.Ilh. cowl
. ‘
"1 -.,

5'» l".

. "n

.1, ‘34..
, “c "at!"
“5.x 1 .
‘5 a)»:
g x _

1.

xi ,. 1;
.gt‘fiarf

: 1;“ a
’13 3" €335

um}:
3%:

x

'A
b

‘1")? .

 

9

m
“

5‘42
. .

3‘9

‘

..
.
”fit:

- 2.35:1?

.15
2“ :1» P 1‘
. 633$“ ‘

. m4-
, ‘4-”

fé‘vfifl

,.

p
t.- s
L . k

x?

Liza

.L
,’

 

 

150? LIBRARY
Michigan State
University

 

 

 

This is to certify that the
dissertation entitled

THEORETICAL AND EMPIRICAL ANALYSIS OF
INTERNATIONAL TRADE
WITH HETEROGENEOUS FIRMS

presented by

Na Yang

has been accepted towards fulfillment
of the requirements for the

PhD. degree in Economics

 

 

QWIM'

Major Pfldfe'ssror’s Sigjature

Arr/5M; 3%

Date

 

MSU is an affirmative-action. equal-opportunity employer

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DAIEDUE

DATEDUE

DATEDUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/07 p:/ClRC/DateDue.indd-p.1

 

 

THEORETICAL AND EMPIRICAL ANALYSIS OF INTERNATIONAL
TRADE WITH HETEROGENEOUS FIRMS

By
NA YANG

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

2008

ABSTRACT
THEORETICAL AND EMPIRICAL ANALYSIS OF INTERNATIONAL
TRADE WITH HETEROGENEOUS FIRMS
By
NA YANG

Chapter 1: The logarithmical form of the trade volume does not acconnnodate
the situation of zero trade l’)etween certain country pairs. This paper starts with
a tl'ieoretical framework of hetm'ogeneous firms to show the true property of zero
trade: “actual" instead of “potential” and proposes a two-part model framework
to deal with zero trade problems. The marginal effect derived from the non-linear
combination of the two parts significantly corrects the bias generated by estimating
the gravity model using conventional methods.

Chapter 2 (with Isao Kamata): In this paper we examine how factor proportions
determine the extensive margin of trade. Different from the existing research
that only analyzes the country-level export pattern in varieties, we explore the
problem at a disaggregate industry level. A quasi—Heckscher-Ohlin prediction for
export varieties emerges from the model: countries export more varieties in the
industries that more intensively use their abundant resources as input factors. The
model also delivers important implications of opening autarky to trade: besides
commonly-accepted facts of larger trade volume for each exporting firm, there is
also a stronger selection of firms into the export market in comparative advantage
industries.

Chapter 3: In this paper we explore the linkage between population health and
inward FDI in a cross-country setting. A semi-IV framework proposed by Frankel
and Romer (1999) is used to account for the endogeneity of FDI. We estimate
FDI with exogenous geographical variables and also investigate the effect of F DI-

indueed openness on health conditions.

To my parents

iii

ACI\'NO\-’\"LEDGEl\-IENTS

I would like to acknowledge many people for helping me during my doctm‘al
work. First and foremost, I thank my advisor. Steven Matusz. for his generous time
and ('ton'nnitment. Throughout my doctoral study he encouraged me to develop
independent thinking and research skills. I have greatly benefited from his valuable
feedback on this thesis and everything else I have written. I would also like to thank
all the members in my (‘lissertation conunittee, Dr. Carl Davidson. Dr. Susan 21111,
Dr. Joseph Gardiner and Dr. Jun-Koo Kang. for their guidance and mentorship.
I thank the Department of Economics for the financial support of my graduate
study all the way through thus making it possible for me to focus on my research.
I particularly thank Zhehui Luo and Dr. Joseph Gardiner. for giving me the.
opportunity to work 011 their projects so that I was extensixv'ely exposed to the
empirical study arena and for sharing with me their insights constantly, not only
on the e(-.()iionietrics problems, but also on life. which helped me to grow as an
individual. I thank Dr. J. Roy Black, to hire me as his Research Assistant when
I was just a first year PhD student and knew only the basics of STATA. Dr. Jun-
Koo Kang generously provided me the Japanese firm-level data he purchased and
loaned me the only copy of the disk he had. I also owe a lot to Professor Amsler for
treating me not only as a student and her 'I‘eaching Assistant. but also as a friend;
I am forever indebted to her encoiiragement when I was very frustrated with my
research.
I extend many thanks to my colleagues and friends, especially my coauthor. Isao
Kamata, and the friends from Lansing Chinese Christian Fellmvship.
I thank my parents because they have been a constant source of love, support
and encouragement. I thank them for their faith in me and allowing me to be as
ambitious as I wanted.

And I give thanks to the LORD, for he is good: for his mercy endureth for ever.

i v

TABLE OF CONTENTS

LIST OF TABLES ................................................................................ vi
LIST OF FIGURES .............................................................................. viii
Chapter 1

Treatment of Zero Trade Volume and Re-estimation of Gravity Equation: an Analysis
with Two-Part Model

1. Introduction ................................................................................. 1
2. Cross-Country Trade Pattern ............................................................. 4
3. The model ................................................................................... 6
4. Choice between Sample-Selection Model and Two-Part Model? ....................... 11
5. Specifications of the two-part model ................................................... 14
6. Estimation result .......................................................................... 19
7. Conclusion ................................................................................. 23
Chapter 2
Explaining Export Varieties: the Unexplored Role of Comparative Advantage
I. Introduction ............................................................................... 25
2. The model ................................................................................. 26
3. The Data ................................................................................... 35
4. Empirical test .............................................................................. 37
5. Conclusion ................................................................................. 40
Chapter 3
Globalized but Unhealthy? A Feedback on Population Health from FDI
1. Introduction ............................................................................... 42
2. How does FDI affect health ............................................................. 44
3. Data and specifications .................................................................. 50
4. Empirical analysis ........................................................................ 55
5. Conclusion ................................................................................. 57
APPENDICES
Appendix A .................................................................................... 64
AppendixB ..................................................................................... 68
BIBLIOGRAPHY ............................................................................... 105

LIST OF TABLES

Table 1: World International Trade and Production ......................................... 5

Table 2: Relationship between Composition of Trade & Country Characteristics, Year

2000 ................................................................................................ 6
Table 3: List of Countries ...................................................................... 74
Table 4: Summary Statistics .................................................................... 75
Table 5: Estimation of Gravity Equations by Conventional Methods .................... 76
Table 6: Estimation Result: 2-part Model without Sampling Weight ..................... 77
Table 7: Estimation Result: 2-part Model with Sampling Weight ........................ 78
Table 8: Marginal Effect Derived from Other Methods .................................... 79
Table 9: Marginal Effect of Iog(distance) and RTA on Iog(trade volume) (with Weight
Factor .............................................................................................. 80
Table 10: Marginal Effect of Iog(distance) and RTA on Iog(trade volume) (without
Weight Factor .................................................................................... 8]
Table 11: Vuong(1989)’s Test for Model Selection ......................................... 82
Table 12: US. Import and Varieties in 1990 ................................................. 83
Table 13: Country List (as of 1990, 115 countries) .......................................... 84
Table 14: Factor Abundance of Countries: Skilled Labor to Unskilled Labor ........... 85 I

