EXPORT SPILLOVER OF FDI IN MANUFACTURING INDUSTRY IN CHINA

Sizhong Sun*

Crawford School of Economics and Government College of Asia and the Pacific

The Australian National University

Phone: 61-2-6125 0179 Fax: 61-2-6125 0767

Email: sizhong.sun@anu.edu.au

* The author would like to thank A/Prof. Ligang Song, Prof. Peter Drysdale, and Prof. Martin Richardson for their constructive suggestions and supervision. Comments and suggestions from ANU Phd Student Seminar and the 19th ACESA Annual Conference 2007 are gratefully acknowledged. All errors are my own.

1

Preface

Title of Thesis: Spillover of FDI and its Determinants in China

Supervisor: A/Prof. Ligang Song

Advisors: Prof. Peter Drysdale, Prof. Martin Richardson

For over two decades, Chinese economy has been growing at an impressive speed, in which

the foreign direct investment (FDI) has been playing an important role. To better

understand FDI’s role, my thesis aims to examine systematically the spillover of FDI. My

thesis first looks at the technology spillover from FDI at both the industry level and firm

level, namely whether FDI promotes domestic firms’ technology level. Then the thesis

examined the export spillover from FDI, namely whether FDI promotes domestic firms’

export.

My thesis is structured as follows:

Chapter 1: Introduction

Chapter 2: Literature Review

Chapter 3: Technology Spillover of FDI at Industry Level

Chapter 4: Technology Spillover of FDI at Firm Level

Chapter 5: Export Spillover of FDI: Cross-sectional Evidence

Chapter 6: Export Spillover of FDI in Non-ferrous Metal Smelting and Pressing

Industry: Evidence from Panel Data

Chapter 7: Policy Implication and Conclusion

Appendix

The following paper is based on Chapter 5.

2

Abstract

This paper studies the export spillover of FDI and its determinants in China, using a firm

level cross-sectional micro-data that consists of over 180 thousands firms in the

manufacturing industry in 2003. The paper first shows in a theoretical model that if FDI

invested firms can reduce domestic firms’ export cost, then FDI will promote domestic

firms’ export. This hypothesis is tested using the sample selection model in which firms’

decisions on whether to export and on how much to export are dependent on each other. It

is found that FDI does generate export spillover, and the scale of spillover depends

positively on firms’ age, size, and ownership structure, and negatively on firms’ average

wage and geographical location.

3

Export Spillover of FDI in Manufacturing Industry in China

Sizhong Sun

It is generally argued that FDI’s (foreign direct investment) presence in the industry usually

promotes domestic firms’ export (e.g. Alvarez, 2006), i.e. there exits export spillover from

FDI invested firms. Compared with domestic firms, FDI invested firms usually possess

some advantages (Dunning et. al., 1988, 1990), for example they are more technologically

superior and have expertise in international business and knowledge about foreign market

etc. These advantages to some extent can spill over to domestic firms by labour mobility or

domestic firms’ learning etc. Hence the export costs, particularly the entry cost, of domestic

firms can be lowered by FDI’s presence, and subsequently their exports are promoted.

Empirically Alvarez (2006) found a positive and significant impact of foreign capital on

firms’ exports in Chilean manufacturing industry, and Athukorala et. al. (1995) found a

positive impact of the third-world multinationals on firms’ export decision in Sri Lanka

manufacturing industry.

Will this be the case in China? Furthermore if they do what determines such spillover? This

paper tries to address these two questions. This paper is organized into 5 sections: Section 1

builds a theoretical model to show that if FDI’s presence reduces domestic firms’ export

cost, for example by knowledge spillover, then domestic firms’ export intensity will be

promoted; Section 2 deploys the econometric specification and estimation strategy used to

test the hypothesis in Section 1; Section 3 describes the data set used and constructs

variables; In Section 4, the empirical results are discussed; and Section 5 concludes the

paper.

1. A simple model of firms’ export behavior

In order to capture the impact of FDI on domestic firms’ export activity, i.e. the export

spillover of FDI, a simple model of firms’ export behavior is developed in this section. In

an industry that contains N firms, which include ( )γ−1N domestic firms and γN FDI

invested firms, where γ denotes the foreign presence in the industry. The FDI invested

firms are invested by foreign firms and are able to choose their output quantity and export

4

intensity independently from their parent firm1, i.e. they can act like a domestic firm. All

firms are homogenous and can sell their products at both the domestic market and foreign

market. At the domestic market, firms play Cournot, and have inverse demand functions as

follows:

( )Qpp = , 0' <p

where Q is the domestic demand and , is firm i’s export intensity, i.e.

the share of export in its output, and denotes firms i’s output. The world market is a

competitive market, and firms are faced with world price P.

∑=

−=N

iii qsQ

1)1( is

iq

In the course of production and export, costs are incurred respectively. For the production

process, firm i's cost function is ( )iqC with ( ) 0' >iqC . For the export process, firm i's

export cost function is with , , , , and

. , shows that firms’ export cost is increasing in its export quantity,

and the speed is increasing. , shows that for given firms’ export quantity

fixed, its export cost is decreasing in FDI invested firms’ activities in the industry, but the

speed of decreasing is also decreasing. states that if FDI invested firms’ activities

in the industry increase, then decreases, which means that for a given fixed increase in

export quantity, the increase in export cost will be smaller due to the increased activities of

FDI invested firms. If firm i is domestic firm, then it enjoys export spillover from FDI

invested firm as its export cost is reduced by the presence of FDI invested firms. The

channel of such benefit can be various, for example domestic firms can learn from FDI

invested firms’ export behavior and subsequently reduce their transaction costs in export.

However such channels are assumed to be exogenous and hence not modeled here.

⎟⎟⎠

⎞⎜⎜⎝

⎛= ∑

=

γN

jjii qqsEE

1

, 0'1 >E 0''

11 >E 0'2 <E 0''

22 <E

0''12 <E 0'

1 >E 0''11 >E

0'2 <E 0''

22 <E

0''12 <E

'1E

1 By this, we can abstract away from the parent firm’s decision. However this is not a strong assumption, as the world market is competitive and the FDI parent firms earn zero profit there and hence the FDI parent firms’ global profit maximization is equivalent to the domestic market profit maximization.

