hand, tari¤s distort prices but do not entail a
physical loss of goods. Trade barriers can take many forms; here I only consider ‡at rate
tari¤s nm which are levied by country n on goods imported from m. Tari¤ revenues are
rebated as lump sum payments to households in n.
Firm Decisions
There is a …xed entry cost to sell in each market (country). Each …rm in each country
decides where to sell its output and what price to set in each market.
Pricing
For simplicity I drop the country index. The variable cost function for a …rm in sector i
1
is ci (z) = B w i rz i , where B = (1 ) 1.
In the home market the …rm sets a price that is the usual markup over unit cost: pi (z) =
1
ci (z). If it sells in a foreign country, it adjusts that price for tari¤s and trade costs:
pix (z) = d (1 + ) 1 ci (z). The potential revenue from selling in either market is

1
pi (z)
sin (z) = ( i Xn ) , n = CN; U S (1)
Pni

where Xn is the total spending of the market and Pni is the price index of the market.
Market Selection
Let Ed and Ex be the Chinese …rms’ entry cost to the domestic and foreign market
respectively. Note the entry costs here are the …xed costs in Melitz (2003). Because the
model here is static, there is no entry costs fe on free-entry condition.
The net pro…ts for …rm z from selling in the domestic and foreign market are:

i sid (z)
d (z) = Ed

i six (z)
x (z) = Ex .

Then the productivity cuto¤s of entering domestic and foreign markets are zdi and zxi ,
respectively:

sid zdi = Ed
six zxi = Ex . (2)

16
 Using the formula for the revenue (1) and combining the expressions of productivity
cuto¤s (2), we get the relationship between domestic and exporting productivity cuto¤s:
1
zxi Pdi Xd Ex 1
= d (1 + ) . (3)
zdi Pxi Xx Ed

Pdi
The literature to date assumes that countries are symmetric, in which case Pxi
= 1, then
zxi
zdi
> 1 due to extra trade costs or higher …xed costs for exporting. Because the existing
empirical evidence …nds that only the most productive …rms export, the productivity cuto¤
for exporting must be higher. But, in theory, this ratio can be less than 1. In particular, if
Pdi zxi
P i is low enough, then zi
< 1 and the domestic market is tougher than the foreign market,
x d
Pi Pi
so that less productive …rms sell to the foreign market. Pdi = PCN
i is decreasing in the labor
x US
intensity of the sector i (Proof is provided in Section 3.2). Therefore, it is more likely that
i
the price ratio is low enough in labor-intensive sectors in China to make zzxi < 1.
d

Figure 12: Productivity Cuto¤s for Market Entry

-
Capital intensive t t t
-
-
zx
zd
>1 0 zd domestic zx f oreign z

-
t t t
-
-
Labor intensive
zx 0 zx f oreign zd domestic z
zd
<1

Figure 12 summarizes Chinese …rms’decisions for the domestic and foreign markets and
for di¤erent sectors. For the labor-intensive sector, PPxd is very small and zzxd < 1.
Intuitively, since wages are generally higher in the main export destinations of Chinese
products than in China, labor-intensive goods are more expensive in those markets. With
frictional trade, the price indices of those goods in foreign markets decrease but are still
higher than that in the Chinese market. Hence, for labor-intensive goods, foreign markets
are less competitive than the domestic (Chinese) market. As a result, less productive …rms
sell only to the foreign market, while survivors in the more competitive domestic market
are more productive. The opposite is true for capital-intensive sectors: Capital is abundant
abroad, so foreign …rms have lower costs. Since the foreign market is more competitive,
only the most productive Chinese …rms can export. Less productive Chinese …rms in those
sectors can sell domestically, but they cannot export.

17
 At the same time, it can happen that for US …rms, the foreign market is tougher in
both the capital and labor-intensive sectors. Therefore, productive US …rms export in both
capital and labor-intensive sectors.11 For Chinese …rms, more productive …rms export in
capital-intensive sectors and less productive …rms export in labor-intensive sectors. Relative
market competitiveness, hence the productivity cuto¤s in (3) also depend on markets size
and entry costs. When the US market is larger or the entry cost is lower, the above case is
more likely to happen.

3.2 Full Model

The baseline model shows that augmenting the Melitz model to allow for factor intensity that
varies by industry can explain the very di¤erent exporting patterns in China and the US.
In this section, I follow Eaton, Kortum, and Kramarz (2009) to introduce …rm-by-market
speci…c shocks in entry costs or/and demand and incorporate Arkolakis’s (2008) formulation
of market access costs. The former helps explain market entry decisions, and the latter helps
explain export intensity. The extended model …ts the data quantitatively.
In the basic model, …rms enter markets according to a perfect hierarchy. Any …rm selling
to a tougher market necessarily sells to the softer market as well. Therefore, when the
domestic market is tougher, any …rm selling in the domestic market should also export.
On the other hand, when the foreign market is tougher, every exporter should sell in the
domestic market as well. Neither the French nor the Chinese data display this rigid pattern,
so in this section I incorporate …rm-by-market speci…c shocks to break the deterministic
ranking of markets.
In the Melitz model, the distribution of export intensity is degenerate. For a n-country
model, the distribution is a step function depending on the number of countries …rms sell
to. Introducing an endogenous marketing costs adds an intensive margin and shows how
relative toughness of the markets a¤ects …rms’sales in di¤erent markets.
Entry Costs
Consider a …rm in country m. In addition to its idiosyncratic productivity shock z, it
has an idiosyncratic shock "n to its marketing cost in country n. I assume …rm (z;~") pays a
11 Pi
Note even with symmetric market sizes and entry costs, Chinese …rms cuto¤ ratio is d (1 + ) Pcn
i and
us
i i 0
i
Pus zx zx
US …rms cuto¤ ratio is d (1 + ) i
Pcn with d (1 + ) > 1. Hence it can happen that zdi
< 1 and zdi
0 > 1
i
zx
for Chinese labor-intensive and capital-intensive goods, respectively, while zdi
> 1 for both US labor and
capital-intensive goods.

