The conventional models

This study uses various forms of the gravity model to analyze the determination of

trade flows. The gravity model of international trade assumes that the values of

bilateral trade flows can be determined as an increasing function of the sizes of

trading economies, a decreasing function of the geographic distance between them,

and other economic and non-economic factors, which may enhance or deter bilateral

trade. A simple form of the gravity model takes the following form:

lnEX_ijkt=α_t+α_k+β₁lnY_it+β₂lnY_jt+β₃lnDIST_ij+β₄LANG_ij

+β₅ADJ_ij+β₆REA_ijt+ε_ijkt , (CS)

where EX_ijkt is the value of exports from country i to country j in year t, and α_t is the

year dummy, DIST_ij is the distance between economic centers of country i and country

j, LANG_ij is a dummy variable assuming the value of one if two countries have a

common official language, and otherwise zero, ADJ_ij is a dummy variable assuming

that the value of one if two countries share a common border and zero otherwise, and

REA_ijt is a dummy variable assuming the value of one if the country join a regional

economic agreement in year t and zero otherwise, ε_ijkt is the residual term. The CS

model is usually estimated by the ordinary least squares (OLS) method.

26

This study uses pooled cross-section (PCS) model as a benchmark to measure the

determinants of bilateral trade flows. The PCS estimation of this study is referred to

Brada & Mendez (1983) in which a cross-section approach is used to study trade

flows. In the following, I use the following four types of gravity model to explain the

determinants of the bilateral trade: (1) controlling for the year (αt) and

industry-specific effects (αk) (Model FEIT), and (2) controlling for the year (αt)and

country-industry pair effects (αjk) (Model FECI), and (3) controlling for the year (αt),

country (αj), and industry specific effects (αk) (Model FE3), and (4) controlling for the

year (αt), country (αj), industry (αk), and country-industry pair effects (αjk)(Model

FEFV).

Controlling for the year and industry specific effects (Model FEIT)

Undoubtedly, pooling cross-section and time-series data increases the number of

observations for the estimation. The pooled cross-section model (PCS) deals with the

problem of additional degree of freedom without significantly increasing the number

of variables, allowing for the intercepts to differ over time. To measure the effects of

trade blocs on bilateral trading pattern, the gravity model could be specified as:

lnEX_ijkt=α_t +α_k +β₁lnY_jt+β₂lnN_jt+β₃lnFDI_OUT_jt+β₄lnFDI_IN_jt

+β₅lnK_jt+β₆lnMANU_jt+β₇lnMANU_LABO_jt+β₈lnDIST_ij

27

+β9lnINDUS_DISTijkt+β10EUjt+β11NAFTAjt+β12AFTAjt

+β13MERCOSURjt+β14APECjt+εijkt , (FEIT)

where EXijkt is the value of industry exports from country i to country j in year t and αt

is the year-specific effect common to Taiwan and other trading countries, αk is the

industry-specific effect, and FDI_OUTjt is the value of outward of foreign direct

investment of country j and FDI_INjt is the value of inward of foreign direct

investment of country j, Kjt is the capitalstock of country j, MANUjt is the value of

manufacturing output of country j, and DISTij is the distances between Taiwan and

trading partners, INDUS_DISTijkt is the trade volume of a specific industry divided by

total volume of trade from Taiwan to its trading partners in year t , MANU_LABO_jt is

the number of manufacturing labor in country j, and the trade bloc dummies,

including EU_jt,,NAFTA_jt, AFTA_jt,, MERCOSUR_jt,, APEC_jt, which are set to considered

to be one if trading partners are acting in that regional bloc in time t and otherwise

zero. And this study further uses EX_lnY_i_lnY_j__ijkt to replaceEX_ijkt, i.e. β₁ is set to be

one to form another regression.

Controlling for the year and country-industry pair effects (Model FECI)

Most of the earlier studies of bilateral trade using cross-sectional models usually yield

biased results due to ignoring the effect of the heterogeneity of country-industry pair

28

effects. This study also estimates the two-way fixed effect model as one of the

estimations. The measure can be specified as:

lnEXijkt=αt+αjk+β1lnYjt+β2lnNjt+β3lnFDI_OUTjt+β4lnFDI_INjt

+β5lnKjt+β6lnMANUjt+β7lnMANU_LABOjt

+β8lnINDUS_DISTijkt+β9EUjt+β10NAFTAjt+β10AFTAjt

+β11MERCOSURjt+β12APECjt+εijkt . (FECI)

In the above equation, (FECI) model is controlling for the year and

country-specific effects, and αjk is the interaction term of country j and industry k. We

also use an alternation to replace EX_lnYi_lnYj_ijkt with EXijkt in equation (FECI), let β1

equals to one and forming the other regressions. And the other variables and dummies

are noted in equation (FEIT).