5

Firm i's problem is to choose its output quantity and export intensity to maximize its profit,

given all the other firms’ output quantity and export intensity, as follows:

{ }( ) ( ) ( ) ⎟⎟

⎠

⎞⎜⎜⎝

⎛−−+⎟

⎠

⎞⎜⎝

⎛−−=Π ∑∑

==

γN

jjiiiii

N

iiiiiisq

qqsEqCPqsqspqsii 11,

,11max

Then domestic firms’ FOCs are:

( ) ( ) ( ) 011 '1

''2 =−−+−+− EsqCPspqsps iiiiii

( ) 01 '1

'2 =−+−−− EqPqpqspq iiiii

FDI invested firms’ FOCs are:

( ) ( ) ( ) 011 '2

'1

''2 =−−−+−+− EEsqCPspqsps iiiiii

( ) 01 '1

'2 =−+−−− EqPqpqspq iiiii

By symmetry, all domestic firms choose the same output quantity and export intensity, and

all FDI invested firms choose the same output quantity and export intensity. But FDI

invested firms’ choice is different from domestic firms’ choice . Let domestic firms’ choice

be and FDI invested firms’ choice be ( dd sq , ) ( )ff sq , . The FOCs become:

( ) ( ) ( ) 011 ',1

''2 =−−+−+− ddddddd EsqCPspqsps (1)

( ) 01 ',1

'2 =−+−−− dddddd EqPqpqspq (2)

( ) ( ) ( ) 011 ',2

',1

''2 =−−−+−+− ffffffff EEsqCPspqsps (3)

( ) 01 ',1

'2 =−+−−− ffffff EqPqpqspq (4)

where is the first derivative of the export cost function with respect to its first

argument, evaluated at domestic firms’ output quantity and export intensity, and and

are those evaluated at FDI invested firms’ choice.

',1 dE

',1 fE

',2 fE

Multiply Equation (2) by d

d

qs−1

and add to equation (1), we obtain:

( ) 0',1

' =−− dd EqCP (5)

6

Multiply Equation (4) by f

f

qs−1

and add to equation (3), we obtain:

( ) 0',2

',1

' =−−− fff EEqCP (6)

The first observation about Equation (5) and (6) is that ( ) ( )ffdd sqsq ,, ≠ , i.e. domestic

firms and FDI invested firms have different equilibrium choice of output quantity and

export intensity, which happens due to the asymmetric impact of foreign presence on firms’

export cost. Furthermore, if FDI invested firms have same production capacity as domestic

firms, i.e. , then FDI invested firms will always export more than their counterparts.

This point can be showed by plugging

df qq =

fd qq = into Equation (5) and (6):

0',2

',1

',1 =−− ffd EEE

which implies as . Since , ',1

',1 fd EE < 0'

,2 <fE 0''11 >E fd ss < .

Total differentiate Equation (5) and (6), holding N, , and constant, we can obtain: dq fq

0'',11

'',12 >×−=

d

f

d

dd

qNq

EE

ddsγ

NEEEE

dds

ff

fff ×+

+−= ''

,12''

,11

'',22

'',12

γ

which shows that for an increase in foreign presence γ , domestic firms will unambiguously

increase their export intensity, while in contrast FDI invested firms’ decision is

undetermined and depends on how their activities affect the export cost function.

This is a stylized model, as it makes simplification in several aspects. For example, all

firms in the model are homogenous, and the FDI invested firms here can act without any

constraint from their parent firms, which seems a little bit unrealistic in reality. Secondly

the FDI’s entry decision is not modeled explicitly, and in the total differentiation above the

total number of firms in the industry, N, is held constant, which implicitly assume that FDI

can only enter the industry by acquiring domestic firms.

7

2. Econometric Specification and Estimation Strategy

Econometric Specification

The theoretical model predicts that if the presence of FDI reduces other firms’ export costs,

for example by the knowledge spillovers on the international market through human

resources movement, then the presence of FDI does promote other firms’ exports. Hence, to

test whether there exists export spillover, the following econometric model is set up:

εβββββ +×++++= XfpfpIXs 43210 (7)

where s denotes firms’ export intensity, namely firms’ export value divided by firms’ total

sale, and ranges between 0 and 1; X is a set of firms’ characteristics, which includes firms’

size, capital intensity, production costs, average labour wage, marketing expense,

ownership structure, regional dummies, and a dummy that indicates firms’ past exporting

experience; I is the industry characteristics that can affect firms’ exports, including the

industry concentration and industry hetereogeneity; and fp denotes the foreign presence that

proxies for FDI’s activities in host country.

X, and I serve as control variables, and fp is the variable of interest. Moreover fp is

interacted with firms’ characteristics X, which allows for empirical search for determinants

of export spillovers of FDI. This is reasonable, as we expect that firms with certain

characteristics, such as bigger firms, are generally more capable to absorb spillovers from

FDI invested firms. The partial impact of fp on firms’ mean export intensity is:

[ ] Xfp

sE43 ββ +=

∂∂

Hence in such kind of setup, testing for the existence of exports spillovers from FDI

follows three steps: first test the joint significance of the coefficient of fp and coefficient of

the interaction terms. If neither of them is significant, then the exports spillover doe not

exist; secondly test the joint significance of the interaction terms. If the interaction terms

are jointly significant, then the happening of export spillovers depends on firms’

characteristic; thirdly, compute the partial effect of fp and evaluate at relevant value of

firms’ characteristics. If the evaluated partial impact of fp is positive, then there exists

8

positive exports spillover from FDI invested firms. If the evaluation is negative, then FDI’s

presence in domestic industry actually does harm to domestic firms’ exports. In addition,

the impact of certain firm characteristic on its capacity to absorb spillovers from FDI

invested firms can be measured from:

[ ]i

ixfpsE

,4

2

β=∂∂

∂

where is the ith firm characteristic in the firm characteristic set X. Obviously, a

significant and positive estimate of the coefficient implies that the firm characteristic helps

the firm to absorb spillovers from FDI invested firms. Otherwise if the coefficient is

significantly negative, then the firm characteristic keeps firms from benefiting spillovers

from FDI invested firms. For example, if the coefficient of fp’s interaction with firm size is

significantly positive, then it indicates that bigger firm is more capable to exploit spillovers

from FDI invested firms.

ix

Besides, fp here is also proxied by FDI invested firms’ output/employment/assets share in

the industry respectively, in order to avoid the inherent pitfall of these three measurements.

Estimation Strategy

One characteristic of the dependent variable s is that it is truncated, and a large amount of s

may take value 0 as it is reasonable to expect a few firms do not export. This makes the

OLS estimation of Equation (7) problematic as the OLS estimation will be biased and

inconsistent, and the prediction of the export intensity s will be easily lower than 0 or

higher than 1.

One widely used approach to handle such situation is the censored Tobit model. The

advantage of the Tobit model is that it does make used of all available information from the

explanatory variables, and in particular it accounts for the fact that a large number of

dependent variables (export intensity) are zero. However, the Tobit model incorporates two

decisions into one model, namely the decision to participate and the decision of level of

participation, which implicitly assumes that the explanatory variables will have same

impact on both the decisions. In the case of exports here, the Tobit model imposes a

9

constraint that the impact on firms’ decision on whether to export by the explanatory

variables is equal to the impact on firms’ decision on how much to export. This may not

necessarily be true, in which case the Tobit model is mis-specified and such kind of

misspecification may have undesirable impact on its estimation (Basile, 2001). This

constraint can be relaxed by separating the problem into two steps: first look at firms’

decision on whether to export, and then explore firms’ decision on how much to export,

conditioned on the fact that firms already decide to export. The validity of such relaxation

of the constraint can be tested (see Lin and Schmidt, 1984, for detail).

Cragg (1971) proposed a two-stage specification that relaxes the constraint of Tobit model.