18
cost En ("n ) to sell goods in country n:

En ("n ) = "n En M (f ) , (4)

where En is the component of the cost common to all the …rms. "n is the …rm-speci…c …xed-
cost shock in market n. M (f ) is the cost for a …rm that choose to access a fraction f of
consumers in the market, and the function M (f ) is the same across destinations. I use the
speci…cation for marketing costs derived by Arkolakis (2008):
1
1 (1 f )1
M (f ) = 1 . (5)
1

The cost of reaching zero consumers in a market is zero. Total cost is increasing and
marginal cost is weakly increasing. Since it is harder to access a higher fraction of consumers,
we can expect more productive …rms to choose a larger fraction of consumers. Export
intensity depends on a …rm’s relative performance in domestic and foreign markets.
Results
I drop the country index for simplicity. The …rm’s optimization problem is:

s (z)
max (z; ") = f "EM (f ) . (6)
f

The FOC is:

s (z) "E
= 1 (7)
(1 f)

Figure 13 shows the marginal cost (right-hand side of the FOC (7)) and marginal revenue
(left-hand side of the FOC (7)) of accessing f fraction of consumers, given "E. Firms with
higher z have higher optimal f , which means more productive …rms choose to access more
consumers.
Optimal f
Using the formula for revenues (1) and the FOC (7), we get the optimal f
8 " # 9
< "E i
c (z)
1 =
f (c; ") = max 1 ;0 .
: iX 1 Pi ;

De…ne 1 1
i i
i X 1
1 i X 1
c (") = P =M P i, (8)
"E "E

19
 Figure 13: Optimal f
Optimal f
20

18 MC
16

14
MR
12 high z
10

8
MR
6 low z
4

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
f

1
which summarizes all the parameters of the market; M = 1
.
A …rm with cost c and shock " enters the market if

c ci (") , (9)

and chooses to reach f fraction of consumers

( 1)
c
f (c; ") = 1 i
. (10)
c (")

Both the entry and access decisions depend on the market’s pro…tability ci (") in (8): if
the price index P i is higher, the market size i X is larger, or the …xed cost "E is lower, then
ci (") is larger, and it is more pro…table and easier for …rms to sell in the market. Equation
(10) shows that, conditional on entering a market, lower cost …rms sell more not only because
they have lower prices, but also because they choose to spend more on marketing to access
a larger fraction of consumers.
Price Index
The price index for sector i in country n depends on prices of varieties from both countries
and weighted by market penetration f of the varieties. Speci…cally,
"Z ! #11
XZ cinm (")
Pni = f ("; c) c1 d nm (c) g (") d" , (11)
1 m 0

20
where

i
nm (c) = nm c ,
i 1
nm = Tm wmi rm i dnm (1 + nm ) . (12)

It can be shown (see EKK (2009)) that the price index in sector i country n is
1 1
1 Xni 1
Pni / i
n , (13)
En
i
where n = Tn wni rn1 i 1
+ Tm wmi rm i dnm (1 + nm ) (14)

is a weighted average of inputs costs at home and abroad for industry i.

The following proposition establishes the relationship between relative factor costs and
relative price indices.

Proposition 1 With frictional trade (d (1 + ) > 1), if CN is more labor abundant than the
US ( wrCN
CN
< wrUUSS ), and sector i is more labor intensive than sector i0 ( i > i0 ), then

0
PUi S PUi S
i
> i0
.
PCN PCN

That is, more labor-intensive goods have a higher relative price index in the US.

Proof. First note that 1 1

1 Xni 1
Pni i
n / i
n ,
En
where
i
n = Tn wni rn1 i 1
+ Tm wmi rm i dnm (1 + nm ) .

Let d1 = dus;cn (1 + us;cn ) > 1 and d2 = dcn;us (1 + cn;us ) > 1, then

0
PUi S PUi S
i
> i0
PCN PCN

is equivalent to

1 1 1 1
Tcn wcni rcn i
+ Tus wusi rus i d2 Tcn wcni0 rcn i0
+ Tus wusi0 rus i0
d2
> .
1 1 1i0 i0 i0 1 i0
Tcn wcn rcn
i i
d1 + Tus wus rus
i i
Tcn wcn rcn d1 + Tus wus rus

21
 i
wus rcn
Letting vi = wcn rus
, it follows that

Tus rus
1 d1 d2 (vi0 vi ) > 0.
Tcn rcn

Since d > 1, this is equivalent to

vi0 > vi .
wus rcn wus wcn
Hence wcn rus
> 1, i.e., rus
> rcn
. Therefore, if d = 1, (i.e., without trade frictions),
0 0
PUi S PUi S wus
i
wcn PU S PUi S
i
PCN
= i
PCN
0 = 1. With trade frictions, if rus
> , i
rcn PCN
> i0 .
PCN

Corollary 2 The competitiveness of the US market in sector i, relative to the Chinese mar-
ket, is decreasing in the labor intensity i .