Controlling for the year, country, and industry specific effects (Model FE3)

Mátyás (1997) firstly introduces the three-way fixed effect of gravity specification

with a dummy of time, and other two dummies of time-invariant, exporting and

importing country effects to study trade flows. Mátyás (1998) further introduces that

the sample with larger cross-section country specific effects should be treated as

unobservable variables.

lnEX_ijkt=α_t+α_j+α_k +β₁lnY_jt+β₂lnN_jt+β₃lnFDI_OUT_jt+β₄lnFDI_IN_jt

29

+β5lnKjt+β6lnMANUjt+β7lnMANU_LABOjt+β8lnDISTij

+β9lnINDUS_DISTijkt+β10EUjt+β11NAFTAjt+β12AFTAjt

+β13MERCOSURjt+β14APECjt+εijkt , (FE3)

lnEXijkt=αt+αj+αk +β1lnYjt+β2lnNjt+β3lnFDI_OUTjt+β4lnFDI_INjt

+β5lnKjt+β6lnMANUjt+β7lnMANU_LABOjt

+β8lnINDUS_DISTijkt+β9EUjt+β10NAFTAjt+β11AFTAjt

+β12MERCOSURjt+β13APECjt+εijkt , (FE3)

where αj and αk are the country and industry-specific effects, respectively. And the

difference of two equations is that the variable of distance between two countries.

Controlling for the year, country, industry, and country-industry pair specific effects (Model FE4)

Model FEFV includes the year, county, industry and country-industry specific

effects. The specification is as follow:

lnEX_ijkt=α_t+α_j+α_k+α_jk+β₁lnY_jt+β₂lnN_jt+β₃lnFDI_OUT_jt+β₄lnFDI_IN_jt

+β₅lnK_jt+β₆lnMANU_jt+β₇lnMANU_LABO_jt

+β₈lnDIST_ijkt +β₉lnINDUS_DIST_ijkt+β₁₀EU_jt+β₁₁NAFTA_jt

+β₁₂AFTA_jt+β₁₃MERCOSUR_jt+β₁₄APEC_jt+ε_ijkt , (FEFV)

30

where in the equation FEFV, αjk is the country-industry pair effects, and the other

variables are denoted in equation FEIT.

4.3. The modified model with two-stage estimation Stage 1.

lnMANUjt=αt+αj+β1lnFDI_OUTjt+β2lnFDI_INjt+β3lnKjt+β4EUjt

+β5NAFTAjt+β6AFTAjt+β7MERCOSURjt+β8APECjt+εjt , (FEIT)

lnMANUjt=αt+αj+β1lnFDI_OUTjt+β2lnFDI_INjt+β3ln(K/L)jt+β4EUjt

+β₅NAFTA_jt+β₆AFTA_jt+β₇MERCOSUR_jt+β₈APEC_jt+ε_jt , (FEIT)

where lnMANU_jt is the manufacturing output of country j, and α_t is the year-specific

effect common to Taiwan and other trading countries, α_j is the country-specific

effects, where (K/L)_jt is the value of labor force divided by capital stock of country j,

and the other variables and dummies are defined in equation (FECI).

Stage 2.

lnEX_ijkt=α_t+α_j+α_k+β₁lnY_jt+β₂lnN_jt+β₃lnMANU_jt+β₄lnFDI_IN_jt

+β₅ln(K/L)_jt+β₆lnDIST_ij+β₇LANG_ij

+β₈lnINDUS_DIST_ijkt+β₉EU_jt+β₁₀NAFTA_jt+β₁₁AFTA_jt

+β₁₂MERCOSUR_jt+β₁₃APEC_jt+ε_ijkt , (FE3-1)

lnEX_ijkt=α_t+α_j+α_k+β₁lnY_jt+β₂lnN_jt+β₃lnMANU_jt+β₄lnFDI_IN_jt

31

+β5lnMANU_LABOjt+β6lnDISTij+β7LANGij

+β8lnINDUS_DISTijkt+β9EUjt+β10NAFTAjt+β11AFTAjt

+β12MERCOSURjt+β13APECjt+εijkt , (FE3-2)

where in the above equations, we adopt three-way FE models to explain the

determinants of bilateral trade flows. The difference of above two equations is that in

equation (FE3-1) we use capital-labor ratio (K/L)jkt as the variable, and in the

equation (FE3-2) , we replace it with labor of manufacturing, MANU_LABOjt.