Table 15: Input Factor Intensity of Industries: Skilled-labor to Unskilled Labor ........ 86

Table 16: List of Countries in Aggregate North and South ................................. 87
Table 17: Skilled-to-Unskilled Labor Ratios of North and South .......................... 88
Table 18: Regressions for Aggregate North and South ...................................... 89
Table 19: Pooled Regression for Individual Exporters ....................................... 90
Table 20: Summary Statistics ................................................................... 96

vi

Table 21: Countries List ......................................................................... 97

Table 22: Predicting FDI: Positive Volume Only ............................................ 98
Table 23: Predicting FDI: Using Two-part Model Approach .............................. 99
Table 24: Life Expectancy (Level) and FDI Ratio (log), Year 2002 ..................... 100
Table 25: Life Expectancy (log) and FDI Ratio (log) , Year 2002 ........................ 101
Table 26: Mortality (level) and FDI ratio (log), Year 2002 ............................... 102

Table 27: Life Expectancy (level) and FDI Ratio (log), Year 2002, With Two-part
Model Including All 170 Countries ........................................................... 103

Table 28: Mortality (level) and FDI Ratio(log), Year 2002, 170 Countries Sample...104

vii

LIST OF FIGURES

Figure 1: Cross—Country Distribution of Trade .............................................. 71
Figure 2: Geographical Distribution of RTAs (in force or under negotiation) .......... 72
Figure 3: Geographical Distribution of RTAs as of year 2000 ............................ 73
Figure 4: Number of Exporters to the US. in Each Manufacturing Industry ............ 90

Figure 5: Scatterplot of Number of Exporters v.5. Industry Skilled-labor Intensity (U.S.
Manufacturing Imports in 1990) ................................................................ 91

Figure 6: Scatterplot of Number of Varieties v.s. Industry Skilled—labor Intensity (U.S.
Manufacturing Imports in 1990) ................................................................ 92

Figure 7: Individual Exporter Regression: Scatterplot of Slope Coefficient v.s.
Abundance of the Country ....................................................................... 93

Figure 8: Exporter’s Relative Factor Abundance, Industry Factor Intensity, and Number
of Varieties in US. Manufacturing Imports in 1990 (1): Selected Skilled Labor-
abundant Countries (relative to unskilled) ..................................................... 94

Figure 9: Exporter’s Relative Factor Abundance, Industry Factor Intensity, and Number
of Varieties in US. Manufacturing Imports in 1990 (2): Selected Unskilled Labor-
abundant Countries (relative to unskilled) ..................................................... 95

viii

Chapter 1: Treatment of Zero Trade Volume and

Re—estimation of Gravity Equation: Analysis Using
Two-Part Models

1. Introduction

The gravity model has a long history in the trade literature. The original
gravity model assumes the goods produced at an origin and attracted to a destina-
tion are proportional to the productions of the two locations, and the friction term
adds the impedance of making travels of various durations or distances. Tinber—
gen(1962) was the first to use gravity model for estimation of bilateral trade flows,
by substituting the production levels with two countries" GDPs and using trade
resistance for the friction term. Over time, his approach has been vastly cited and
furnished with better empirical estimation teclmiques and used to explain other
aspects of international trade than the trade volume only. For example, Anderson
and van Wincoop (2003) showed that the gravity model should be augmented with
exporter and importer fixed effects because the traditional one does not take the
multiple resistance terms into account.

In his seminal paper, Krugman formalized the role played by the geographical
proximity in the regionalization process. In Frankel and Romer (1999), they showed
that regionalization could be explained by geographical proximity and preferential
trade agreements, with country sizes being constant. Moreover, Rauch (1999)
showed that differentiated products could exhibit. stronger geographical proximity
effects than homogeneous products within a gravity model structure.

Though the gravity model has been widely recognized for its empirical success
in predicting the trade volume and estimating the effects of the factors that impede
bilateral trade, it initially did not. have a strong theoretical background. 1Recently
there has been an increasing trend to use both the traditional and new trade

theories to derive the gravity model. For instance, Deardorff (1995) derived it

from a traditional Hechscher—Ohlin perspective while Eaton and Kortum (1997)
used a Ricardian framework. Additionally. Helpman(1987) used both monopolistic
competition model and gravity equation for an analysis and argued that the close
of fit. of the gravity model of trade variable could serve as supportive empirical
evidence for the 111011<)1i)()listi(- competion model.

For empirics. Learner(1974) used both the gravity equation and the Heckscher-
Ohlin model to motivate the explanatory variables in a regression analysis of the
trade flows. Notably in Helpman(1987). he. applied his test to data on trade
of the OECD countries, which yielded supportive evidence to the monopolistic
competition model. Hummels and Levinsohn (1995) expanded the test to a much
wider variety of countries, using different data and different estimation methods
to test whether the data still supports the theory.2

All in all, these studies have improved our 1.111derstanding of the gravity equa-
tion as a tool to model and analyze bilateral trade. But what is common in the
existing research is that the. analysis of gravity model was only applied on sample
countries that have positive trade flows between them. However, a direct result. of
discarding zero trade volumes could result in biased estimates. In this paper we
developed a heterogeneous firm model following Melitz (2003) then used the theo-
retical predictions to show how to amend the gravity model to deal with zero trade
flows. The heterogeneity is introduced in a similar way to Melitz (2003): firms face
uncertainty about their future productivities: the entry fees for both producing for
the domestic market and exporting to foreign markets are costly and irreversible.
The model delivers a set of implications for export probability and export volume.
Similar to the l\"Ielitz's model, this paper allows the fact that none of the firms in
one. country has such productivity level that they can break even when they enter
any other foreign market. As a result, the model incorporates the likelihood of

zero volume of trade l’)etween some country pairs as well as positive trade volume

in one direction from country ’1'. to country j, but zero trade flow from country j
to country 2. For predicted positive two-way trade, the model generates a gravity
equation in which the size of trade flow is proportional to the sizes of the partners,
but dampened by bilateral barriers.

Using various sources, I then assembled a dataset on bilateral trade and gravity
measures for all the countries. The availability of the gravity measures makes it
possible for us to accurately construct gravity measures for both country samples
with zero trades and positive trade volumes.