Proof. 1
i
i X 1
c (") = Pi
"E
0
ci ci
Proposition 1 implies ciU S > ciU0 S for i > i0 . Thus the US market for labor-intensive goods
CN CN
is relatively (w.r.t. the CN market) less competitive than for capital-intensive goods.
Market Selection
Market selection is the same as in the simple model in Section 3.1. Tougher markets select
more e¢ cient …rms. Market toughness depends not only on …xed costs and transportation
costs, but also depends on market size and the competition in the market. For labor-intensive
goods, the foreign market is less competitive than the Chinese market in the sense that the
price index is higher in the foreign market. As a result, less productive …rms sell only to
the foreign market while survivors in the more competitive domestic market are the more
productive …rms. The opposite is true for capital-intensive sectors.
Export Intensity
The model also explains the distribution of export intensity among Chinese exporters.
Note that the expression for sales in the domestic market is:
1
c
sid = f ("; c) ( i X) . (15)
1 Pi

22
Then the export intensity for exporters is:
1
Pxi
six fx Xex
= 1
six + sid Pxi 1
fx Xex + fd (Pdi ) Xd
( 1) 1
cx Pxi
1 cix
Xex
= ( 1) 1 ( 1)
. (16)
cx Pxi c 1
1 cix
Xex + 1 cid
(Pdi ) Xd

Without marketing costs, conditional on selling in the market, fd = 1, fx = 1, and the

export intensity is a constant for all …rms. With marketing costs, f 2 (0; 1). For labor-
i
intensive sectors, cid < cix ; fd < fx , and exporters’ export intensity ri r+r
x
i is more likely to
x d
i i
be close to 1. In capital-intensive sectors, cx < cd ; fd > fx and exporters’export intensity
rxi
rxi +rdi
is more likely to be close to 0.
Intuitively, as the foreign market is less competitive than the domestic market for labor-
intensive sectors, the exporters are relatively more pro…table in the foreign market and choose
to access more consumers. A relatively larger fraction of output is sold in the foreign market.
Again the opposite is true for exporters in capital-intensive sectors.
Figures 14, 15, and 16 show a simulated example of productivity by market and the distri-
butions of export intensity, respectively. For Chinese …rms, non-exporters are more produc-
tive in labor-intensive sectors while exporters are more productive in capital-intensive sectors.
Figure 15 shows the histograms of Chinese exporters’export intensity, in labor-intensive and
capital-intensive sectors, respectively. They are consistent with the distributions in the Chi-
nese data. Figure 16 plots the histograms of US exporters’export intensity, where in this
example US exporters always sell mainly in the domestic market.
If foreign-owned …rms have lower entry costs to foreign markets (the distribution of the
entry shock has a lower mean) than domestic …rms, then the export intensity of foreign …rms
are always more concentrated to 1. Nonetheless, their export intensity distribution is still
bimodal (see the following Figure), which is again consistent with the observations in Section
2.2.

23
 Figure 14: Productivity of Firms Entering Di¤erent Markets (Model)
South Firms Average Productivity by Market
4.45
Domestic Market
4.4 Foreign Market

4.35

4.3

4.25
z

4.2

4.15

4.1

4.05

4
0.3 0.4 0.5 0.6 0.7
Kshare

Figure 15: Distribution of Chinese Firms Export Intensity, Labor-Intensive and Capital-
Intensive Sectors (Model)
Histogram of Export Intensity β=0.7Histogram of Export Intensity β=0.3

0.6 0.6

0.5 0.5
Share of Firms

Share of Firms

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0 0.5 1 0 0.5 1
Export Intensity Export Intensity

24
Figure 16: Distribution of US Firms Export Intensity, Labor-Intensive and Capital-Intensive
Sectors (Model)
Histogram of Export Intensity β=0.7Histogram of Export Intensity β=0.3

0.6 0.6

0.5 0.5
Share of Firms

Share of Firms
0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0 0.5 1 0 0.5 1
Export Intensity Export Intensity

H is t o g r a m o f E x p o β=
r t 0I n. 7t e n s it y H is t o g r a m o f E x p o β=
r t 0I n. 5t e n s it y H is t o g r a m o f E x p o β=
r t 0I n. 2t e n s it y
7 4 2 0

D o m e s ti c fi r m D o m e s ti c fi r m D o m e s ti c fi r m

F D I fi r m s 1 8 F D I fi r m s
F D I fi r m s
3 .5
6

1 6

3
5
1 4

2 .5
1 2
4

2 1 0

3
8
1 .5

6
2
1

1
0 .5
2

0 0 0
- 0 .2 0 0 .2 0 .4 0 .6 0 .8 1 1 .2 - 0 .4 - 0 .2 0 0 .2 0 .4 0 .6 0 .8 1 1 .2 1 .4 - 0 .2 0 0 .2 0 .4 0 .6 0 .8 1 1 .2

3.3 Estimation
In this section, I estimate the model to …t three moments for each sector: (1) the productiv-
ity di¤erence between exporters and non-exporters, (2) the fraction of exporters that only
export, (3) the mean export intensity.
Because I focus on sector variations, I use parameter values from EKK (2009) for 1 , ,
and the shock distributions. I assume for sector i that the cost threshold for selling in the
i
foreign market relative to the domestic market is ccxd = a bi, which means the relative
cost threshold decreases with the sector’s capital intensity. Figure 17, 18, and 19 shows how
the model …ts the three sets of moments taken from the data with only 2 parameters (a; b).

25
Figure 17: Data Versus Model: Di¤erences Between Exporters and Non-Exporters in Value
Added per Worker Across Sectors
Export productivity advantage in VA per worker
-.2 0 .2 .4

1 1.5 2 2.5
K/L
Data Model

Figure 18: Data Versus Model: Percentage of Exporters That Only Export Across Sectors
.6
Percentage of exporters only export
0 .2 .4

1 1.5 2 2.5
K/L
Data Model

26
 Figure 19: Data Versus Model: Average Export Intensity Across Sectors

.8 .6
Export intensity
.4.2
0

1 1.5 2 2.5
K/L
Data Model

4 General Equilibrium and Trade Liberalization

4.1 General Equilibrium
I now close the model by making factor returns endogenous, and use it to examine the
aggregate e¤ects of a tari¤ reduction on the decisions of individual …rms. To do so we need
to consider how such changes would a¤ect factor returns and prices through market clearing
conditions.
The factor allocation in sector i and country m satis…es12 :

i 1 i i
rm km = wm lm . (17)
i

Factor market clearing conditions are:

X
i
km = Km ,
i
X
i
lm = Lm . (18)
i

12
Results in Section 3 can be generalized to a model with N countries.

27
 Given factor prices, trade shares are:

1 i
Tm wmi rm i
dnm (1 + nm )
i
nm = , (19)
X
N
1 i
Tk wk i rk i
dnk (1 + nk )
k=1

i
where dnm is the iceberg costs for shipping goods from country m to country n, and nm is
the tari¤ applied by country n to goods imported from country m in sector i.
The goods market clearing condition for sector i is:

X
N
1
Ymi = i i
nm Xn i
,
n=1
1+ nm

where

Xni = i
Xn
!
X
I X
N i
i nk
Xn = wn Ln + rn Kn + n + Xn 1 ,
|{z} | {z } |{z} 1+ i
nk
i=1
Total spending factor returns …rms’net pro…ts | {zk=1 }
Tari¤ rebate to consumers

Ymi = wm lm
i i
+ rm km + m
i
.
|{z}
net pro…t

It can be shown that, for m = 1; :::; N and i = 1; :::; I, goods market clearing conditions can
be re-expressed as:
1 XN
(wn Ln + rn Kn ) 1
i i
wm lm = nm i I (20)
i n=1
X X i 1 + inm
N
i nk
1+ i
nk
i=1 k=1
P P
I normalize world GDP = m i Ymi to 1 (or we can normalize wage or rental in one
country to be 1). De…ne an equilibrium as follows:

De…nition 3 Given Km ; Lm ; inm ; inm ; i ; i , an equilibrium is a set of factor prices and

i i
factor allocations fwm ; rm ; km ; lm g that satisfy the factor allocation conditions (17), factor
market clearing conditions (18), and goods market clearing conditions (20).

Remark 4 There are I N factor allocation conditions (17), 2N factor market clearing
conditions (18), and I N goods market clearing conditions (20). Since the goods market
clearing conditions sum to an accounting identity, this is a system of N (2 + 2I) 1 equations
in the same number of unknowns.

28
4.2 E¤ects of Import Tari¤ Reduction
Proposition 5 If China reduces its import tari¤s, then:

1. Both Chinese and US trade volumes increase;

2. In China, labor moves to labor-intensive sectors;

3. Price indices in all sectors in China decrease relative to the US. Chinese markets become
more competitive, and exporters’productivity relative to non-exporters decreases;

4. Real returns increase for both factors in the US.

As a direct e¤ect of import tari¤ reduction, Chinese imports increase. Chinese exports
also increase through two indirect e¤ects: factor prices adjust to the deterioration of terms
of trade, and demand increases as foreigners have more pro…ts.
A reduction of trade frictions leads to an increase in the relative demand for a country’s
comparative advantage goods. Thus labor moves to labor-intensive sectors in China. The
relative nominal reward of the abundant factor also rises.
Relative price indices in both sectors in China decrease as it becomes easier for for-
eign …rms to enter, which increases competition and drives prices down. Less productive
…rms cannot survive and the average productivity of domestic …rms increase. As a result,
exporters’productivity advantage falls.
We can test these predictions using the observations when China joins the WTO in 2001.
As shown by Figure 20, China reduced the average tari¤ it applied on its imports by 30%
upon joining the WTO in 2001. On the other hand, the export tari¤ applied by the rest of
the world to China did not change much.
Figure 21 plots the changes in labor employment share by sector in China from 2000
to 2007. On the x-axis is the median capital labor ratio of the sector and on the y-axis
is the percentage change of labor employment share over the time period. Observe that
labor-intensive sectors have employment share gains while capital-intensive sectors have em-
ployment share losses. This is consistent with prediction (2) in Proposition 5 that trade
liberalization moves labor to the sectors with comparative advantage.
Figure 22 reports the average productivity (in terms of value added per worker) for Chi-
nese exporters and non-exporters across di¤erent capital-labor-ratio bins for the years 2000,
2005, and 2007. The stylized relationships between capital-labor ratio and productivity as
well as exporters (dis)advantage remain similar over time: average productivity (both ex-
porters and non-exporters) increases with capital-labor ratio, but the slope for exporters
is steeper than that of non-exporters. Therefore exporters have lower productivity than

29
 Figure 20: Tari¤ Rate from 1998 to 2007
Tariff applied by
16 China to World US to China EU to China

16

16
14

14

14
12

12

12
10

10

10
Percentage

Percentage

Percentage
8

8
6

6
4

4
2

2
0

0
1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008 1998 2000 2002 2004 2006 2008
year year year

Figure 21: Percentage Change of Sectors’Employment Share from 2000 to 2007.

Furniture
1.2
Percentage Change of the Employment Share

Computer
.6 .9

Furs
Timber
Plastics
Electric
.3

Clothes Instruments
Toys
Metal
Foods Processing
0

Crafts Machinery Foods

Non-ferrous
Textile
Transportation Priting
Rubber Medicals
Special Papermaking Petroleum
Smelting
RawChemicals
Nonmetal
-.3

Beverage Chemicals

Tobacco
-.6

1 1.5 2 2.5
K/L in sectors

30
 Figure 22: Productivity of Exporters and Non-Exporters Over Time
Year 2000 Year 2005 Year 2007

3
2.5

2.5

2.5
VA per worker

VA per worker

VA per worker
2

2
1.5

1.5

1.5
1

1
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
K/L K/L K/L

Exporters Nonexporters Exporters Nonexporters Exporters Nonexporters

Figure 23: Productivity Di¤erences Between Exporters and Non-Exporters Over Time
Year 2000 Year 2005 Year 2007
.4

