An Anatomy of China’s Trade Growth∗

Mary AmitiFederal Reserve Bank of New York and CEPR

Caroline FreundInternational Monetary Fund

Preliminary and Incomplete

Abstract

China’s exports grew 450 percent between 1992 and 2006. This paper decomposesChina’s export growth into the extensive margin (new varieties of exports) and theintensive margin (more exports of the same varieties). At the HS 6-digit level, we findthat nearly all of China’s export growth was in the intensive margin. Using highlydisaggregated U.S. trade data at the HS 10-digit level, we find that 5-15 percent ofexport growth was in new varieties; almost all of China’s export growth to the U.S.was in the intensive margin. Within these margins, for both exports to the U.S. and tothe world, we find that a small share of goods accounts for most of the trade growth.We also find that the skill content of China’s exports has increased.

Key Words:JEL Classifications:

∗Amiti, International Research, Federal Reserve Bank of New York, email: mary.amiti@ny.frb.org; Fre-und, Research Department, Trade and Investment Division, International Monetary Fund, 700 19th Street,Washington DC 20431, Ph 1-202-623-7767, Fax 1-202-589-7767 , email cfreund@imf.org. We would like tothank Jin Hongman of Customs, China for providing us with the data. Prepared for NBER conference. Theviews expressed in this Working Paper are those of the authors and do not necessarily represent those of theIMF or those of the Federal Reserve Bank of New York or the Federal Reserve System.

1. Introduction

China’s exports have increased by a factor 5 over the last 15 years. Was this growth ac-

complished by expanding trade of existing goods or by developing new export varieties?

Traditional theory highlights the expansion of existing products (the intensive margin) as

the only source of export growth. New trade theory gives a dominant role to an expansion

of the number of export varieties (the extensive margin), providing an additional channel

for welfare gains from trade. China’s dramatic export rise offers a unique opportunity to

evaluate these predictions.

This paper decomposes China’s export growth from 1992 to 2005 along various dimen-

sions. Using highly disaggregated trade data, we find that increased exports of existing

goods account for at least 85 percent of trade growth. We begin with HS 6-digit data on

exports from China to the world and find that the extensive margin is responsible for less

that one percent of trade growth. However, at the 6-digit level, many new export products

may go unobserved because the level of aggregation is not fine enough. To ensure that our

results are not driven by aggregation, we also use U.S. 10-digit import data. These data will

show a larger extensive margin both because the level of disaggregation is finer and because

bilateral data define all products that are new in the U.S. market as new exports. Despite

a forty percent increase in the number of varieties imported from China since 1992, we find

that the extensive margin accounts at most 15 percent of China’s export growth to the U.S.

In sum, most of China’s export growth between 1992 and 2005 occurred in products that

were already traded in 1992.

Still, there is evidence of a restructuring of China’s exports. Ranking (6-digit) exports

from smallest to largest in 1992, the export share of the top 10 percent of products declined

substantially, with an offsetting export share increase in the next 40 percent of products.

The change in the export share of the bottom 50 percent of products was negligible. We

also show that China’s export rise was associated with an increase in specialization. The

2

top 100 exports in 2005 account for 50 percent of trade, as compared with just 45 percent

of exports for the top 100 in the early 1990s. Finally, we find that China’s exports have

become more skill intensive. High-skilled products increased from just over 10 percent of

exports in the early 1990s to one-third of exports today.

Our results are supportive of traditional theory, which places endowments, productivity,

and terms of trade effects at the center of trade growth. This has important welfare impli-

cations: it limits the potential gains to importers from China’s expansion, while expanding

the potential losses to China’s competitors.

The importance of the intensive margin stands in sharp contrast to other recent empir-

ical papers. In seminal work, Feenstra (1994) provides insights into how to measure the

intensive and extensive margin. Using the Feenstra index, Hummels and Klenow (2005)

find that variation in the size of exports across countries are largely due to variation in the

extensive margin, suggesting that countries expand along the extensive margin as exports

grow. However, their results are not directly comparable with ours since the forces that

generate product distribution across countries may be quite different from the forces at work

within countries.