Along the lines of research on gravity equations. the paper fundamentally dif-
fers from the existing research in two dimensions. Theoretically, it uses Melitz(2003)
”s heterogeneous firm model to derive the gravity equation for both the export
probability and volume besides the established Hechscher—Ohlin structure and tra-
ditional monopolistic competition model. With the heterogeneous firm setting,
we rationalized the true feature of zero trade, which is significant suggesting the
proper handling of zeros. The generalization of the gravity model accounts for the
asymmetries between the volume of exports from 2' to j and the trade volume from
j to 2'. But different from Helpman, Melitz and Rubinstein (2006), the firm-level
l’ieterogeneity term does not enter our estimation specification on the country lev 31.
Empirically, we propose a two-part model, which has been intensively used in other
fields so that we can better address the problem of zero trade and correct the po-
tential bias generated by the conventional estimation methods. We also explained
how to compute the correct marginal effects of covariates on actual outcomes with
different distributional specifications. A Vuong’s test. is used to sort. out the best
fit of the model for actual trade data. In a. recent paper, Helpman et al. (2006)
proposed a technique similar to sample selection model. We argue that two part
model is more appropriate than sample selection in handling the problem of zero

trade based on the fact that trade values are actual instead of potential. In two

papers by \Yesterlnnd and \Vilhelmsson (2006), Silva and Tenreyro (2000), they
proposed a Poisson maximum likelihood estimator(Pl\ILE). Although the PLME
outperforms conventional OLS by removing the need to linearize the model, it
makes strong assumptions on count data regression that might be inapplicable
when dealing with continuous trade volume.

The remainder of the paper is organized as follows. Section 2 presents the
cross country trade pattern. In section 3 we use. the model of heterogeneous firm
to clarify the true. property of zero trade— actual instead of potential outcomes,
and it is addressed that there is no selection bias problem when modeling the actual
trade. outcomes. Section 4 studies the. econometric problems raised by conventional
methods and section 5 proposes two-part estimation techniques with different sets
of parametric distributions. We also explain how to compute the correct marginal
effects for covariatesein this paper, distance and regional trade agreement. Addi-
tionally. a Vuong test is used for model selection. Section 6 provides the informa-
tion of data and estimation results. The results are. compared with those. generated
by OLS, adjusted OLS and non—linear least squares estimations. Section 6 contains

concluding remarks.

2. Cross-country trade pattern

Since 19908, international transactions are playing an increasingly important
role in world trade. Table 1 shows the world GDP and world export growth path
from 1982 to 2001. The volume of export approximately constitutes around 20%

of total world GDP. and grows at a steady rate in levels.

Next we show the composition of country—pairs according to their trade sta-

tus in figure 1 - country-pairs with two-way trade. one-way trade and no trade

 

 

 

 

 

 

 

 

 

1982 1990 1996 2001
World GDP (in billion dollars) 11.758 22,610 29,024 31,900
W'orld export (in billion dollars) 2.247 4.261 6,523 7,430
World export as 0/0 of GDP 19 19 22 23

 

 

Table 1: World International Trade and Production

at all. Years 1980-1997 are considered. From 1980-1997, the country pairs that
are involved in either bilateral or unilateral trade constitute around 3070—5001, of
all possible country pairs.3However. the proportion of one way trade relationships
stays at a fairly constant level of 10% along the early 90s. These years also wit-
nessed an increasing share of country pairs with positive two-way trade.

Generally, the gravity approach suggests that trade volume is a function of
trading partners’ sizes and trade barriers. _ The GDPs are usually used to reflect
the sizes of exporters and importers. The importer’s market size. represents the
market demand for bilateral trade, and exporter’s size reflects the potential com-
modity supply. Geographic distance is usually used as the term of resistance,
along with other binary variables to proxy other aspects of economics integration
factor or trade barrier factor. Using Feenstra(1995)’s trade dataset for year 2000
and country-pair characteristics, in Table 2 we show the correlations between the
possibility to trade and gravity variables.

Particularly, the average distance among the group of country pairs that do
not have any trade is much bigger than country pairs that. have positive trade.
The average GDP products are also substantially larger for countries with positive
trade. In fact, the GDP products for country pairs with two way trade are almost.
as twice as that of country pairs with one way trade. and almost 20 times of the
products for countries with zero trade. Two-way trade country pairs include larger
proportion of both countries affiliated with the same regional trade agreement
compared with one-way trade and zero trade. However, the benefit of sharing the

common language is not quite clear, this could stem from the fact, of the large

 

 

 

 

 

 

 

 

 

 

 

 

two— way one-way zero

trade trade trade
Mean distance(in kilometers) 6959 8027 8495.82
(X with common language 12.10 12.82 17.96
(7t with connnon border 2.95 0.69 1.14
GDP products(in e+19 US $) 17400 8220 922
(it of landlocked exporters 13.15 15.58 21.05
(7(. of landlocked importers 13.15 18.22 20.87
mean of # of exporter major cities 21.10 22.92 23.85
mean of # of importer major cities 21.05 23.69 23.85
Ct of country pairs with common RTA 7.25 4.14 6.42
Number of country pairs 8,940 1.739 25,611
ft of all country pairs 24.63 4.79 70.57

 

 

 

 

 

 

Table 2: Relationship between composition of trade and country characteristics,
year 2000

economic variation among Ei‘iglish—speaking countries. In summary, the evidence
presented in the previous tables and figure 1 suggest that size and geography are
very important factors to explain the existence of trade and the. level of export

volumes.

3. The model
The model of trade with l’ieterogeneous firms is built 111) based on Melitz
(2003)4 The world is comprised of J countries, j=1,2 ....... .l. ()n the demand side.
the preferences of a. representative consumer are given by a standard C.E.S. utility
function over consumption of a continuum of goods x (n varieties) indexed by l.
The varieties of x are imperfect substitutes. where (7 = 1/ (1 — p) > 1 indexes the

substitution pattern between varieties of x. Country j‘s utility is expressed as:

,n, 1
UJ- = [/Jrawdzjfi (1.1)
0

\Vitli c(.)nsumers maximizing their utilitities subject to the budget constraints,

it. leads to the demand of variety I in country j:

6

.rJ-(l) = riff—115(1)“ (1.2)

n 1
where Y]- is the total income of countryj and Pj = [f p(1)1—0(11]1——? denotes
0
the aggregate price index; pj is the price of variety 1 in country j.

For production, labor is the only factor of production and is inelastically
supplied in a market. The unit price of labor in country j is j‘)arameterized at cj.
There is a large pool of 1,)ropective entrants into the industry.

The production of (‘lifferentiated good .1‘ is characterized by monopolistic com-
petition. All the potential entrants are the same ex ante. To enter the market,
firms in country j need to incur a sunk cost fncj so that they can draw their
productivities (,3 from a distribution C(99). Once the firm knows its productivity,
it needs to choose to produce or exit. Successful entrants could make nonnegative
profit. with high enough productivity; however, if the productivity draw is below a,
cutoff level. it is best. off exiting at. once.