.4

.4
.2

.2

.2
Exporters Advantage in VA per worker

Exporters Advantage in VA per worker

Exporters Advantage in VA per worker

0

0
-.2

-.2

-.2
-.4

-.4

-.4

0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100

K/L K/L K/L

31
Figure 24: Productivity Di¤erences Between Exporters and Non-Exporters Over Time
(Model)
Exporters ProductivityRelative to Non-exporters

0.09 High Tariff

Low Tariff
Exporters/Nonexporters
0.06

0.03

-0.03

0.4 0.5 0.6 0.7

Kshare

non-exporters for low capital intensity goods, but higher productivity for high capital inten-
sity goods. However, over the time period studied, non-exporters become more productive
relative to exporters.
Figure 23 displays the productivity of exporters relative to non-exporters across di¤erent
capital labor ratio bins. The productivity advantage of exporters clearly declines from 2000
to 2007, as predicted by part (3) in Proposition 5. I also simulated the model and plot the
simulated exporter advantage over time in Figure 24. By comparing it to Figure 23, we
can see that the overall pattern of the predicted exporters’advantage matches the empirical
observations as China enters the WTO.

4.3 Quantify the E¤ects of Tari¤ Reduction

In this section, I use Chinese and US data on capital labor endowments Km ; Lm , bilateral
trade inm , and tari¤ inm to calculate the e¤ects of import tari¤ reduction in China. I apply
the method used in Dekle, Eaton, and Kortum (2008). To get the baseline wm ; rm , I solve the
i i
factor prices and factor allocations fwm ; rm ; km ; lm g from (17), (18), and (20), given Km ; Lm
i i
and nm ; nm in 2001.
Denote the value after import tari¤ reduction as x0 and the change of any variable as

32
 x0
x^ = x
. We can rewrite the equilibrium conditions:

0 1 i 0
r^m rm kmi = w^m lmi
i
X 0
i
k m = Km
i
X 0
lmi = Lm
i

i
w^mi r^m
1 i di )
(1 +
0i nm nm
nm =
X
N
i
w^k i r^k
1 i di )
(1 +
nk nk
k=1

1 0i
X
N
0i (w^n wn Ln + r^n rn Kn ) 1
w^m wm lm = nm i 0i
i n=1
X X 0i
I N 1+ nm
i nk
0i
1+nk
i=1 k=1
n o
Given wm ; rm ; Km ; Lm ; i
; (1 di ) , solve
+
0 0
w^m ; r^m ; kmi ; lmi . Table 4 shows the re-
nm nm
sults.13

Table 4: Model Simulation for Tari¤ Reduction

Country
Changes China US
r
w
0.96 1.05
Relative Wage 1 1.02
Relative Rental 1 1.11
Price Index (Most Labor-Intensive Industry) 0.98 0.98
Price Index (Least Labor-Intensive Industry) 0.95 1.03
Real Wage 1.02 1.00
Real Rental 0.98 1.05

5 Conclusion
Firm-level trade patterns in China are di¤erent from those in the US and France, and they
may seem to be at odds with existing theoretical models. They can be explained, however,
by a hybrid model that considers relative factor abundance. This model encompasses the
existing trade model as a special case and explains the data from Chinese and developed
countries simultaneously. Tougher markets –either foreign or domestic markets –select the
13
In this exercise, I leave out the trade de…cit, which is an important issue for future work.

33
most e¢ cient …rms. Relative toughness of markets depends on the comparative advantage
as well as trade costs. Foreign market can be the main market for some …rms and exporters
can be less productive than non-exporters. This is more likely to happen in a country’s
comparative advantage sectors. The model’s predictions on the e¤ects of a tari¤ change are
also consistent with observations before and after China joined the WTO in 2001.
By looking at foreign-owned, private …rms and SOE separately, at …rms in tari¤ free
zones and other areas, as well as at share of processing trade across sectors, I provide some
reassuring evidence on concerns about Chinese government policies, but future work can try
to investigate this issue more fully. In addition, further quantitative analysis of the e¤ect
of trade liberalization on wages and welfare in both China and other countries would be
interesting, as it could aid trade policy formulation.

References
[1] Arkolakis, Costas (2008), “Market Penetration Costs and the New Consumers Margin
in International Trade,”Yale University.

[2] Aw, Bee Yan; Chung, Sukkyun; and Roberts, Mark (2000), “Productivity and Turnover
in the Export Market: Micro Evidence from Taiwan and South Korea,”The World Bank
Economic Review, 14(1), 65–90.

[3] Aw, Bee Yan; and Hwang, A.R. (1994), "Productivity and the Export Market: A Firm-
Level Analysis," Journal of Development Economics, 47, 313–332.

[4] Bernard, Andrew B.; Eaton, Jonathan; Jensen, J. Bradford; and Kortum, Samuel
(2003), “Plants and Productivity in International Trade,”American Economic Review,
93, 1268–1290.

[5] Bernard, Andrew B.; and Jensen, J. Bradford (1995), “Exporters, Jobs, and Wages in US
Manufacturing: 1976–87,” Brookings Papers on Economic Activity: Microeconomics,
67–112.

[6] Bernard, Andrew B.; and Jensen, J. Bradford (1999), “Exceptional Exporter Perfor-
mance: Cause, E¤ect, or Both?”, Journal of International Economics, 47, 1–25.

[7] Bernard, Andrew B.; and Jensen, J. Bradford (2004), “Why Some Firms Export;”
Review of Economics and Statistics, 86, 561–569.

34
 [8] Bernard, Andrew B.; Jensen, J. Bradford; and Schott, Peter K. (2006), “Survival of
the Best Fit: Exposure to Low Wage Countries and the (Uneven) Growth of U.S.
Manufacturing Plants,”Journal of International Economics, 68, 219–237.

[9] Bernard, Andrew B.; Redding, Stephen J.; and Schott, Peter K. (2007), “Comparative
Advantage and Heterogeneous Firms,”Review of Economic Studies 74, 31–66.