Also using the Feenstra index, Broda and Weinstein (2006) examine U.S. import growth

from 1972-2001. They find that 30 percent of growth was the result of an expansion in the

extensive margin, and that China was the largest contributor to growth in U.S. varieties.

While their results are suggestive of a larger role for the extensive margin, they also find that

the bulk of growth in new varieties was in the period before 1988. Overall, the extensive

margin was responsible for just 5 percent of U.S. import growth from 1988-2001.

In related work, Kehoe and Ruhl (2002) explore the responses of the intensive and ex-

tensive margin to trade liberalization. They find that trade liberalization leads to a large

increase in the share of trade accounted for by goods that were little traded before liber-

alization, and interpret this as evidence that the extensive margin has increased. We also

analyze how trade shares adjust; however, we argue that it is not clear that this can be

3

interpreted as growth of the extensive margin because the bottom 10 percent of goods cover

more than half of the product categories that are exported in both periods.

The rest of the paper is organized as follows. Section 2 sets out the methodology; and

section 3 describes the data. The results are presented in section 4, and section 5 concludes.

2. Methodology

This section describes the methodology employed to calculate the intensive and extensive

margin.

Let Vti be the value of trade at time t in product i (Vti = ptiqti ).

IEt0 is an indicator variable that is one if the product was exported in both period t and

period 0 (existing products).

IDt0 is an indicator variable that is one if the product was exported in period 0 and not

in period t (disappearing products).

INt0 is an indicator variable that is one if the product was exported in period t but not in

period 0.

Pi Vti −

Pi V0iP

i V0i=

PVti(I

Et0)−

PV0i(I

Et0)P

V0i−P

V0i(IDt0)P

V0i+

PVti(I

Nt0 )P

V0i(2.1)

This is an identity where total growth in trade relative to the base period is decomposed

into three parts: (i) the growth in products that were exported in both periods, the intensive

margin; (ii) the reduction in export growth due to products no longer exported, disappearing

goods; and (iii) the increase in export growth due to the export of new products, the extensive

margin. The disappearing category can either be counted separately or incorporated into

the intensive margin, as they are part of old categories.

4

3. Data

We use bilateral trade data at HS 8-digit level in current U.S. dollars from 1992-2006, source:

China Customs, Beijing. We aggregate the data into a panel at the 6-digit level, as there

are reclassifications of 8-digit codes every year. After adjusting the data by the U.S. CPI to

generate a constant dollar series, we create a 7-period panel of 2-year averages in order to

reduce idiosyncrasies that may occur in any given year.

There was a change in HS classifications in 1996 and 2002. We use the appropriate

concordances to convert all data to HS-1992 6-digit data. We also use mirror import data

from the world (at HS 6-digit 1992 codes) and from the United States (HS 8-digit and HS

10-digit) to check the robustness of the results.

4. Results

Overall, China’s real export growth was just over 450 percent from the first period (1992-

1993) to the last period (2004-2005). In this section, first, we decompose the growth into

the intensive and extensive margin. Second, we examine the extent to which specialization

was important in trade growth. Third, we explore whether growth was broad based or fueled

by a few products. Finally, we examine how the skill intensity of exports has changed.

4.1. Intensive vs. Extensive Margin

We calculate the extensive and intensive margin, using equation 2.1. To calculate the share

of trade growth attributed to each margin, we divide each component by the total growth in

trade. Table 1 records the intensive and extensive share of trade growth. From 1992-2006,

the extensive margin accounts for less than one percent of total trade growth, using 6-digit

HS 1992 data from China or the mirror import data.

To examine whether there is more variety growth to a single market and more disaggre-

gated data we recalculate the margins with U.S. data. The extensive margin is more likely

to be present in bilateral data, as exports can expand by entering new markets with old

5

varieties. Additionally, part of the reason the extensive margin estimated above may be

small is because 6-digit data may not be detailed enough to observe exports of new varieties.