The productivity t;(l) for firm to produce variety I is drawn from a Pareto
distribution with the range of [ 1,9113ij and shape parameter [3.5

The total cost for production of variety 1 at (1 quantity is comjn‘ised of two
parts: the variable cost and fixed cost ( f D units of labor).

qc-

53(7) + chj (1.3)

In a monopolistic competition characterized economy, the pricing rule of a

 

profit-maximizing firm with efficiency level to is:

15(1) = —’— (1.1)

For a given firm, we could simply express the profit of the firm (with product

I) from domestic sale as:

c-
,. 7 1—0
win=e—pn-e—e—) —hx- (Le
J J WUlpj J
The same derivation applies to the profit from trade, where Tj, is the “melting
iceberg" cost which is bigger than one; equally speaking le- units of goods need

to be shipped in order for one unit to arrive. Thus the profit for the firm (with

product I) from country j to export to cmmtry i is:

man=o—«ngg§%§fica—tmj (Le

f ji is the units of labor needed as fixed cost wl‘ien firms produce at country j
and export to country i, and it is larger than f D, which is needed to produce for
domestic market. Evidently with this assumption, this profit is positive for sales
in the domestic markets, because fji is bigger than f D- It also follows that only

a proportion of country j’s firms could export to country i.

The free entry condition for firms from country j implies:

fer = (1“ G(s3}))fgh(ej)+(1- Gwzillfji‘jh(WEfl) (1-7)

and similarly for firms from country 2',

feCi = (1‘ affillfC—ihbifl + (1 — C(QLJUfijCihf‘PLJ‘) (1-8)

~‘

,,, , >1:
where h.(;,9*) = ( YE; ))U—1—1 and 99:17 is the benchmark productivity needed

 

for exporting to happen from country j to country i. Similarly, 99:1}. is for export
from country i to country j. (,9: identifies the lowest productivity level to produce
at home.

Given these conditions. we can obtain the following:

Lemma 1:

 

@2131 " lei—”>07 —=91 (1.9)
* f

99.171? fji .—1 (‘1' 01,-

9”? f (Ci) Yz‘ 2

Proof: See appendix A

By substituting (1.9) and (1.10) into (1.7) and (1.8), it leads to a system of

equations with two unknowns (,9: and (,9; :

 

 

 

 

. 99L k
/.9 :19 (K —1)(:}.> +fJ.9(K—1)(Q:.,,2> (1.11)
fe(i: f0.“ A — 1)(:§)k +fsz-i(K _ 1)(¢::51)k (1.12)

where K = 11%;. Solving the system yields:

6 k
, _. (If—1w "
(.9) "= L (1.13)

f f2?!
f+ 13277275?

From equation (1.8) and (1.12). the export. productivity cutoff is determined

 

by :

* f__’-_] 7.0— 1 010(K—1)9‘;[: -1
Y’rlj: f (P ) ( 792

j f+ f‘flT—fi—T

where {21 and 92 are given by equations (1.9) and (1.10). which only consist

 

(1.14)

of exogenous variables.

Define the ratio of cutoff productivity to export. and the highest productivity

1"

V’Hi as:

 

 

Dji = ‘ (1.15)

The. probability for country j to export to i is determined by prob(DJ-,j < 1).
in another word. under the circumstances that the cutoff point to export (from j to
i) is bigger than 99;, the bilateral trade volume equals zero. The proportion of the
firms that could be involved in exporting activities in a given country is determined
by function: A((5,j. (ij. 60-): where 9k (k = i,j) is country-specific characteristics

and 61-]- is bilateral gravity covariates. From equations (1.9)—(1.10), the term (.931.

 

and 93:0- export productivity cutoffs are also determined accordingly.

When trade is possible for two countries, or at least for trade in one direction,

the level of trade would be:

0.7..
J 1’ Nit-

 

Tji =( (2”‘l(IG(-.a) (1-16)

\' 491(-

991' i

This function is again a function of two countries’ variables and bilateral
gravity covariates, which we. will generalize to the function TM 1‘, 63-, (SI-j).

For empirical fran'iework. equations (1.15) and (1.16) are related to export.
probability and export volume separately for (,:ross-l:)or(ler transactions from coun-
try j to 1'. Both the probability to trade and trade volume between two countries
could be decomposed into three components as Helpman, Melitz and Rubinstein
(2006) showed: one that depends on importer characteristics. one that depends on

exporter characteristics, and a. third that depends on the country pair characteris-

10

tics. The decompositition resembles Anderson and van W incoop (2003)”s gravity
equation that embodies all three sets of variables. But unlike Helpman, l\1‘Ielitz and
Rubinstein, we do not include the additional term that controls sample selection
bias and proportion of the exporters. Instead. we address the question using a two-
part model. The distinction between the sample selection model and the two part
model is discussed in section 4.2. Other than model selection due to the property
of zero trade. empirically. the two part model also outperforms sample selection
because the downward bias for the coefficients of the covariates could arise from
the high colinearity between the inverse-Mills ratio and the countries’ covariates if

sample sele("tion is used.

4. Choice between sample selection model and two-part model

The two-part model has been used intensively in the field of health economics
to deal with data that. includes a large fraction of zero values, such as the cigarette
demand, hospital utilization and health insurance coverage. The sample selection
model is often misapplied to the corner solution problem. thus deriving the wrong
marginal effect as their main interest.

This debate on the choice over Heckman sample selection model and two part
model (hereafter, 2PM) went back to the famous “cake debate” of the 19803. Jones
in Handbook of Health Econon‘iics documented an excellent history of the “cake
debate”The sample selection model has dominated much of the literature in mi-
croeconomics and the Heckit estimation procedure is routinely adopted to analyze
the problems involving censoring and selection bias. The earlier comparisons are
mostly based on theoretical issues 6and the recent investigations have. turned to
Monte. Carlo simulation experiments7. There seems to be. a general 111isunder-
standing of the terms “censored” and “selected" samples, especially when applied
to the two-part model setting. As \Vooldridge (2002) points out. a second kind

of application of censored regressirm models appears more often in econometrics,

11

and tmfortunately. is where the label “censored regression” is the least appropri-
ate. Though in many situations the problems we are trying to solve arise from an
optimization problem and the true feature is corner solution. the name “censored
models" appears to be more entrenched. Besides V‘Vooldridge (2002), Dovr and
Norton (2003) also clearly addressed several issues regarding the merits and usage
of the Heckit and two-part model when applied to data with a large chunk of zero
values.

\Vhen choosing between the two models, we need to first. distinguish between
potential values and actual values. Both methods are used when dealing with
continuous outcome variables with a large portion of zeros. But the choice over
the two models depends 1_)rimarily on the distinction between actual value and
potential value.

\Vhen dealing with trade volume, an actual zero volume is observable, whereas
positive volumes generally exhibits a skewed-to—tlie-right continuous distribution.
For example, in a specified period patients either have zero health expense or a
positive expense, but not negative expenditure; and in the same fashion the zero
trade values in our heterogeneous firm model are corner solutions ——-t,.rue zeros
instead of missing values, therefore we do not have a sample selection issue to
address.

In contrast, the potential outcome is a latent variable that could not be fully
observed. Zeros do not represent the fact that true values should be zero. In
labor economics, observations without positive wage outcomes do not. imply that
these people would work for zero wages; instead these wages are non—observable.
In the same way. potential expenditures that have never occurred would not affect
the health care budget. Dow and Norton (2003) also gave an example of this
type. For a person with zero health expenditure, it does not mean his potential

expenditure would be zero if he had been examined by a doctor and indeed had

12

sought any health care. The Heckit model would work better in this situation
because the observed working people are likely to be different. from the unobserved
non—working people.