lnEXijkt=αt+αjk+β1lnYjt+β2lnNjt+β3lnMANUjt+β4lnFDI_INjt

+β5ln(K/L)jt+β6lnDISTij+β7LANGij

+β₈lnINDUS_DIST_ijkt+β₉EU_jt+β₁₀NAFTA_jt+β₁₁AFTA_jt

+β₁₂MERCOSUR_jt+β₁₃APEC_jt+ε_ijk , (FECI-1)

lnEX_ijkt=α_t+α_jk+β₁lnY_jt+β₂lnN_jt+β₃lnMANU_jt+β₄lnFDI_IN_jt

+β₅lnMANU_LABO_jt+β₆lnDIST_ij+β₇LANG_ij

+β₈lnINDUS_DIST_ijkt+β₉EU_jt+β₁₀NAFTA_jt+β₁₁AFTA_jt

+β₁₂MERCOSUR_jt+β₁₃APEC_jt+ε_ijkt , (FECI-2)

where in the above equations, we adopt two-way FE models. And the difference

between them is the variables of capital-stock ratio, K/L_jkt, and labor of

manufacturing, MANU_LABO_jt.

32

lnEXijkt=αt+αj+αk+αjk+β1lnYjt+β2lnNjt+β3lnMANUjt+β4lnFDI_INjt

+β5lnK/Ljt+β6lnDISTij+β7LANGij

+β8lnINDUS_DISTijkt+β9EUjt+β10NAFTAjt+β11AFTAjt

+β12MERCOSURjt+β13APECjt+εijkt , (FEFV-1)

lnEXijkt=αt+αj+αk+αjk+β1lnYjt+β2lnNjt+β3lnMANUjt+β4lnFDI_INjt

+β5lnMANU_LABOjt+β6lnDISTij+β7LANGij

+β8lnINDUS_DISTijkt+β9EUjt+β10NAFTAjt+β11AFTAjt

+β12MERCOSURjt+β13APECjt+εijkt . (FEFV-2)

in the above equations, (FEFV-1) and (FEFV-2) are the four-way fixed effect

models, controlling for the year, importer, industry specific effects, and

country-industry interaction terms. Equations (FECI-1) and (FECI-2) are

controlling for the year and country-industry pair effects.

33

Population (N_jt) Measured in one thousand people per unit.

In stock of US million dollars. UNCTAD Handbook of statistics, 1985-2010.

Distance (DIST_ij) Measured in kilometers. The World Factbook 2010 computed distance by The Chuck Taylor Web Site.

Trade Bloc Dummy (EUjt , NAFTAjt , AFTAjt , MERCOSURjt , APECjt)

1 if trading partner belongs to the specific trade bloc in year t.

The official website of each trade bloc.

34

Table 9. List of partner country of Taiwan

Partner countries Continent Partner countries Continent

Argentina South America Japan East Asia

Australia Oceania S. Korea East Asia

Austria Central Europe Malaysia Southeast Asia

Belgium West Europe Mexico North America

Brazil South America Netherlands Northwest Europe

Canada North America New Zealand Oceania

Chile South America Norway North Europe

China East Asia The Philippines Southeast Asia

Denmark North Europe Poland East Europe

Egypt North Africa Portugal Southwest Europe Finland North Europe Singapore Southeast Asia

France West Europe Slovakia Central Europe

Germany West Europe Spain Southwest Europe

Greece South Europe Sweden North Europe

Hong Kong East Asia Thailand Southeast Asia

India South Asia United Kingdom West Europe Indonesia Southeast Asia United States North America

Ireland Northwest Europe Viet Nam Southeast Asia Italy South Europe

35

Table 10. Trade agreements and selected member countries

Trade bloc Year (2004)*, Slovenia (2004), Slovakia (2004)*, Bulgaria (2007), Romania (2007).

AFTA 2002 Brunei Darussalam (2002), Cambodia (2002), Indonesia

(2002)*, Laos (2002), Malaysia （2002）*, The Philippines

（2002）*, Singapore (2002)*, Thailand (2002)*, Viet Nam (2002)*.

NAFTA 1989 Canada (1989)*, USA (1989)*, Mexico (1994)*.

APEC 1989 Australia(1989)*, Brunei (1989), Canada (1989)*, Chile

(1994)*, China (1991)*, Hong Kong (1991)*, Indonesia (1989)*, Japan (1989)*, Korea (1989)*, Mexico (1993)*, New Zealand (1989)*, Papua New Guinea (1993), Philippines (1989)*, Peru (1998), Russia (1998), Singapore (1989)*, Taiwan (1991)*, Thailand (1989)*, USA (1989)*, Viet Nam (1989)*.

MERCOSUR 1994 Argentina (1994)*, Brazil (1994)*, Paraguay (1994), Uruguay (1994), Venezuela (2006).

Note: Total number of member countries of each trade bloc is shown in a parenthesis in column one. In column three the member in a parenthesis is the year when a member country enters into the trade bloc. * denotes a country that is selected member in the dataset.

36

在文檔中 貿易流量之研究-以台灣製造業為例 (頁 31-42)