There is, however, an intrinsic problem using 8-digit or 10-digit data, as there are numerous

reclassifications over time. We follow Debaere and Mostashari (2006) and consider only cat-

egories that are permanently present in the U.S. trade data. This will reduce the problem

of product reclassifications, assuming that reclassifications do not happen among incumbent

product codes. The permanent categories at the HS10-digit account for 65 percent of the

total HS codes and 47 percent of total exports between 1992 and 2005; for the HS 8-digit

codes, the permanent categories at the HS 8-digit account for 75 percent of the total HS

codes and 63 percent of total exports. The last four rows of Table 1 report the extensive

margin using U.S. data. Using HS 6-digit data, we still observe a small extensive margin,

with less than 2 percent of trade growth attributable to new products. The next two rows

show the results using the permanent codes at the HS 8-digit and HS 10-digit level. As would

be expected, the extensive margin is larger when we use more disaggregated data; however,

it is still quite small, with the intensive margin accounting for more than 95 percent of trade

growth.

An important concern is that concorded data will understate the extensive margin if new

exports are also primarily in new classifications. The last row of Table 1 reports growth in ex-

tensive margin using a traditional Feenstra index of variety on unconcorded HS 10-digit data

from the United States. Specifically, we calculate the expenditure share in the last period

on goods available in both the first and last period relative to the expenditure share in the

first period on goods available in both periods (Feenstra index of variety = Vti(IEt0)/ Vti

V0i(IEt0)/ V0i).

The index will be equal to one if there is no growth in varieties relative to the base period

and less (greater) than one if the number of varieties has grown (declined). Variety growth

is 1-Feenstra index. This measure has the nice feature that if classifications are simply split,

and their share of total trade remains unchanged, the index remains unchanged. However,

if growth classifications tend to be split to a greater extent than shrinking classifications are

6

merged, the index will tend to overstate the extensive margin. Using this index we find an

extensive margin of 15 percent. We think of this as an upperbound on the extensive margin.

Our estimates of the extensive margin in trade are considerably smaller than has been

found in previous literature. Exports of new varieties (at the HS-10 digit) from China to the

U.S. accounted for 40 percent of the total number of categories in 2004/2005, however these

new varieties account for at most 15 percent of the total value of China’s exports to the U.S.

4.2. Diversification versus Specialization

Next, we examine whether exports became more or less specialized over the period. Figure

1a shows inverse cumulative trade shares for all products. We graph it as an inverse function,

with products ranked from largest to smallest, in order to focus on the top 1000 and 500

products. Figure 1b and 1c magnify the image, showing the cumulative trade shares when

we keep only the top 1000 and 500 categories, respectively. The figures show that overall

trade diversification is roughly unchanged; however, if we look at the top 500 products,

which account for nearly 80 percent of trade in each period, concentration has increased.

To examine specialization in more detail, we estimate a Gini coefficient of export equality

in each period. It is defined as

Gini = 1− 1n

Xi

(csharei−1 + csharei),

where there are n products, i is a product’s rank (1 is smallest and n is largest), and csharei

is the cumulative share of exports of the ith product. The Gini coefficient uses the trapezoid

approximation to calculate the area between a 45 degree line and the cumulative distribution,

weighting each industry as an equal share of the population of industries (1/n). A Gini

coefficient of one reflects complete inequality and a Gini coefficient of zero reflects complete

equality.

Table 2 reports the Gini coefficient for the first and last period for various different

samples. When we include all products, the Gini coefficient remained constant over the

period at 0.85. Ranking the goods and including only the largest 70 percent, the Gini

7

coefficient increased from 0.46 to 0.55. Alternatively, examining the top 100 products, which

account for nearly 50 percent of trade in the second period and 45 percent in the first period,

the Gini coefficient increased from 0.35 to 0.50. Thus, over the period we see enhanced

specialization, especially for the products that accounted for the largest export shares.

An alternative way to examine specialization is to graph ln trade versus ln rank (the

highest trade share receiving a rank of 1), a kind of Zipf’s law for trade. We use the top 70

percent of trade in each period to eliminate the very small values of highly ranked products.

Figure 2 shows the plot for the first, fourth, and final periods. As shown in the figure, the

plot gets somewhat steeper over time (statistical testing confirms this), implying that moving

to a lower rank is associated with a greater percentage increase in trade in the final period

than in the first period.