[10] Bernard, Andrew B.; and Wagner, Joachim; (1996), "Exports and Success in German
Manufacturing," MIT Dept. of Economics Working Paper No. 96–10.

[11] Chaney, Thomas (2008), “Distorted Gravity: Heterogeneous Firms, Market Structure,
and the Geography of International Trade,”American Economic Review, 98, 1707–1721.

[12] Clerides, Sofronis; Lach, Saul; and Tybout, James R. (1998), “Is Learning by Exporting
Important? Micro-dynamic Evidence from Columbia, Mexico and Morocco,”Quarterly
Journal of Economics, 113, 903–947.

[13] Eaton, Jonathan; Kortum, Samuel; and Francis Kramarz (2009), “An Anatomy of
International Trade: Evidence from French Firms,”NBER Working Paper.

[14] Fox, Jeremy; and Smeets, Valerie (2009), “Does Input Quality Drive Measured Di¤er-
ences in Firm Productivity?”Working Paper.

[15] Lentz, Rasmus; and Mortensen, Dale T. (2005), "Productivity Growth and Worker
Reallocation". International Economic Review, 46(3), 731–749.

[16] Melitz, Marc J. (2003), “The Impact of Trade on Intra-industry Reallocations and
Aggregate Industry Productivity,”Econometrica, 71, 1695–1725.

[17] Roberts, Mark; and Tybout, James R. (1997), “The Decision to Export in Colombia: An
Empirical Model of Entry with Sunk Costs,”American Economic Review, 87, 545–564.

[18] Romalis, John (2004), " Factor Proportions and the Structure of Commodity Trade,"
American Economic Review, 94(1) 67–97.

[19] Syverson, Chad (2010), "What Determines Productivity?" Working Paper.

[20] Van Biesebroeck, Johannes (2008), “The Sensitivity of Productivity Estimates: Re-
visiting Three Important Productivity Debates,” Journal of Business and Economic
Statistics, 26(3), 321–38.

35
 Appendix

A Firm Ownership and Processing Trade

This section considers and rejects several explanations of the di¤erence between US and
Chinese exporters.
Firm Ownership
A possible explanation for the di¤erence in export patterns between China and the United
States is the di¤erence in …rms’ownership structure: some manufacturing …rms in China are
state-owned. If state-owned enterprises behave di¤erently, this might cause the di¤erences
in overall export patterns.
Table 5 shows the ownership patterns and the export participation of di¤erent kinds of
…rms in China in 2000 and 2005.14 Note that the percentage of SOEs and collective …rms
falls from 2000 to 2005. However, export participation remains stable over the same period.

Table 5: Composition of Firms

2000 2005
Fraction(%) Exporters(%) Fraction(%) Exporters(%)
SOE 24 12 5 15
Collective 42 19 26 20
Private 15 20 47 21
Foreign 8 61 11 63
HK/Macau/Taiwan 11 60 11 62
Total 100 100

A closer look at the SOEs reveals that they have smaller sales, are more capital intensive,
but are less productive in terms of value added per worker. Most SOEs are non-exporters and
are less productive, which implies that among non-SOEs the patterns described above are
even stronger. Among non-SOEs, exporters are still less productive than non-exporters (see
Figure 9 for details). Thus the presence of SOEs cannot explain the low relative productivity
of exporters in China.
SOEs are also not the explanation for the U-shaped distribution of export intensity
in China. Figure 25 plots the export intensity for both SOEs and non-SOEs. While SOEs
typically export a smaller fraction of their output than non-SOE exporters, both distributions
are still bimodal.
14
Using the ownership classi…cation code in the data, I group …rms into …ve categories. The SOE in this
table includes those with ownership codes 110, 141, 143, 149, 151. Results for non-SOE in this section are
similar if I include all the …rms with positive state asset as SOE. Hong Kong, Macau, and Taiwan …rms are
200, 210, 220, 230, 240. Foreign-owned …rms are 300, 310, 320, 330, 340.

36
 Figure 25: Histogram of Export Intensity: SOE vs. Non-SOE

30
Percentage of firms
10 0 20

0 .2 .4 .6 .8 1
SOE non-SOE

Figure 26: Histogram of Exporter Export Intensity: Foreign-owned vs. Domestic Firms
40 30
Percentage of firms
10 200

0 .2 .4 .6 .8 1
FDI or HK Firms Domestic

Another possible explanation for the di¤erence in exporting patterns: many manufactur-
ers are foreign-owned …rms that serve as an exporting platform for foreign countries. Indeed,
Table 5 indicates that most of these …rms export. To see if this fact can explain the pattern,
I group …rms into domestic and foreign-owned …rms (for our purpose, Hong Kong, Macau,
and Taiwan …rms are considered as foreign-owned …rms since they receive the same treat-
ment as foreign investors) and plot the distribution of export intensity for both groups.15
The histogram is given in Figure 26. Foreign owned …rms do sell a larger fraction abroad
than domestic exporters. But among domestic …rms the distribution of export intensity is
still U-shaped, with a large fraction of …rms selling most of their output abroad. We lack
15
Including all the …rms with positive foreign asset as foreign-owned …rms does not change the result.

37
explanation for the group of …rms that export most of their output.
Tari¤ Free Zones
To further investigate the impact of processing trade, I also classify exporters by their
locations: in tari¤ free zones or other regions. Firms in tari¤ free zones can import inter-
mediate materials without a tari¤ and usually work on assembling or processing trade, with
their main markets being the foreign markets. Figure 27 plots the export intensity of …rms
in the two groups. Again, while exporters in tari¤ free zones do sell a higher fraction abroad
than exporters in other regions, the export intensity distributions are still bimodal. There
are still many …rms in other regions that export most of their output.