In Cragg’s specification, the first stage is to estimate a Probit model in which the dependent

variable is firms’ decision on whether to export, namely a binary variable that takes value 1

when firms export. At the second stage, the estimation was made only over the firms that

do export, in which the dependent variable is the export intensity. As the export intensity

takes value only between 0 and 1, the dependent variable is doubly truncated, and hence the

truncated estimation procedure is applied in the estimation. However, Cragg’s

specification implicitly assumes that the two stages are independent of each other. If this is

not true, then the truncated estimation at the second stage will be biased, not only for the

true whole population but also for the group of population that has been selected (Basile,

2001). The possibility of correlation between two stages can be accounted for by the

sample-selection model, in which the first stage is carried out using Probit model, while in

the second stage the sample selection bias is corrected, using the inverse mills ratio

computed from the Probit model in the first stage.

3 The Data and Variable Construction

Variable construction

The data set is a cross-sectional firm level micro-data that covers 181,188 firms in

manufacturing industry in China in 2003. From the data set, two categories of variables are

constructed, namely the firm characteristics and the industry characteristics. In addition, a

10

dummy variable, firms’ past exports experience, is also constructed for the purpose of

identifying the sample selection model in the estimation.

Firm characteristics include firm size, which is proxied by firms’ output, firm age, capital

intensity, which is equal to total assets divided by firms’ number of employees, production

cost, which is equal to firms’ total production and operation cost divided by its output, the

average wage, which is equal to firms’ total wage divided by its number of employees and

can proxy for the composition of labour structure in the firm, namely the human capital, the

unit labour cost , which is equal to total wage divided by firms’ output, the marketing

expense, which is equal to firms’ total marketing expenses divided by their total sales

revenue, the ownership, and the regional hetereogeneity.

The ownership is a dummy variable, which is constructed from firms’ registration type, and

takes value 1 if the firm is private firm or foreign invested firm. The regional

hetereogeneity consists of two regional dummies, the variable western which takes value 1

if the firm is located in West China, and the variable middle which takes value 1 if the firm

is located in Middle China. West China includes Chongqing City, Sichuan Province,

Guizhou Province, Yunnan Province, Tibet Autonomous Zone, Shannxi Province, Gansu

Province, Qinghai Province, Ningxia Autonomous Zone, Inner Mongolia Autonomous

Zone, Xinjiang Autonomous Zone, and Guangxi Autonomous Zone. Middle China includes

Shanxi Province, Henan Province, Anhui Province, Jiangxi Province, Hunan Province,

Hubei Province, Jilin Province, and Heilongjiang Province. The rest of China is classified

as coastal region, including Beijing City, Tianjin City, Shanghai City, Liaoning Province,

Hebei Province, Shandong Province, Jiangsu Province, Zhejiang Province, Fujian Province,

Guangdong Province, and Hainan Province.

For firm size, a positive sign is expected, as the international trade allows for the

exploitation of economy of scale since international trade can be viewed as a way of

extending the market (Krugman, 1979), and firm size is a good indicator of such economy

of scale (Hirsch and Adlar, 1974, Glejser et. al. 1980, Lall and Kumar, 1981). Empirically

people find an inverted U shaped relationship between firm size and export (Willmore,

11

1992, Kumar and Siddharthan, 1994, Wagner, 1995, Wakelin, 1998, Bernard and Jensen,

1999, Sterlacchini, 1999, and Roper and Love, 2002), which reflects the possibility that

firm size may be important for firms to overcome the initial cost barriers in establishing

their export channels, however with the establishment of export channels, the initial

investment is regarded as sunk cost and firm size may be less important in determining

firms’ export intensity.

For firm age, capital intensity, production cost, average, and unit labour cost, a positive

sign is also expected, as previous empirical studies confirms the positive relationship

(Wakelin, 1998, Roper and Love, 2002, Athukorala et. al., 1995, Alvarez, 2006, Roberts

and Tybout, 1997, Basile, 2001). The sign for the coefficient of marketing expense is

expected to be negative, as this signals firms’ sale strategy. If a firm has a large marketing

expense, then the firm’s sales strategy focuses more on domestic market, and hence its

exports will be lower. For regional hetereogeneity, it is expected that firms located in

Coastal region are more likely to export than firms located in Middle China and West

China, which is due to both the transportation cost and different level of economic

development in these regions.

Industry characteristics include the Herfindahle index, which is calculated at 4 digit

industry level and equal to the sum of squared firms’ share in the industry total sale, the

exports concentration ratio, which is computed using the formula of Herfindahle index, the

foreign presence, which proxy FDI’s activities in the industry, and the industry

hetereogeneity. The industry hetereogeneity is computed at 2 digit level (see the following

table), as is reasonable to believe that firms within the same 2 digit industry have same

exports behaviour. The sign of coefficient of Herfindahle index is expected to be negative,

as the Herfindahle index proxies for domestic market concentration and with higher

Herfindahle index firms have power over domestic market and are hence less willing to

export. For the export concentration ratio, it is expected to positively affect firms’ exports.

For the foreign presence, which is the focus of this study, it is also expected positively

affect firms’ exports, which is demonstrated in the above theoretical model.

12

Table 1 Coverage of Manufacturing Industries Code Industry Name

13 Food processing industry 14 Food manufacturing industry 15 Beverage manufacturing industry 16 Tobacco industry 17 Textile industry 18 Textile garments, shoes and hats manufacturing 19 Leather, fur, feathers (cashmere) and the apparel industry 20 Timber processing and wood, bamboo, vines and brown grass-products industry 21 Furniture manufacturing 22 Paper and paper products industry 23 Printing and reproduction of recorded media 24 Cultural, educational, sporting products manufacturing 25 Oil processing, coking and nuclear fuel processing industry 26 Chemical materials and chemical products industry 27 Pharmaceutical industry 28 Chemical fiber industry 29 Rubber products industry 30 Plastic products industry 31 Non-metallic mineral products industry 32 Smelting and rolling of ferrous metals industry 33 Smelting and rolling of non-ferrous metals industry 34 Fabricated metal products 35 General equipment manufacturing industry 36 Specialized equipment manufacturing 37 Transportation equipment manufacturing 39 Electrical machinery and equipment manufacturing industry

40 Communication equipment, computer and other electronic equipment manufacturing

41 Instrumentation and cultural, office machinery manufacturing 42 Crafts and other products industries 43 Waste resources and recycling materials processing industry

Note: 29 industry hetereogeneity dummies are constructed from this classification

Descriptive statistics

Table 2 presents the descriptive statistics of variable constructed from the data set. For

firms’ exports intensity, 2,558 firms have missing value, which happens because they have

zero sale. In the rest of firms, the average exports intensity is 0.18, and 71.7% of firms do

not export, as expected. For the production cost, 2673 firms have missing value, which is

again due to these firms have zero output. As the production cost is equal to firms’

production and operation expenditure divided by output, a value of production cost that is

bigger than 1 indicates the firm is suffering loss. In total 181,188 firms, 18.9% of firms

suffer loss, which significantly increases the mean and standard deviation of the