From our theoretical derivation of the last section, we are able to show that
the true feature of zero trades here is indeed corner solutions for country pairs
because countries choose not. to trade when all the firms in their counties have
productivities below the necessary benchmark. Therefore two-part modeling is
more appropriate when zeros are the actual observed trade volume between two
countries and we are interested in the determinants of the actual trade.

In mathematical terms, both models consist. of two equations. For Heckit
model, the first equation models the probability of having a positive value (se-
lection equation), and the second equation expresses the mean trade volume in
the sub—population with the positive trade volume (the conditional equation), the

conditional equation usually takes the form of:

E(yly > 0, X) 2 X232 + 71/\(X1;31) (1.17)

where /\(X1,r31) is the inverse-Mills ratio term under the assumption of normality
of latent variable that denotes the potential trade volume. However, in the two
part model, the second equation does not include this term due to the nature of

the zeros, which makes it simply:

E(y|y > 0, X) 2 X233 (1.18)

Therefore, when sample selection is inappropriately invoked for actual values.
0 {'32 will only equate ,8; under special case where there is no selection bias(71 = 0).
The two part model should not be confused with standard Tobit model because

the two part model allow the hurdle decision (part one) to be separate from the

13

level decision (part. two). Some studies that followed the practice of standard
Tobit to estimate gravity equation with zero flows include Rose (2004). Soloaga

and \Vinters (2001) and Anderson and l\.lar('touiller (2002).

r

0. Specifications of the two-part model

Two-part model is sometimes mentioned in other fields in alternative ways—
two-tier model or hurdle model. Different from sample selection8, the property of
latent variable. is actual instead of potential. Parametrically, we model the first
part in the following specificationsprobit and logit. For the distribution of the
error term in the second part, we will use lognormal and exponential distributiong.

The set up and the merits of the models are discussed below. A Vuong’s test is

used at the end of section 5 for model selection.

5.1 Lognormal Distrilimtion
The estimation procedure works as follows. We define the dummy variable

10, otherwise

from our trade theory as Di, therefore Di means two countries trade
no trade exists. In the second level equations, we use Ti for observation ‘2'. (a unique

country pair)‘s trade volume.

I.Probit and Lognormal:

 

When part one is probit and second part is lognormal, two equations take the

f(i)llowing forms:

I I ,
Pa), = 1|X1,-) = magi/31 + 51 > 0) = <I>(X1,:x31) (1.19)

I
E(T,:|D, = 1, X”) = exp(X2,I-;32 + 0.502) (1.20)

14

where the error terms follow the distrilmtions: €1~N(0, 1), and 111(52) ~ N(0, 02)

The unconditional expected value would be:

, _ . r . y! _ _ r/
E(Til.\1,‘..\2;) = E(T1lDi = 1,}121') P1‘(DZ' = 1'21“) 2 CXl’b’x‘2).v"32+0-002)(D(AM531)
(1.21)

The log-likelihood function will change accordingly to

I
111 Ti — Xog-HQ

0'

 

I
(71(Tl') =111(<I>(X11-:31))+ln((.t)( )) — 1110le (1.22)

and marginal effect of variable .rk is:

013(3)
(9:121C

 

I I I I
: <I>(X1.l-131);32k exp(X2,-;32 + 0.502) + @(Xlil31)131kexp(X2z-/32 + 0.502)

(1.23)
For estimation. when we use total trade volume for level of Ti and for example,

.rk is the log form of distance. the elasticity of distance 011 trade volume is:

0E(ln trade)

I I I
= <1) X 43 13 .. X #3 )3 . X )3 1.24
0(lndistance) ( 1" 1) 2k +6“ 11* 1) 1A( 21 2) f )

 

The marginal effect of Regional Trade Agreement would take a similar form

to average treatment effect:

r/ I, _ I . 'I :- 2 fl /
exp(.X2’-132 + 0.:102)<I>(X1’-131)|RTA:1 — mph/31213132 + 0.00 )<I>(.X1,i,131)|RTA:0
(1.25)

11. Logit and Lognormal

 

Similarly, we can also compute the 1.1nconditional expected value when part

one is legit and second part. is lognormal,

 

 

 

I
, ,I exp X 4'3
E(T,'|.X1.Ij. XQI) = GXI')(A22-5132 + 0.502) (I 1’ 1) (1.26)
exp(.X'17.,z31) + 1
And the log-likelihood function is:
(v' a ) 1 T X' 3
ex) . 11 .- — -:.
ln(Ti) : In( I ‘I 1" 1 )+ ln(cb( l 2" 2)) — lnaTi (1.27)
exp(XlI-f31) + 1 a

and marginal effect of ln(distance) would become:

’ .
exp(X1l-_x31)
I
(exp(X1_l.)31) + 1)2

I ,
I
ex1)(X1_i,:'31) + 1

 

 

I I
1321. exp ( X2]: :32 +0502 ) + 1311., exr)(X2.,3132 +0.50?)

 

 

 

 

(1.28)
It can be shown that the elasticity of distance on trade is:
(X' 3 ) (r' 3 )
exp ,7)" 1 / exp 1 31’ 1 , r/
I 1,1 (5211: + I _1? 21131114912132) (L29)
9Xp(X1i/31) + 1 (GXP(X11'(31) + 1)
while the effect of RTA 011 trade is:
(Y' 3 >
I V 9 exp .4) 1
(¥X})(1Y2?-1’32 + 0.50“) I if iRTAZI_
ex1;)(X1i/31) + 1
(X' 3 )
I , exp 1
e.rp(X2,iI32 + 0.502) 1’ 1 lRTAzO (1.30)

I ,
(axr)(X1i;}1) + 1
For the standard gravity estimation, which uses positive trade volumes only,
the marginal effect of distance is 1321., which would be a biased result if we want.

to know the true effect of distance on the level of trade. What we are interested in

16

HE (111 trade)

should be 0E (1n trade trade>0)
L ' ' 8(111 distance

8(ln distance)

 

 

 

) instead of . In another word, we need
to investigate the effect of the barriers on overall trade for all countries, not. just
the sub-sample of positive trades. For both sets of models (probit. and lognormal,
logit and lognormal), it can be shown that I). affects E (T,) in three ways: through

its effect in the hurdle equation, captured in ,[31 ; through its direct effect in the

conditional equation (captured in 1'32); and through the density function of equation
I
eXp(X1,'(31)
I
(expuumal)?
The two part model is different from sample selection in that it does not

 

I I I I
one (oh-‘11,)? 1) 111 model one and 111 model 2).
include an inverse-l\eIills—ratio part due to the zero trade property, therefore there
is also efficiency gain in 2PM because inverse-Mills ratio term usually generates
multicollinearity problem. When dependant variables are. actual values and sample
selection is mistakenly used, the nus-specification is analogous to adding a single

higher order term of X1 in the second equation.