4.3. Stable or Spotty growth

We now know that growth was mainly of goods that were traded in the first period (tiny

extensive margin). We also know that growth led to a distribution of exports that was

more skewed at the top. This picture could be generated by different growth patterns

across products. To understand the growth process we examine the distribution of growth in

more detail. Figure 3 shows the cumulative share of trade growth by the cumulative share

of products, when growth is ranked from smallest to largest. Just 20 percent of product

categories (or 478 products) make up more than 90 percent of trade growth. The top 5

percent of products account for nearly 30 percent of trade growth. Thus, we know that

growth was skewed, with a small share of products accounting for most of the growth.

Next, we determine where these products lie in the initial distribution. Figure 4a shows

a plot of rank in period 1 versus rank in the final period to see whether there is a lot of

churning. It shows that rank is highly correlated, but also that there are numerous outliers.

Figure 4b shows the cumulative export share in the last period when products are ranked

by the cumulative export share in the first period. This picture shows that the bottom part

8

of the distribution does account for a sizeable share of trade growth.

Figure 5 examines the reallocation of trade from 1992-3 to 2004-5 in more detail. Figure

5a divides trade into deciles according to the value of trade in 1992-3. We then examine what

share of trade in 2004-5 is accounted for by the varieties falling into each of these deciles, as in

Kehoe and Ruhl (2002). The figure shows that the categories that accounted for the bottom

two deciles have increased their share of trade, while the categories in the other deciles have

contracted. Figure 5b divides trade into deciles according to the categories of trade in 1992-

3. For example, the tenth decile is the top ten percent of product categories when products

are ranked by value. It shows that the reallocation was largely in the top 50 percent of

categories; products in the top decile shed trade while the next four deciles expanded. How

are these two pictures consistent? The answer is that trade shares are highly skewed and

the bottom 20 percent of trade by value accounts for 88 percent of product categories. In

sum, trade value of the mid-upper rank products expanded the most. These are products

that were in the bottom 20 percent by value and in the mid-range by rank.

4.4. Skill Intensity of Growth Products

Finally, we examine whether the skill content of China’s export growth has increased over

the sample period. Assuming no factor intensity reversals, we use the U.S. data on the

ratio of production workers to total employment from the NBER SIC database for 1992 to

rank industries from low to high skill on the horizontal axis, and then plot the cumulative

export share on the vertical axis, following Chun and Trefler (2005). Figure 5 show that

in 1992/1993, 80 percent of China’s exports were accounted for by 40 percent of the least

skill-intensive industries. The shift of the curve to the right indicates that the skill content

of China’s exports has increased over time. For example, in the first period, 50 percent of

the least skill-intensive industries produced 88 percent of China’s export share. By the last

period, the export share that these industries produced fell to 66 percent.

9

5. Conclusions

China has experienced rapid export growth over the last 15 years. We find that this growth

was largely driven by expanding trade in existing goods. Despite the large growth in number

of varieties, these only account for a small share of China’s trade growth. Even using the

most disaggregated trade data (HS-10 digit) available for the United States, we find that the

new varieties only accounted for at most 15 percent of total export growth over a 15 year

period.

We also find evidence of export specialization and an increase in the skill intensity of

exports.

References

[1] Broda, C. and D.Weinstein (2006). “Globalization and the Gains fromVariety”Quarterly

Journal of Economics, 541-585.

[2] Chun Zhu, S. and D. Trefler (2005). “Trade and Inequality in Developing Countries: A

General Equilibrium Analysis", Journal of International Economics 65, 21-48.

[3] Debaere, P. and S. Mostashari (2006?) “Do Tariffs Matter for the Extensive Margin of

International Trade? An Empirical Analysis” Mimeo University of Texas, Austin.

[4] Feenstra, R. and H.L. Kee (2006) “Trade Liberalization and Export Variety: A Compar-

ison of Mexico and China”. Mimeo World Bank.

[5] Hillberry, R. and C. McDaniel (2002) “A Decomposition of North American Trade

Growth Since NAFTA”, ITC Working Paper 2002-12-A.