13

distribution of production cost. For the capital intensity and average wage, 950 firms report

zero employee, which makes the corresponding variables missing values. For the marketing

expense, it can be seen that most firms much focus on domestic market, with an average

value of 0.89. For the domestic market structure, the average Herfindahle index and its

standard deviation are rather low (0.0002 and 0.01 respectively), indicating most markets

are quite competitive. Only two industries are monopolized, namely the nuclear radiation

processing and the agricultural, forestry, and fishery specialized equipment and instrument

manufacturing. In the previous industry, the monopoly firm does not export, while in

contrast in the latter industry the monopoly firm exports 72.3% of its output. For exports

concentration ratio, similar with the Herfindahle index, its mean value and standard

deviation are both very low, indicating a competitive exporting market. However, there are

5 industries that do not have exports, namely the artificial oil production, the tire retread

processing, the asbestos and cement products manufacturing, the nuclear and nuclear

radiation measurement equipment manufacturing, and the nuclear radiation processing

industry, which makes 153 firms have missing value. For the foreign presence, which will

be elaborated in detail in the following section, the three measurements appear to give

consistent proxy on true FDI invested firms’ activities in domestic industry, as their means

are well within one standard deviation of each other.

The bottom panel of Table 2 is a set of dummy variables that are used in estimation, and the

percentage is the proportion of firms in the sample that take value 1 for the variable. For

example, 57.3% of all firms are privately owned, and 22.1% of firms have exporting

experience in the previous year. For the regional dummies, they indicate that 71.6% of

firms are located in Coastal region, which obviously shadows the gap of regional

development.

Table 2 Descriptive Statistics Variables Obs Mean Std. Dev. Min Max Exports intensity 178630 0.18 0.35 0 1Firm size 181188 70289.45 540901.30 0 5.5E+07Firm age 181188 10.71 12.11 0 403Capital intensity 180238 266.43 3263.43 0 874735Production cost 178515 4.66 1249.10 0 527000Average wage 180238 10.95 11.93 0 2832.09

14

15

Unit labour cost 178515 0.26 40.97 0 17138.5Marketing expense 178533 0.89 0.56 0 174.25Herfindahle index 181188 0.00 0.01 0 1Exports concentration ratio 181035 0.00 0.01 0 1fpo 181188 0.34 0.18 0 1fpe 181188 0.28 0.19 0 1fpa 181188 0.36 0.18 0 1

Dummy variables % Dummy variables %

Dummy variables %

Ownership 57.3 d21 1.13 d32 2.27Western 10.9 d22 3.07 d33 1.86Middle 17.5 d23 2.25 d34 5.38Past exporting experience 22.1 d24 1.39 d35 6.92d14 2.56 d25 0.73 d36 3.93d15 1.76 d26 7.62 d37 4.57d16 0.14 d27 2.24 d39 5.74d17 8.20 d28 0.52 d40 3.23d18 5.36 d29 1.11 d41 1.39d19 2.49 d30 4.63 d42 2.35d20 1.93 d31 8.94 d43 0.06Note: (1) fpo/fpe/fpa is the foreign presence measured in terms of

output/employee/assets share; (2) Variables starting with letter d are the industry dummies, for example d14 denotes whether the firm is in Industry 14; (3) Percentage for dummy variables denotes the proportion of firms that take value 1, for example for dummy Western, 10.9% of firms are located in West China.

Table 3 presents the correlation matrix for key variables. From the table, we can find that

for most of variables the correlation is rather low, except for the production cost and unit

labour cost. The correlation between the production cost and unit labour cost is as high as

0.992, which may present multicollinearity problem. Besides, for the three proxies of FDI,

there are highly correlated with each other, which is expected and desired.

Table 3 Correlation Matrix for Key Variables

Firm

size

Firm

age

Capital

intensity

Production

cost

Average

wage

Unit

labour

cost

Marketing

expenditureHerfindahle

exports

concentrationfpo fpe fpa ownership western middle

past

exporting

experience

Firm size 1

Firm age 0.049 1

Capital intensity 0.120 -

0.007 1

1

Production cost 0.000 0.008 0.000 1

Average wage 0.107 -

0.021 0.229 -0.002 1

Unit labour cost -

0.001 0.008 -0.001 0.992 -0.001 1

Marketing

expenditure

-

0.004

-

0.010 -0.003 -0.003 -0.016 -0.003 1

Herfindahle 0.108 0.022 0.021 0.000 0.026 0.000 -0.003 1

exports

concentration 0.065 0.012 0.018 0.000 0.026 0.000 -0.001 0.168 1

fpo 0.030 -

0.126 0.017 -0.002 0.085 -0.002 0.007 0.002 -0.014 1

fpe 0.013 -

0.143 -0.012 -0.002 0.064 -0.002 0.011 0.002 -0.015 0.921 1

fpa 0.013 -

0.148 -0.001 -0.003 0.063 -0.003 0.010 -0.003 -0.018 0.954 0.920 1

ownership -

0.022

-

0.369 0.004 -0.004 0.066 -0.004 0.007 -0.016 0.003 0.202 0.230 0.223

western -

0.006 0.099 -0.006 0.008 -0.064 0.009 -0.008 0.000 0.002

-

0.159

-

0.174

-

0.170 -0.152 1

middle -

0.007 0.085 -0.033 -0.001 -0.123 -0.002 -0.005 -0.002 -0.005

-

0.176

-

0.191

-

0.182 -0.149 -0.159 1

past exporting

experience 0.084 0.031 0.012 -0.002 0.098 -0.002 0.004 0.021 0.052 0.240 0.292 0.243 0.148 -0.118 -0.155 1

Source: Stata calculation

16

Exports intensity in manufacturing industry

Table 2 gives the summary statistics for exports intensity over the whole sample set, and

it indicates that on average firms’ exports intensity is quite low (0.18). In this section, the

distribution of firms’ exports intensity will be further examined. Table 4 presents the

summary statistics of firms’ exports intensity by 2 digit industry. From Table 4, several

observations on the exports intensity can be concluded, as follows:

Table 4 Summary Statistics for Exports Intensity by Industries

Industry Code

Number of firms Mean

Std. Dev. Min Max

% of firms with 0 exports intensity

% of firms with 1 exports intensity

% of firms with missing value

13 11193 0.11 0.29 0 1 79.66 4.09 2.90 14 4636 0.12 0.29 0 1 72.35 3.88 3.21 15 3194 0.05 0.19 0 1 85.82 1 2.44 16 255 0.01 0.04 0 0.53 76.86 0 3.92 17 14863 0.24 0.37 0 1 60.61 7.93 1.31 18 9717 0.52 0.46 0 1 37.31 28.37 0.55 19 4518 0.50 0.46 0 1 39.35 30.01 0.73 20 3501 0.18 0.36 0 1 74.06 9.08 1.43 21 2046 0.30 0.43 0 1 60.70 16.32 1.12 22 5570 0.05 0.19 0 1 87.97 1.80 1.65 23 4084 0.04 0.19 0 1 90.40 1.91 1.71 24 2516 0.60 0.44 0 1 27.31 36.37 0.64 25 1323 0.01 0.09 0 1 91.38 0.30 2.12 26 13803 0.08 0.23 0 1 78.90 2.09 1.48 27 4063 0.08 0.22 0 1 78.12 1.33 1.80 28 937 0.06 0.20 0 1 83.14 1.81 0.53 29 2016 0.15 0.31 0 1 71.28 5.75 1.29 30 8382 0.19 0.36 0 1 71.14 9.60 0.70 31 16245 0.07 0.23 0 1 86.01 3.30 1.52 32 4119 0.03 0.15 0 1 89.39 1.02 1.97 33 3367 0.07 0.20 0 1 81.68 1.49 1.34 34 9746 0.20 0.37 0 1 69.24 7.94 1.09 35 12546 0.12 0.27 0 1 74.45 3.04 0.97 36 7129 0.07 0.20 0 1 78.73 1.54 1.66 37 8281 0.09 0.24 0 1 78.95 2.46 1.29 39 10400 0.18 0.35 0 1 68.89 7.58 1 40 5857 0.32 0.41 0 1 49.02 13.95 1.33 41 2515 0.28 0.41 0 1 56.78 12.49 1.31 42 4259 0.56 0.46 0 1 33.79 36.82 0.66 43 107 0.02 0.11 0 1 95.33 0.93 0.93

Note: (1) Table 1 presents the name of corresponding Industry Code. (2) The mean and standard deviation are computed excluding missing values.

17

First, there are four industries with average exports intensity higher than 50%, namely the

cultural, educational, sporting products manufacturing (24), the crafts and other products

industries (42), the textile garments, shoes and hats manufacturing (18), and the leather,

fur, feathers (cashmere) and the apparel industry, on all of which China has comparative

advantages. Secondly, except for the tobacco industry (16), all the other industries have

firms that specialize in exports, i.e. with 100% exports intensity. The highest percentage

of firms that specialize in exports is 36.82% in the crafts and other products industries

(42), and in contrast the lowest percentage is 0 in the tobacco industry (16). Thirdly, there

are five industries with average exports intensity lower than 5%, namely the printing and

reproduction of recorded media industry (23), the smelting and rolling of ferrous metals

industry (32), the waste resources and recycling materials processing industry (43), the

oil processing, coking and nuclear fuel processing industry (25), and the tobacco industry

(16). Fourthly, the distribution of exports intensity across industries displays a rather

similar pattern, i.e. a large proportion of firms have zero exports intensity.

Foreign presence in manufacturing industry

Conventionally, there are three proxies for measuring FDI’s activities in domestic

industries, namely the share of FDI invested firms’ output/employees/assets in the

domestic industry. Either of these three measurements has its own pitfalls. For example,

for the assets share, it is argued that FDI tends to flow into capital intensive industry and

FDI invested firms are usually more capital intensive, and thus it can over-estimate the

foreign presence. To avoid such kind of pitfall, all three measurements are used in this

study. In addition, this section looks into the consistency of these three measurements.

Ideally, we would expect all these three proxies rank the industries consistently. For

example, if one industry is ranked no. three in terms of output share, then it will be also

ranked no. three in terms of both employee share and assets share. If such kind of

consistency is satisfied, then it is expected that they shall be able to measure the true

foreign presence accurately. From Table 2, it can be seen that the three proxies give

similar measurement of foreign presence in that their mean and standard deviation are

quite close to each other.

18

Table 5 presents the ranking of foreign presence of the three proxies. The ranking of fpo

is consistent with that of fpa. However, unfortunately the ranking of fpe is significantly

different. As it is reasonable to believe that FDI invested firms are usually more capital

intensive and hence fpe can underestimate the true foreign presence, the final empirical

results obtained using fpe as proxy of FDI’s activities shall be interpreted with caution.

Table 5 Ranking Consistency of Three FDI Proxies

Industry Code fpo rank (fpo) fpe Rank (fpe) fpa Rank (fpa) 40 0.75 1 0.66 1 0.66 1 24 0.59 2 0.61 2 0.65 2 19 0.54 3 0.57 3 0.58 3 21 0.50 4 0.51 5 0.53 4 18 0.52 5 0.52 4 0.53 5 30 0.45 6 0.41 7 0.50 6 42 0.42 7 0.46 6 0.46 7 41 0.50 8 0.39 8 0.42 8 29 0.37 9 0.32 10 0.42 9 22 0.35 10 0.23 18 0.41 10 14 0.40 11 0.29 11 0.40 11 39 0.40 12 0.36 9 0.40 12 15 0.39 13 0.27 14 0.38 13 28 0.33 14 0.25 16 0.38 14 34 0.33 15 0.27 12 0.36 15 20 0.28 16 0.23 17 0.36 16 23 0.35 17 0.27 13 0.35 17 17 0.32 18 0.27 15 0.35 18 37 0.33 19 0.19 21 0.32 19 43 0.20 20 0.08 29 0.31 20 13 0.26 21 0.20 19 0.30 21 35 0.26 22 0.18 22 0.29 22 26 0.28 23 0.16 24 0.28 23 27 0.27 24 0.20 20 0.26 24 36 0.24 25 0.16 23 0.23 25 31 0.18 26 0.11 26 0.22 26 33 0.18 27 0.12 25 0.20 27 25 0.13 28 0.10 27 0.17 28 32 0.17 29 0.09 28 0.15 29 16 0.02 30 0.04 30 0.03 30

19

4 Empirical Results

The sample selection model is estimated using Heckman two-step method, and the

estimation is made following a general-to-specific approach, namely first include all

variables in the model and then sequentially drop insignificant variables in which the

Wald tests are conducted to examine whether the dropping is appropriate. The process is

repeated until the most parsimonious specification is reached. In every step, the estimated

correlation between the export participation (the Probit model) and the export intensity

(the truncated regression model) is significantly different from zero, and in the final

estimation the estimated correlation is -1. This indicates that firms’ decision on whether

to export is not independent of their decision on how much to export. Hence Craigg’s

specification of the Tobit model, which assumes the independence between firms’ export

participation decision and export intensity decision, is inappropriate here. In addition, the

existence of sample selection bias, i.e. the significance of inverse mills ratio, is tested in

every step and it is found that the sample selection does exists. The sequential search for

parsimonious model is made using FDI invested firms’ output share in the industry as

proxy for foreign presence. When the most parsimonious specification is reached,

estimations using the other two proxies are made according to the parsimonious

specification, in which further dropping of variables are made in the estimation using the

employment share as a proxy for foreign presence. Furthermore, the Tobit restriction, i.e.

explanatory variables have equal effect on both firms’ export participation and export

intensity decisions, is tested on the final parsimonious specification. Table 6, 7, and 8

report the final estimation of the sample selection model, using output share, employment

share, and assets share as proxy for foreign presence respectively. The test statistics for

Tobit restriction are 6076.92, 5921.35, and 6022.64, which rejects the null that Tobit

restriction is appropriate at 1% significance level. In summary, diagnostic tests suggest

that firms with sale abroad can not represent the sample of exporting firms, and

explanatory variables have different effects on firms’ decision on whether to export and

how much to export.