5.2 Exponential distribution:
Another leading case for functional forms that ensure a positive value for the
second part is exponential function. In that case, our second part’s function would

take the form of:
I
E(T,:IX2,:. Ti, > 0) = 0XP(X2;/‘32)

A commonly used parameterization is to define the probability density func-

tion (pdf) of an exponential distribution as:

1 ‘s
f(.r) : Xexp(—§) if .r >2 0

17

= 0 if .‘I‘ < (l

where A > 0 is a parameter of the distribution. The exponential distribution
is used to model Poisson process, in which situation an object initially in state A
can change to state B with constant probability per unit time A. The exponen-
tial distribution may be also viewed as a continuous counterpart. of the geometric
distrilmtion that describes the number of Bernoulli trials necessary for a discrete
process to change state.

A detailed description of the derivations of log likelihood functions and marginal

effects are discussed in Appendix A2.

5.3 Vuong’s test for model selection:

In this section, we invoke Vuong (1989)”8 test. to select from the four sets of
modelslland choose one that is the closest to the true trade data generating pro-
cess. The models are considered non-nested if neither models can be represented
as a special case of the other. Models with different non-nested distributions and
models with different non—nested functional forms for the conditional mean are
called strictly non-nested. The formal definition is given in Vuong (1987) and Pe-
saran (1987). In the econometrics literature, starting from Cox (1961, 1962a), the
hypothesis testing is performed in a non-standard framework. A brief review is
given in Davidson and MacKinnon (1993). Mizon and Richard (1986) proposed
the encompassing principle, which leads to a quite general framework for testing
one model against the other. VVooldridge (1990b) derived encompassing tests for
the conditional mean in nonlinear regression models with heteroskedasticity. In
this paper, we follow Vuong (1989) for model selection by discriminating between
models on the basis of their distance from the true data—generating process, where
distance is measured using the Kulback-Liebler information criterion. The follow-

ing statistic is proposed:

18

TI '7 )-

(1),].1 1 f(_(__'__/,l.r,,91 f(g,|.‘r,.6)
TLR1\N=\/-Zl “H"; ————)2) -(;;1n—ffi_—A—)2}

(Uill 9(9le ’1) :1 “Milt/7)

where

TLR.NN —> 1N70'r'rll.(11(0, 1)

under the null hypothesis:

f(;I/il1'2:~ 9)

H :E
0 h‘[g(yzi|fl‘i~?)

(:0

We will reject at significant level 5% the null l'iypothesis of equivalence of the
models in favor of F being better than G if TL RN N > 35% (or TL R, N N < —:5% ).

The null hypothesis is not rejected if lTL R, N Ni < 22.5% .

6. Estimation Result

6.1 The data

The goal is to get consistent. estimates of the parameters 011 observable barriers
and calculate bilateral costs of export. Considering the fact that. country-specific
variables and barrier variables. such as distance and common language, do not
vary much in time dimension. we instead investigate the cross-sectional feature of
the model without looking at the panel data.

We use year 2000 world trade data, which makes the analysis cover a cross
section of 191 countries. Therefore the data consists of 36290 (191*190 country
pairs) observations of bilateral trade flow. Table 3 provides the list of country
names. Out of this number, 25611 unique combinations of these countries have
zero trade. which is around 70%.

Information on bilateral trade comes from World trade. data compiled by

19

Feenstra etc. (2005). Countries GDPs are from Penn \Vorld Table 6.2 and com-
1_)le1nented by the World Bank's “forld Development Indicators (2002). Gravity

12and colonial ties13

measures. including dummy variables for contiguity come from
various sources: C EPII, CIA’S \Vorld Factbook and Jon Haveman’s website. The
bilateral distance is calculated following the great circle formula. which uses lati-
tudes and longitudes of the most important city (in terms of population) or of its
official capital.14The number of major cities for both exporters and importers are
from Hendersons \Vorld City Data. Table 4 1.)rovides the summary statistics of
the variables.

The data on regional trade agreement (RTA) is constructed from World Trade
Organization documents on RTA dated on Oct. 11, 2000. Figure 4 shows the level
of engagement of individual countries and customs territories in RTAS in the year
“2000. The figures indicate the total number of each country’s trade agreements.
Such a map allows for a quick comparison of different countries and regions. Some
countries are not involved in any RTA, while others are signatories to more than a
dozen. We use a simple. dummy variable to represent the RTA relationship between
country i and j. so I},- 21 if country i and j are in the same RTA and 0 otherwise.

6.2 The result.

The exact specifications of the 2-part model of the gravity equation used are
as follows:

E(DJ-, = 1) = f[)30+1‘31*ln(I",;)+;32*ln(YJ-)+1'33*ln(dist,j)+54*e.r_la:n(llk+
135 * I'm -10 mil}; +136 >1: com .10 n +137. * com_bordcr+1’38 * Clone/rial +139 * c.r-citics +
1310 * inLc'itics +1310 * III]

In T], = 1’30 +131 >1: ln(Y,) + [32 * ln(Yj) +133 * ln(dist,j) +134 * (11:10thde +135 *
iIIIJIlIId/k + (36 * comeL + 137 * ("(IIII._b(‘)-I'der + 538 * Clo-name] + 139 * (5’.z'_c'itics +

[310 * IIII,_(,“lifI(J'.S' + .1310 * [N + EU
where. the subscript. 'i. and j denote the trading partners, exporter for j, im-

20

porter for i:
Y1. (I; = 21]) is the GDP:
distij is the distance between the trading partners;
co-I'IIJII'II‘I'It—‘I‘ is the. dummy variable for common border;
Clonom’al is the dummy variable for colonial relationship after 1945;
eJ‘Jand/k takes the value of unity if the exporter is landlocked;
imJand/k takes the value of unity if the importer is landlocked;
e.r_citics is the number of cities in exporter country;
-iI‘I2_(‘-Ities is the number of cities in importer country;
I fl = 1 if country 2' and j are in the same RTA;

E ,j j represents the omitted other factors that influence the bilateral trade;

The functional form of f(.) is probit or logit, and error term 5,- j will follow
the 1'>reviously specified lognormal or exponential distrilmtions.

Table 5 presents the estimation outcomes of various techniques proposed by
traditional 111ethods. The first column corresponds to the OLS estimates by using
the logarithm forms of exports as dependent variable, and only using the subgroup
of country pairs with positive trade value. The second column uses OLS as well
but the dependent variable is in the logarithm form of sum of one and trade value,
in this way we could also include zero trade observations in the estimation to deal
with the problem of no log form for zero. The third and forth columns correspond
to Poisson 1)seudo-maximum likelihood estimation proposed by Silva and Tenrcyro
for all observations and for positive trade only.

In table 6 and 7, we show the results from two part models. Table 6 contains
our benchmark estimation. In another word, we treat all the country pairs as
equally important without assigning any weight. However, it is not realistic to
treat the trading relationships equally for country pairs like US and China and

country pairs like Haiti and Uganda, two small developing crmntries. Therefore in

21

table 7, we use two countries’ GDP product. as a weight, so that larger country
partners are given bigger weight in the 2PM estimation.