Figure 27: Histogram of Exporters Export Intensity: Tari¤ Free Zones vs. Other Regions
30
Percentage of firms
10 0 20

0 .2 .4 .6 .8 1
Tariff free zone Other regions

Overall, …rms ownership and location explain little of the di¤erence between US and
Chinese exporters.16
Processing Trade by Sectors
This section looks at Chinese processing trade by sector. Processing trade has been an
important part of China’s export. It accounted for more than half of its exports in recent
years. However, more than 70 percent of processing trade is handled by foreign-owned …rms.
I have shown that foreign-owned …rms and …rms in tari¤ free zones explain little for the
di¤erence in pattern between China and the US.17 This section examines processing trade
by sector and shows that the share of processing trade is not correlated with the variation (of
export intensity and productivity di¤erences between exporters and non-exporters) across
sectors.
16
Taking both foreign …rms and …rms in tari¤ free zones into account still explains little.
17
Ideally we can further investigate the impact of processing trade by analyzing well-matched …rm-level
data and custom data.

38
 Processing trade data are aggregated from Chinese custom monthly transaction data for
the year 2005.18 The custom data categorize trade into many types, and here "processing
trade" indicates the summed value of two classi…cations: "processing and assembly" and
"processed with imported materials". The share of processing trade in total trade for all HS
2-digit industries are given in Figure 28.

Figure 28: Share of Processing Trade in Total Trade (HS 2-digit)

1
.8
.6
.4
.2 0

-49

l 84 -83
-43

-63
4

-67
8

-71

-97
9
1-2

-40
7

8-3

-85

6-8
5-2

44

ctr ls 72
41

50

64

68

90
39
ls 2
ffs

n8
s2

od
er

es

ar

ass

us
stu

ta
rs

ath
ica

Wo

e
xtil

atio
ica
uct

e
be

eo
otw

/Gl
od

M
em

Le

Te
rod

ub

lan
ort
Fo

Fo

ne

Ele
Ch

s/R

nsp
lP

cel
Sto

ry/
era

stic

Mis
Tra
ne
Min

Pla

chi
Ma

Figure 29: Share of Processing Trade in Total Trade (ISIC 2-digit, sectors ranked by their
capital intensity)
1
.8
.6
.4
.2
0

tob oke
n e ach od

r
ipm ereytal

se
trate r
ne er
ma le s

utic ng

co
ure

als
ry

pe
e

min ort

mefoo

che al
the

ry

ilexrtil

tal
l
b
chi ath

era
ne
wo

ac
acerinti
en

pa
rub

mic

c
qu inm
nit

llic sp
clo

chi
fur

eta tran

arm p
ted
ma

nd
atio l m

rica
nm er
sa
ice

no oth
nic rica

ph
fab
icle
off

mu lect

veh
com e

By using the correspondence between HS1996 and ISIC Rev.3, I match the HS 6-digit
products to ISIC 4-digit sectors, then calculate the share of processing trade in total trade
18
I thank Anna Wang for providing this data.

39
for ISIC 2-digit sectors. Figure 29 ranks the 2-digit sectors (on the y-axis) by their capital-
labor ratio (from the …rm-level data) and plots the shares. The share of processing trade in
total trade is not correlated with sectors’capital-labor ratio, hence the variation of export
intensity and exporter productivity (dis)advantage across sectors.

B Robustness to Productivity Measures

This section shows that the patterns observed are robust to using alternative measures of
productivity.
Labor Quality
Labor quality may di¤er across …rms. A concern is that productivity will be mismeasured
if labor quality varies in a systematic way across …rms. We can use a …rm’s wage bill to
correct this source of variation in our productivity measure.19 De…ne the number of quality
adjusted workers for …rm j as
w (j) l (j)
l (j) = ,
w
P
j w(j)l(j)
where w(j)l(j) is the total wage bill of …rm j, and w = P is the average wage markets
P P j l(j)
paid. Quality adjusted workers satisfy j l (j) = j l (j).
VA
Value added per quality adjusted worker is de…ned as l(j) . Figure 30 shows the dis-
tributions of value added per worker and value added per quality adjusted worker. The
distribution of value added per quality adjusted worker is slightly more concentrated, but
not much. This …nding is consistent with that of Lentz and Mortensen (2005) and Fox and
Smeets (2009) for Danish …rms.20
For Chinese …rms, value added per quality adjusted worker is highly correlated with value
added per worker. Figures 31 and 32 plot the productivity of exporters and non-exporters,
and di¤erences between exporters and non-exporters in the di¤erent capital-labor-ratio bins
using value added per quality adjusted worker. They are similar to Figures 7 and 8.
The patterns observed are robust to various productivity measures (domestic sales, TFP,
value added per worker, and value added per quality adjusted worker). As van Biesebroeck
19
Fox and Smeets (2009) suggests that including the wage bill alone as a measure of labor inputs does
almost as well as including the full array of their human capital measures in their data.
20
Lentz and Mortensen (2005) suggests that di¤erences in the labor quality do not seem to be the essential
explanation for cross-…rm productivity di¤erentials. Correcting for wage di¤erences across …rms does reduce
the spread and skew of the productivity distribution somewhat, but both distributions have high variance
and skew, and are essentially the same shape.
Fox and Smeets (2009) uses Danish matched employer-employee data to control for workers education,
gender, experience, and industry tenure in the production function estimation. While these labor-quality
measures have signi…cant coe¢ cients in the production function, accounting for their in‡uence only decreases
the average within-industry 90-10 percentile productivity ratio from 3.74 to 3.36.