Existence of exports spillover from FDI and its determinants

20

As stated previously, testing for the existence of exports spillover from FDI to domestic

firms is made in three steps: first test the joint significance of foreign presence and its

interaction terms, secondly test the joint significance of the interaction terms, which is

also a test for the determinants of exports spillover, thirdly compute the marginal effect

of the foreign presence, which is a function of the determinants and indicates that

different firms can benefit differently from FDI. This marginal effect is then evaluated at

sample average, which tells on average how FDI affect firms’ exports.

For the estimation using output share as proxy for foreign presence, the test statistics for

joint significance of foreign presence and its interaction terms are 3916.79 with p-value

of 0 in the export intensity equation and 4571.32 with p-value of 0 in the export

participation equation, which is a necessary condition for the existence of spillovers.

Secondly, the test statistics for joint significance of interaction terms are 2043.99 with p-

value of 0 in the export intensity equation and 4504.62 with p-value of 0 in the export

participation equation, which confirms that whether the export spillover will happen

depends on firms’ characteristics. Thirdly, the marginal effect of foreign presence in the

export intensity equation is:

middlewesternownership

eaveragewagsizeagefpos

×−×−×+

×−×+×+=∂∂

35.031.031.0

003.001.00024.02.0

where age denotes firms’ age, size denotes firms’ size, averagewage denotes firms’

average wage, ownership is a ownership dummy, western and middle are regional

dummies. This marginal effect formula indicates that firms’ age, firms’ size, average

wage, ownership structure and geographical location determine the happening of export

spillovers from FDI. Firms’ age and size have positive impact on export spillovers. In

contrast, firms’ average wage, a proxy for firms’ employment structure, has negative

impact on export spillovers, which may due to the fact that China has comparative

advantages in labour intensive goods. Private firms can benefit more from foreign

presence in the industry than public owned firms, and firms located in Coastal China are

more capable in absorbing such export spillovers. If the marginal effect formula is

evaluated at sample average, we obtain Table 9. Table 9 shows that for an average firm,

21

the foreign presence need not necessarily be beneficiary to its export. If the average firm

is privately owned and located in West China, then 1% increase in foreign presence will

increase its export intensity by 0.19%. If the privately owned average firm is located in

Middle China, the increase in export intensity is 0.15%. For public owned average firms,

foreign presence does harm to their exports. 1% increase in foreign presence will lower

the firm’s export intensity by 0.12% if it is located in West China, and by 0.16% if it is

located in Middle China. Firms located in Coastal China appear to benefit most from

foreign presence. For an average firm located in Coastal China, 1% increase in foreign

presence will increase 0.5% of its export intensity if it is privately owned, and 0.19% if it

is publicly owned.

Table 9 Marginal Effect of Foreign Presence on Export Intensity West China Middle China Coastal China fpo fpe fpa fpo fpe fpa fpo fpe fpa Private 0.19 0.3 0.22 0.15 0.19 0.17 0.5 0.59 0.52 Public -0.12 -0.12 -0.09 -0.16 0.01 -0.15 0.19 0.41 0.21 Note: The marginal effect is evaluated at sample average.

In regard to the export participation equation, the marginal effect of foreign presence on

the probability of firms’ participating in exports is:

( ) ( )( westerneaveragewagsizeageXfpo

E×+×−×+×−−=

∂∂ 47.005.034.14903.027.0Pr βφ )

)where denotes the probability that the firm participates in exports, ( )EPr ( βφ X is the

pdf of normal distribution evaluated at βX , and age, size, averagewage, and western are

as previous. The formula shows that the impact of foreign presence on firms’ probability

of participating in exports depends on firms’ age, size, average wage and geographical

location, conditioned on a same probability density. Evaluated at the sample average of

all explanatory variables in the export participation equation, the marginal effect is

φ36.9 for an average firm located in Middle or Coastal China, and φ82.9 for an average

firm located in West China, where φ denotes the normal probability density evaluated at

sample average. Both evaluated marginal effects are positive, indicating foreign presence

increase firm’s probability in exporting, and furthermore such kind of impact is bigger in

West China than in Middle and Coastal China. Besides, compared with the export

22

participation equation and export intensity equation, foreign presence’s interaction terms

with ownership and regional dummy middle are insignificant.

For the estimation using assets share as proxy for foreign presence, the Wald tests for

joint significance of foreign presence and its interaction terms both reject the null of

insignificance at 1% level. The marginal effect of foreign presence in the export intensity

equation is:

middlewesternownership

eaveragewagsizeagefpas

×−×−×+

×−×+×+=∂∂

36.03.031.0

003.001.00025.022.0

which is not significantly different from the marginal effect of foreign presence in the

estimation using output share as proxy. Table 9 also reports the marginal effect evaluated

at sample average, and again there is no significant difference. As to the export

participation equation, the marginal effect of foreign presence is:

( ) ( )( westerneaveragewagsizeageXfpa

E×+×−×+×−−=

∂∂ 46.005.014.14803.019.0Pr βφ )

which also displays no significant difference from the estimation using output share as

proxy.

For the estimation using employement share as proxy for foreign presence, the Wald tests

for joint significance of foreign presence and its interaction terms both reject the null of

insignificance at 1% level. The marginal effects of foreign presence in the export

intensity equation and export participation equation are:

middlewesternownership

eaveragewagsizeagefpes

×−×−×+

×−×+×+=∂∂

4.029.018.0

003.002.00015.046.0

( ) ( )( westerneaveragewagsizeageXfpe

E×+×−×+×−−=

∂∂ 46.005.077.14703.02.0Pr βφ )

in which the marginal effect in the export participation equation does not appear to be

significantly different from the other two estimations, while the marginal effect in the

export intensity equation displays some difference. However, if the marginal effect is

evaluated at sample average (see Table 9), it gives same sign with prediction of the other

23

two proxies, except for publicly owned firms located in Middle China. Even for publicly

owned firms located in Middle China, the scale of marginal effect is pretty small.

In summary, the following conclusion can be made in regard to the export spillover of

FDI in China: the export spillover of FDI in China does exit, but the scale of the spillover

depends on such firm characteristics as firms’ age, size, average wage, ownership

structure, and geographical location. Furthermore, the existence of export spillovers from

FDI and its determinants are robust to different proxies of FDI.

Impact of other factors on firms’ export behaviour

In the process of sequential search for parsimonious specification using output share as

proxy for foreign presence, altogether 53 variables, including foreign presence’s

interaction terms, are incorporated into regressions, and in the final model there are 39

variables in the export intensity equation and 14 variables in the export participation

model. The insignificant variables in the export intensity equation are firms’ size, age,

production cost, unit labour cost, marketing expense, regional dummy western, and

industry dummies d29, d39 and d40. The insignificant variables in the export

participation equation are firms’ size, age, production cost, unit labour cost, marketing

expense, regional dummies western and middle, and industry dummies d15, d16, d19,

d20, d21, d22, d23, d24, d25, d26, d27, d28, d29, d30, d31, d33, d34, d35, d36, d37, d39,

d40, d41 d42, and d43.