The various estimations reveal that the coefficients of log of the GDPs, are
consistent with general belief, which is close to one. The coefficients of gravity mea-
sures, however. differ from each other in different models. But the general message
is that countries further apart trade less, while larger countries trade more. The
effects of GDPs and distance measures are large and highly statistically significant,
which is in line with the estimates from the literature. Countries belonging to the
same regional trade association trade more, as do countries sharing the same lan-
guage. the same border, and a shared colonial history. Other gravity effects are
quite mixed. The table also clearly shows that the same variables affect both the
probability of trade and trade volumes.

Comparing the second part of 2PM and OLS for positive trade sub-sample, the
following observations are in order. The elasticities of exporter/ importer GDPs,
and distance are substantially smaller under 2PM. Common colonial ties have
stronger effects under OLS. The negative coefficients on number of major cities in
exporters and importers in both models lead to some puzzling results.

However. the major interest of this research rests on the marginal effect of the
distance and RTA on the level of trade. The results from other models are reported
in table 8. And our results from 2PM are reported in Table 9 and 10 separately
for estimations without and with sampling weight. Though the marginal effects of
distance are quite similar for results of unweighted 2PM and weighted 2PM, the
weighted 2PM marginal effect of RTA is relatively smaller than the unweighted
result. This could be due to the correlation between existence of RTA for certain
country pairs and their trade volume. Therefore, countries that are involved in
RTA are also assigned a larger weight. However, this correlation for distance is

less apparent. The two model spet‘fifications for 2PM yield larger marginal effect

than traditional methods except OLS on positive samples only.

The result for the Vuong test is reported in Table 11. The test is applied to
probit versus logit and then lognormal versus exponential distributions separately.
As shown by LR statistics, there is no significant difference between the methods.
The LR, tests suggest that we can not reject one model against the other. As
a matter of fact. how to specify the parameterization seems to be rather a less
important issue compared with the choice of an appropriate model to deal with
the zeros. To summarize, we have been able to find evidence that traditional

methods generate biases and are economically incorrect.

7. Conclusion

In this paper, we provide a thorough analysis and discussion on the methods
applied on zero trades, in light of prevailing knowledge on the theoretical foun-
dation of gravity model and the econometric techniques. The general message
is that estimates are. biased if zero values are discarded. Additionally, applying
two part model to deal with the zero trade problem is more appropriate than the
conventional models, such as sample selection. We showed persuasive evidence
by ctnnljmrisons between our estimated result and the existing research. The es-
timated elasticity of distance 011 trade value and average treatment effect of RTA
are significantly different. in two part models from the existing models.

To understand the correct marginal effect of trade barriers is essential for our
understanding of international trade pattern and policy makings. What policy
makers should analyze is the marginal effect of a trade agreement on overall trade
level for all possible country pairs, as we showed in the paper. not just for countries
that. are involved in bilateral trade.

Several caveats are in order. Due to the limitation of data. we can only explore

the total trade up to the year 2000. A potential future work is to include more

23

years when data becomes available and look at the time series. Another problem
caused by the unavailability of trade data is the precision of the RTA effect. As
Baier and Bergstrand (2006) showed, a foreign trade agreement approximately
doubles two members" bilateral trade after 10 years. which could not be addressed
in our cross sectional data. However, the essential problem we want to address in
this paper is the appropriateness of different models for application of zero trade
volume and we want to provide the policy makers with more alternatives on an
unbiased estimates of the marginal effects of distance and average treatment effect
of RT A.

The supportive evidence for 2PM against other methods we presented may
seem surprising in light. of the prolifr—u‘ation of the traditional ways to deal with
zeros in international trade. We hope the further research could acknowledge the
importance of zero trades and come to appreciate the need to adjust by using

appropriate econometric techniques.

24

Chapter 2: Explaining Export Varieties: the
Unexplored Role of Comparative Advantage

1. Introduction

The recent trade literature on export / import variety has grown rapidly.
The seminal work by Krugman (1979) first brought product variety into focus
through a. monopolistic competition model of international trade. Although the
i1‘1creases in product varieties have long been known as important source of gains
from trade. empirical studies on the significance of the growth of the product va-
rieties, or “extensive margin” of trade, in international trade are relatively new.
For example, Kehoe and Ruhl (2003) show that. the trade of new goods (exten-
sive margin) explains a larger proportion of the growth of trade following trade
liberalization than the increase in the volume of previously-traded goods (inten-
sive margin) does. A series of empirical studies by Funke and Ruhwedel (2001a,
2001b, 2005) indicates that the growth of product. variety in exports has a signifi-
cant effect. on the economic growth in various countries and regions. Feenstra and
Kee (2004) also provide evidence supporting the positive impact of export variety
on productivity growth for a large sample of developed and developing countries.
Broda and VVeinstein (2004) empirically show how much the increase in imported
variety mattered for the welfare of United States. Their results suggest that the
US. welfare has increased by 3% due to the increase in the extensive margin of
trade.

Although these previous studies have examined the cross-country pat—
terns of product varieties in international trade, few explored the trade patterns
of product varieties across countries. In this paper we examine whether the tra-
ditional theory of comparative advantage explains the cross-industry patterns of
product varieties in the exports of countries. Our approach also considers the mod-

ern framework of firm-level heterogeneity. We first construct. a theoretical model

in which countries vary in factor emlowment, industries differ in factor intensity,
and firms belonging to the same industry are heterogeneous in productivity. This
model is used to derive a prediction that. relates product varieties in a country’s
exports to the degree of relative factor intensity of industries. To empirically test
the prediction we en‘lploy the data 011 US. imports in 1990 from Feenstra, Romalis,
and Schott (2002), which finely classifies imported commodities according to the
10-digit Harmonization System (HS). we also use the data on input factor use
in various industries from 1992 US. Census of l\*Ianufactures, as well as the data
from Hall and Jones (1999) for factor abundance of countries. The empirical tests
support our semi-Heckscher—Ohlin prediction for product varieties in trade; that
is, countries export more varieties in the indutries that more intensively use their
abundant resources as input factors.

This paper contributes to the literature by extending the theoretical
model of Bernard, Bedding and Schott (2006), which integrates a heterogeneous
firm model by l\-'Ielitz (2003) into the 2-country, 2—factor and 2-sector framework, to
a multi—industry setting as Dornbusch, Fischer and Samuelson (1980) and Romalis
(2004). The paper also goes empirically further than others by explicitly linking
the factor endowment and industry-wise factor use to the number of varieties in
their exports.

The paper proceeds as follows. Section 2 develops the theoretic model in
order to provide an implication for the relationship between factor proportions and
export variety. Section 3 proposes an empirical approach to test the theoretical
prediction, and Section 4 describes the data. The results of the empirical tests are

also presented in Section 4. Section 5 concludes.

2. The Model
This paper adopts the monopolistic competition model with CBS function

and a. fixed cost of exporting to account for market entry. The model features a

26

framework where countries differ in endowment and increase in the exposure to
trade leads to inter-firm and inter-industry reallocation toward more productive
firms and industries with more comparative advantage.