40
Figure 30: Distribution of VA per Worker and Quality Adjusted VA per Worker

.6 1
.8
Density
.4 .2
0

-4 -2 0 2 4 6
VA per Worker, log scale
VA per worker
Quality Adjusted VA per Worker
kernel = epanechnikov, bandwidth = 0.0334

Figure 31: Quality Adjusted VA per Worker: Exporters vs Nonexporters

Year 2005
2.2
Quality Adjusted VA per worker
1.4 1.6 1.8 2

0 20 40 60 80 100
K/L
Exporters Nonexporters

41
 Figure 32: Exporters Advantage in Quality Adjusted VA per Worker
Year 2005

Exporters Advantage in Quality Adjusted VA per worker

-.25 -.2 -.15 -.1 -.05 0

0 20 40 60 80 100
K/L

(2008) and Syverson (2010) point out, the inherent variation in …rm-level microdata is typi-
cally large enough to swamp small measurement-induced di¤erences in productivity metrics.
High productivity producers tend to look e¢ cient regardless of the speci…c way their pro-
ductivity is measured.

C Some Evidence from Indonesian Plant-Level Data

As another example of a developing country, Indonesian plant-level data also show some
patterns signi…cantly di¤erent from the US, France, and other countries in the literature,
but similar to China. Indonesia’s manufacturing survey is conducted annually and covers all
establishments with more than 20 employees.21
For exporters, how much do they sell abroad and how is the export intensity related to
…rm’s e¢ ciency? Figure 33 and 34 plot the distribution of export intensity for Indonesia
exporters. The dots in Figure 33 (Figure 34) indicate the mean value of log domestic sales
(log value added per worker) within each export intensity bin. Firstly, the distribution of
export intensity is signi…cantly di¤erent from the US, France, and other countries in the
literature. More than 30 percent of exporters only export and among them 75 percent do
not import. Second, …rm’s productivity in terms of sales in one market is decreasing with the
export intensity. In a standard n-country Melitz model and if foreign markets are tougher,
we would expect that more productive …rms have higher export intensity.
21
I thank Chang-Tai Hsieh for providing Indonesian plant-level data. Results shown in this section are
for the year 1996.

42
Figure 33: Distribution of Export Intensity for Indonesian Exporters and the Mean Domestic
Sales by Export Intensity

30

7 6.5
mean domestic sales
percentage of firms
20

6
10

5.5
0

5
0 .2 .4 .6 .8 1
Share of Exports in Output
# Firms Mean Domestic Sales

Figure 34: Distribution of Export Intensity for Indonesian Exporters and the Mean Value
Added per Worker by Export Intensity
4.2
30

4.1
mean VA per worker
percentage of firms
20

3.9 4
10

3.8
3.7
0

0 .2 .4 .6 .8 1
Share of Exports in Output
# Firms Mean VA per Worker

43
 In the sample, on average Indonesian exporters are more productive than their non-
exporting counterparts with higher domestic sales and value added per worker. But in
the comparative advantage sector –Furniture sector –exporters have smaller domestic sales
than non-exporters. Figure 35 plots the distribution of export intensity for the Furniture and
Chemicals sectors. The patterns are similar to what I found for Chinese sectors: comparative
advantage plays a role in …rms’market selection and export intensity.

Figure 35: Distribution of Export Intensity: Comparative Advantage Sector and Compara-
tive Disadvantage Sector
Furniture Chemicals
60

60
40

40
percentage of firms

percentage of firms
20

20
0

0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1
Share of Exports in Output Share of Exports in Output

D Simulation and Estimation

The simulation algorithm for each sector is the same as that in Eaton, Kortum, and Kramarz
(2010) except that I draw potential producers who sell in at least one market, which can be
the domestic or exporting market.
De…ne
u(j) = T z(j) , (21)

as …rm j’s standardized unit cost. The standardized costs have a uniform distribution that
doesn’t depend on any parameters.
The unit cost for a Chinese …rm in sector i in market n is:
1=
w i r1 i dn;CN (1 + n;CN ) u(j)
cin (j) = = i
: (22)
z (j) n;CN n

44
It follows that !1=
uin
cin = i i
. (23)
n;CN n

Substituting (8) and (13) for cin and Pni into (23), we get uin . Firm j will enter market n if
its u(j) and "n (j) satisfy:

i i
n;CN Xn
u(j) uin (j) = c1 "n (j) 1 . (24)
En

Conditional on …rm j’s passing this hurdle, …rm j’s sales in market n is:
" #
=e 1=e
u(j) u(j)
sin;CN (j) = "n (j) 1 En , (25)
uin (j) uin (j)

where e = 1 > 1:
Given values of the parameters, for each sector I create an arti…cial set of Chinese …rms.
Each …rms has the option to sell in the domestic or/and the foreign market.
The simulation and estimation steps are:

1. Fixed the realizations of the stochastic components of the model: Draw realizations
of v(j)’s independently from the uniform distribution U [0; 1], which will be used to
construct standardized unit cost u(j); Draw realization of …rm-by-country speci…c
entry shocks from log normal distribution.

2. Given the parameters for relative hurdles across industries a bi, I construct the relative
i 1
ux (j) "x (j)
entry hurdles ud (j)
= (a bi) "d (j)
for each …rm j.

3. Calculate u(j) = maxfun (j)g; the maximum value consistent with selling in some
market. Construct u(j) = v(j)u(j). A measure u (j) of …rms has a standardized unit
cost below u (j), so the arti…cial …rm j gets weight !(j) = u(j).

4. Calculate which indicates whether

n;CN (j), ( arti…cial …rm j enters market n, as deter-
1 if u(j) un (j)
mined by the entry hurdles: n;CN (j) = . Conditional on entry,
0 otherwise.
calculate sales in market n using (25).

5. Now we know for each arti…cial draw of …rms, its weight, where it sells, and how much
it sells. For each sector, I construct the following moments to match with the data:
the di¤erences of value added per worker between exporters and non-exporters, mean
export intensity among exporters, and the fraction of exporters that only export.

45