For firms’ size, age, and regional dummy western, even though they are insignificant

themselves, their interaction terms with foreign presence is still significant, which

indicates that they do have impact on firms’ decisions on whether to export and how

much to export. For the regional dummy middle, it is significant in the export intensity

equation, but not in the export participation equation. For firms’ production cost, unit

labour cost and marketing expense, both they and their interaction terms with foreign

presence are insignificant, which contradicts with the prior expectation. The

insignificance of firms’ production cost and unit labour cost may be due to the fact that

China has been enjoying the cost advantage and hence conditioned on this a lower cost

24

will not give firms further incentive to export. The insignificance of firms’ marketing

expense may owe to the fact that most of firms in China are domestic firms and do not

have a global sale strategy.

In addition, compared with the final specification of both equations, we can find that the

export participation equation displays much less industry hetereogeneity than the export

intensity equation, in the sense that only five industry dummies are significant in the

participation equation while all industry dummies are significant in the intensity equation.

This is reasonable, as different industries have different products, which mean their

export procedure is different, and this must be explicitly considered when firms are

deciding how much goods to export, while this is less important in their decision on

whether to export.

For all the other significant variables, the capital intensity, which is equal to total assets

divided by the number of employees, has positive impact on the export intensity, namely

the increase of capital intensity by 1 will spur the export intensity by 0.000005. In

contrast, it has a significant and negative impact on firm’s decision on whether to export.

Furthermore, this is robust to different proxies of foreign presence. China is a country

with rich labour endowment, which makes firms with lower capital intensity, i.e. more

labor intensive, easier to export. However in regard to how much to export, firms with

higher capital intensity usually have bigger production capacity, and hence tend to have

higher export intensity.

The average wage is a proxy for human capital. Higher average wage indicates higher

human capital. The coefficient of average wage is all significant and positive in both the

export participation equation and export intensity equation and in all three estimations,

which shows that firms with higher human capital not only are more likely to participate

in export but also tend to export more. Furthermore, the magnitude of the coefficients is

the same across three estimations, except in the export intensity equation in the

estimation using employment share as proxy, where the coefficient (0.0017) is

significantly bigger than the other two (0.0009).

25

For the Herfindahle Index, it has negative impact on firms’ export intensity, which is

consistent with the prior expectation, as firms with domestic market power are more

willing to exploit such market power than export. The sign of the coefficient are the same

across estimations using three different proxies, and the magnitude is very similar, which

states that 1% increase in Herfindahle index will decrease firms’ export intensity by

1.07%. For exports concentration ratio, which is computed according to the formula of

Herfindahle index, it has significant and positive impact on export intensity in all three

estimations and the magnitude of the coefficient is similar. This is reasonable, as in a

industry with higher export concentration ratio those dominant export firms are easier

targets of learning for other domestic firms. Such learning process helps firms to export.

Firms’ ownership structure plays a role in both the export intensity equation and

participation equation. In the export participation equation, the coefficient is 0.36, 0.35,

and 0.36 in estimations with output share, employment share, and assets share as proxies

respectively, which is quite consistent with each other in the sense that one is well within

one standard deviation of the others. However, in the export intensity equation, the sign

of the coefficient is positive in estimations with output share and assets share, in contrast

it is negative in the estimation with employment share. The reverse in the sign of

coefficient indicates the impact of firms’ ownership on firms’ export intensity is sensitive

to proxies of foreign presence. However, as the foreign presence measured in terms of

employment share usually underestimates FDI invested firms’ activities since FDI

invested firms are usually more capital intensive than their counterparts, the coefficients

estimated using output share and assets share respectively, which are consistent with each

other, is more convincing. As ownership’s interaction with foreign presence is also

significant in the export intensity equation, its marginal effect depends on the level of

foreign presence, which is -0.09 when evaluated at the sample average of foreign

presence in estimations using output share and assets share. The evaluated marginal

effect of ownership indicates that a private firm will have 9% lower export intensity than

publicly owned firms on average, but in contrast a private firm is more likely to

participate in export.

26

Tabe 6 Estimation Results Using Output Share as Foreign Presence Exports Intensity Exports Participation Variables Coefficient Std.Err. Coefficient Std.Err.

Capital Intensity 5.01E-06 1.76E-06 -1.05E-044.37E-

06 Average Wage 0.0009 0.0002 0.03 0.003 Herfindahle Index -1.07 0.21 Exports Concentration Ratio 1.19 0.08 fpo 0.20 0.01 -0.27 0.09 fpo*Firm Age 0.0024 0.0003 -0.03 0.002 fpo*Firm Size 0.01 0.003 149.34 2.34 fpo*Average Wage -0.003 0.0005 -0.05 0.01 fpo*Ownership 0.31 0.01 fpo*Western -0.31 0.01 0.47 0.10 fpo*Middle -0.35 0.02 Middle 0.02 0.01 Ownership -0.02 0.005 0.36 0.03 d14 -0.02 0.01 -0.14 0.05 d15 -0.07 0.01 d16 -0.07 0.03 d17 0.08* 0.005 -0.16* 0.04 d18 0.26 0.01 0.19 0.08 d19 0.24 0.01 d20 0.05 0.01 d21 0.08 0.01 d22 -0.09 0.01 d23 -0.09 0.01 0.28 0.07 d24 0.31 0.01 d25 -0.07 0.01 d26 -0.04 0.005 d27 -0.02* 0.01 d28 -0.10 0.01 d30 -0.01* 0.01 d31 -0.01 0.005 d32 -0.05 0.01 -0.22 0.07 d33 -0.03 0.01 d34 0.04 0.01 d35 -0.01* 0.005 d36 -0.05 0.01 d37 -0.05 0.01 d41 0.06 0.01 d42 0.36* 0.01 d43 -0.10 0.04 Past Exports Experience 0.32 0.05 Constant 0.05 0.01 1.78 0.03

27

Observations 180089 inverse mills ratio -0.45 0.02 rho -1 sigma 0.45 Test for Tobit Restriction (chi2) 6076.92 Note: * denotes significance at 5% level, and all the other coefficients are significant at 1% level.