[6] Hummels, D. and P. Klenow (2005). “The Variety and Quality of a Nation’s Exports”

American Economic Review, 95, 704-723.

10

[7] Kehoe, T. and K. Ruhl (2002) “How Important is the New Goods Margin in International

Trade?” Federal Reserve Bank of Minnesota.

11

12

Table 1: Extensive and Intensive Margin of China's Exports, 1992 to 2005 Extensive Intensive Data

0.2 99.8 China, 6-digit HS 92 0.1 99.9 Mirror, 6-digit HS 92 1.1 98.9 U.S., 6-digit HS 92 2.2 97.8 U.S. permanent, 8-digit HS

4.9 15.0

95.1 85.0

U.S. permanent, 10-digit HS U.S. permanent, all codes (Feenstra Index)

Table 2: Gini Coefficient for China's Exports Period All Top 70% Top 100

1992-1993 0.85 0.46 0.35 2004-2005 0.86 0.55 0.50

13

Figure 1: China’s Export Specialization 1a. All Products

020

4060

8010

0

Cul

ulat

ive

Sha

re o

f Tra

de

0 1000 2000 3000 4000 5000

Rank of Product

Cumulative share of trade, 1992-1993Cumulative share of trade, 2004-2005

Note: Data uses HS 6-digit classifications. Rank is largest to smallest by value.

1b. Top 1000 Products

020

4060

8010

0

Cul

ulat

ive

Sha

re o

f Tra

de

0 200 400 600 800 1000

Rank of Product

Cumulative share of trade, 1992-1993Cumulative share of trade, 2004-2005

Note: Data uses HS 6-digit classifications. Rank is largest to smallest by value.

1c. Top 500 products

2040

6080

100

Cul

ulat

ive

Sha

re o

f Tra

de

0 100 200 300 400 500

Rank of Product

Cumulative share of trade, 1992-1993Cumulative share of trade, 2004-2005

Note: Data uses HS 6-digit classifications. Rank is largest to smallest by value.

14

Figure 2: Zipf’s Law for Trade

1820

2224

ln(tr

ade

valu

e)

0 2 4 6ln(rank)

ln(trade value), 1992-1993ln(trade value), 1998-1999 ln(trade value), 2004-2005

Note: Data uses HS 6-digit classifications.

15

Figure 3: The Distribution of Trade Growth Across Products

0.1

.2.3

.4.5

.6.7

.8.9

1C

umul

ativ

e sh

are

of g

row

th

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1Share of product categories

Note: Data uses HS 6-digit classifications.

16

Figure 4: Trade Growth and the Reallocation of Trade

4a. Rank Correlation of Trade, 1992-93 vs. 2004-05

010

0020

0030

0040

0050

00

Ran

k of

pro

duct

, 200

4-20

05

0 1000 2000 3000 4000 5000

Rank of product, 1992-1993Note: Data uses HS 6-digit classifications. Rank is smallest to largest by value.

Figure 4b. Adjustment in Export Shares Over Time

0.1

.2.3

.4.5

.6.7

.8.9

1C

umul

ativ

e E

xpor

t Sha

res

2004

-200

5

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1Cumulative Export Shares 1992-1993

Export Share

Note: Data uses HS 6-digit classifications.

17

Figure 5: Trade Shares and the Reallocation of Trade

Figure 5a. By Share of Value

0.0

5.1

.15

.2.2

5S

hare

of t

otal

val

ue in

200

4-20

05

1 2 3 4 5 6 7 8 9 10

Note: Data uses HS 6-digit classifications.Decile by value, 1992-1993

Figure 5b. By Share of Product Categories

0.2

.4.6

.8S

hare

of t

otal

val

ue

1 2 3 4 5 6 7 8 9 10

Note: Data uses HS 6-digit classifications.

Decile by number of products, 1992-1993

Share of total value, 1992-1993

Share of total value, 2004-2005

18

Figure 6. The Skill Intensity of China’s Exports

020

4060

8010

0C

umul

ativ

e ex

port

shar

e

0 .2 .4 .6 .8 1Skill intensity

Cumulative export share, 1992-1993Cumulative export share, 2004-2005

Note: Data uses HS 6-digit classifications.