Tabe 7 Estimation Results Using Employment Share as Foreign Presence Exports Intensity Exports Participation Variables Coefficient Std.Err. Coefficient Std.Err. Capital Intensity 6.57E-06 1.74E-06 -1.05E-04 4.37E-06 Average Wage 0.0017 0.0002 0.03 0.003 Herfindahle Index -1.08 0.20 Exports Concentration Ratio 1.21 0.08 fpe 0.46 0.01 -0.20* 0.08 fpe*Firm Age 0.0015 0.0003 -0.03 0.002 fpe*Firm Size 0.02 0.003 147.77 2.33 fpe*Average Wage -0.006 0.0004 -0.05 0.01 fpe*Ownership 0.18 0.01 fpe*Western -0.29 0.01 0.46 0.10 fpe*Middle -0.40 0.02 Middle 0.04 0.01 Ownership 0.02 0.004 0.35 0.03 d14 -0.16 0.05 d15 -0.04 0.01 d17 0.09 0.004 -0.15 0.04 d18 0.23 0.01 0.20* 0.08 d19 0.19 0.01 d20 0.06 0.01 d21 0.04 0.01 d22 -0.06 0.01 d23 -0.07 0.01 0.28 0.07 d24 0.25 0.01 d25 -0.05 0.01 d28 -0.08 0.01 d30 -0.03 0.01 d31 0.02 0.004 d32 -0.21 0.07 d34 0.05 0.00 d35 0.02 0.004 d36 -0.02 0.01 d41 0.08 0.01 d42 0.32 0.01 Past Exports Experience 0.33 0.05

28

Constant -0.01* 0.00 1.76 0.03 Observations 180089 inverse mills ratio -0.44 0.02 rho -1 sigma 0.44 Test for Tobit Restriction (chi2) 5921.35 Note: * denotes significance at 5% level, and all the other coefficients are significant at 1% level.

Tabe 8 Estimation Results Using Assets Share as Foreign Presence Exports Intensity Exports Participation Variables Coefficient Std.Err. Coefficient Std.Err. Capital Intensity 5.16E-06 1.75E-06 -1.05E-04 4.37E-06 Average Wage 0.0009 0.0002 0.03 0.003 Herfindahle Index -1.06 0.20 Exports Concentration Ratio 1.20 0.08 fpa 0.22 0.01 -0.19* 0.08 fpa*Firm Age 0.0025 0.0003 -0.03 0.002 fpa*Firm Size 0.01 0.003 148.14 2.34 fpa*Average Wage -0.003 0.0005 -0.05 0.01 fpa*Ownership 0.31 0.01 fpa*Western -0.30 0.01 0.46 0.10 fpa*Middle -0.36 0.02 Middle 0.02 0.01 Ownership -0.02 0.004 0.36 0.03 d14 -0.02 0.01 -0.15 0.05 d15 -0.07 0.01 d16 -0.07* 0.03 d17 0.08 0.005 -0.15 0.04 d18 0.26 0.01 0.18* 0.08 d19 0.23 0.01 d20 0.04 0.01 d21 0.07 0.01 d22 -0.11 0.01 d23 -0.09 0.01 0.28 0.07 d24 0.29 0.01 d25 -0.07 0.01 d26 -0.04 0.005 d27 -0.02* 0.01 d28 -0.11 0.01 d30 -0.03 0.01 d31 -0.02 0.005 d32 -0.04 0.01 -0.22 0.07 d33 -0.03 0.01 d34 0.04 0.01 d35 -0.01 0.005 d36 -0.05 0.01

29

d37 -0.05 0.01 d41 0.08 0.01 d42 0.35 0.01 d43 -0.12 0.04 Past Exports Experience 0.32 0.05 Constant 0.05 0.005 1.77 0.03 Observations 180089 inverse mills ratio -0.45 0.02 rho -1 sigma 0.45 Test for Tobit Restriction (chi2) 6022.64 Note: * denote significance at 5% level, and all the other coefficients are significant at 1% level.

5 The Conclusion

This paper studies the export spillover of FDI and its determinants, using a firm level

cross-sectional micro-data that consists of over 180 thousands firms in the manufacturing

industry in China in 2003. The paper first shows in a theoretical model that if FDI

invested firms can reduce domestic firms’ export cost, for example by the knowledge

spillover, then domestic firms will benefit from FDI’s presence in the domestic industry

in the sense that their exports will be promoted. Then this hypothesis is tested empirically

using the sample selection model in which firms’ decisions on whether to export and on

how much to export are dependent on each other. The sample selection model is

estimated using Heckman two-step procedure. The estimation finds that FDI in the

manufacturing industry does generate export spillover, and furthermore the scale of the

spillover depends positively on firms’ age, size, and ownership structure, and negatively

on firms’ average wage and geographical location.

30

References:

Alvarez, R. (2006), Explaining export success: firm characteristics and spillover effects,

World Development, doi:10.1016/j.worlddev.2006.11.002

Arthukorala, P., S. Jayasuriya, & E. Oczkowski (1995), Multinational firms and export

performance in developing countries: some analytical issues and new empirical

evidence, Journal of Development Economics, vol.46, pp.109-122

Basile, R. (2001), Export behaviour of Italian manufacturing firms over the nineties: the

role of innovation, Research Policy, vol.30, pp.1185-1201

Bernard, A.B. & Jensen, J.B. (1999), Exceptional export performance: cause, effect, or

both, Journal of International Economics, vol.47, pp.1-25

Cragg, J.G. (1971), Some statistical models for limited dependent variables with

application to the demand for durable goods, Econometrica, vol.39, no.5, pp.829-

844

Dunning, J.H. (1988), Explaining international production (Unwin Hyman, London)

Dunning, J.H., B. Kogut, & M. Blomstrom (1990), Globalization of firms and the

competitiveness of nations, Lund University Press, Lund

Glejser, H., A. Jaquemin & J. Petit (1980), Exports in an imperfect competition

framework, an analysis of 1446 exporters, Quarterly Journal of Economics, vol.94,

no.3, pp.507-524

Hirsch, S. & Z. Adler (1974), Firm size and export performance, World Development,

vol.2, no.2, pp.41-46

Krugman, P. (1979), Increasing returns, monopolistic competition, and international trade,

Journal of International Economics, vol.9, pp.469-480

Kumar, N. & N.S. Siddharthan (1994), Technology, firm size and export behaviour in

developing countries, Journal of Development Studies, vol.32, pp.288-309

Lall, S. & S. Kumar (1981), Firm-level export performance in an inward-looking

economy: the Indian engineering industry, World Development, vol.9, no.5, pp.453-

463

Lin, T. & P. Schmidt (1984), A test of the Tobit specification against an alternative

suggested by Cragg, Review of Economic Statistics, vol.66, pp.174-177

31

Roberts, M.J. & J.R. Tybout (1997), The decision to export in Colombia: an empirical

model of entry with sunk costs, American Economic Review, vol.87, no.4, pp.545-

564

Roper, S. & J. H. Love (2002), Innovation and export performance: evidence from the

UK and German manufacturing plants, Research Policy, vol.31, pp.1087-1102

Sterlacchini, A. (1999), Do innovative activities matter to small firms in non-R&D-

intensive industries? An application to export performance, Research Policy, vol.28,

pp.819-832

Wagner, J. (1995), Exports, firm size, and firm dynamics, Small Business Economics,

vol.7, pp.29-39

Wakelin, Katherine (1998), Innovation and export behaviour at the firm level, Research

Policy, vol.26, pp.829-841

Willmore, L. (1992), Transnationals and foreign trade: evidence from Brazil, Journal of

Development Studies, vol.28, pp.314-335

32