We consider a world of two countries, two factors, multiple industries and a
continuum of heterogeneous firms. Countries share the same technologies but differ
in terms of factor endowments. \Ve use H for home country that is skill-abundant
and F for foreign country that. is skill-scarce. so that I— > IT” where S stands

- . , H
for skilled-labor and U for unskilled-labor.
2.1 Consumption

The representative consumer derives his utility from consumption over the
output. from all M industries. And each industry consists of a large number of
differentiated varieties with each variety indexed by one firm. In what follows, we
use 7? to index firms. and m for industries. The utility function takes the following

form:

N
0: (JV (1'
U = (71102 2....Cfi” 2 am 2 1 (2.1)
m=1

C m represents the consumption index over industry m, which produces a set

of 9m individual varieties, with quantity of each variety as (1(1)

1
em, =[ / qmpdw (2.2)

i697”
Accordingly, the price index Pm over individual varieties is defined as
1—0 T—l -
Pm = [ / 1),”(1) dz] ‘0 (2.3)

’16 Qm

27

where 0 : l-1——p is the constant elasticity of substitution across varieties.

2.2 Production

As l\-Ielitz(2003)‘s model, we model the production cost as a combination of
two parts, fixed cost and variable cost. The fixed costs are the same for all firms in
the same industry within a country. however. the. variable. costs vary across firms
with their productivities ,9 E (0. 3c). Therefore. we assume the cost function for

firm i of industry In is:

(I
9")(m, i)

3712,11,:‘(31713 1 > ,3,” > fill—1“" > 0 (2.4)

Total cosf(m. i) = f + si

where s is the wage for skilled labor, w is the wage for unskilled labor, A = H , F
is the country index. By ranking the skill intensity of industries(,x3,,,), we assume
the industries are more skilled-labor intensive as 77). gets larger. The fixed cost and
variable cost of each firm features the same factor intensity. There is no difference
in factor proportion use within the same industry across countries.

For domestic production, the profit maximization rule implies that equilibrium

price for domestic sale will equal to a constant mark-up over marginal cost:

I" _ "
git-3m, 1 (3m

. ‘/\ w :-
1’m.i(‘f9) : [)(b (2.0)

 

With the pricing rule. the firms“ equilibrium domestic revenue is in the fol-
lowing form:

.5377) . 1713171.

.‘a ll,}

7‘/\'II),,17(""") Z ("-iyAf i

 

1-0‘
, ) (2-6)
p‘i’PA'I'n

YA is the total income of country A. Each firms revenue is ii'icreasing pro-

portionately with its idiosyncratic productivity of). the domestic aggregate income

[\3
30

1, the industrv price index P,” and the inverse measure of mark-up The profit
of each firm would equal the revenue minus fixed cost of production and variable

cost:

Tm. 1'__(_C5 )

, 3 —3.
"m. 1(9) 2 0 _3f)\1 ml 1 111

There is a sunk entry cost for each firm in order to draw a productivity param-
eter from distribution C(91). which is a Pareto distribution with shape parameter k
with the support. of [1,511. (1le- The entry cost takes similar form as fixed production

cost. which used the same proportion of the two factors:

#833111, (,1 13711. (2.8)

In sunnnary. there is common factor intensity for fixed, variable and entry
cost. In the equilibrium, the firms keep entering until the profit is zero, therefore
the benclunark productivity for domestic production is determined by zero-profit

productivity cutoff:

1 _ .13771

, 1’3,
7"111 (0:111) : afsAm “iA (2-9)

In country A, all the firms with productivity higher than or equal to (1'): will
continue producing while firms with (,1) lower than the benchmark productivity will

exit.

The value of the firm is also determined 011 a discount basis of future profit.

flow, with (5 as the probability of death, we have:

29

x .

- — Ci)

1),,(0) :max{0,Z(1— o)7r,,,(a)} 2 max{0, 17%)} (2.10)
i=0

In the long run equilibrium, the expected value V,” should equal the sunk
cost in each industry. The expected future value of entry is the er ante expected
probability of successful entry multiplied by expected profit, therefore we get the

free—e111‘ry condition:

[1 — C(Oimfl—Li—m— : fc.s:\37”‘111/1\_dm (2.11)

where f,,,(c')) stands for the average or expected profit. from successful entry
for industry 111.
By combining the free entry condition and the zero profit condition, we could

simply express the function of (5):,” in one equation:

f 0C
, (f)
g [( ("1*

‘ Am

, *
oAm.

 

)‘7—1 - 1l9(6‘>)de = f(.) (2.12)

where 9(0) is the probability density function of Pareto distribution. The left
hand side of equation (2.12) monotonically decreases with the increase in the value

of o: with the right hand side of equation (2.12) being constant. therefore a

"I
o v * o ‘ ‘ ‘
umque \alue of 0A,," 1s guaranteed.

For export, we assume the export cost. includes both the fixed and variable
parts. In order to export a. manufacturing variety to another country, a. fixed
export cost must be spent for each firm, and it uses the same factor intensity as
production cost. Additionally. the variable trade cost takes the standard iceberg
form with parameter 7' > 1. this term is synnnetric across industries and cmmtries.

\N’ith the export costs, the equilibrium price is still a constant 111ark—up over

30

the marginal cost,

 

A , A , Tsi'nzui—d'nr ‘ ‘
p111.1.r(o) E Tun-.110) : p65 (2'13)

For firms from home. depends 011 their productivity levels, their profits will

be domestic profit only or both domestic profit and export. profit:

ff}; (0) = 7"{),(c$) for domestic production only
= 1%,(96) + -7‘,);,j;,_1(cb) for both domestic production and export

To solve the cutoff points of export as well, we need similar equations for
zero-profit condition and free entry condition.

However, the expected profit now consists of two parts:

A

7Tm .1totul

(1,1)) = figmm) +1113X1017TA1..11(¢)} (2.14)

TI

The zero profit condition consists of two equations:

Zero-profit domestic production condition, which involves 05:7”
., ,8 1— ’3
1~.,1,,(1:,,,) = 0 101911, ’ "1 (2.15)

Zero-profit. export condition, which involves (1‘):an

,1’3711. 1‘ #3111

1111110153,”) = 0.115, 1, (2.16)

Equations (2.6), (2.15) and (2.16) jointly determine the relationship between

cutoffs (1):," and (937,", which is

1 ——k , —k H .
99*Hm. 2 0:71111' A7" (2'17)

31

—A _ —k F
qF'm QHmr A"? (2'18)

where A51, 2 [Tl—073(%%)0];—k1, A7}; = [Tl—072(fi—g)0]m,A/\ =
5:3,” “ifljm

Assumption 1: fr > f

With this assumption, there is a selection effect into export market, which
means only a portion of firm with successful entry could export. Of all the firms
from the borne country, a fraction of C(01).) will exit because their revenues could
not (mer the fixed production cost, a fraction of C(QAmr) — C(éAm) firms will
serve market A only but could not be able to cover the higher export cost. Only
the most, productive ones, those who could draw high enough productivity will
export.

The free entry condition is also modified because now the expected value of
entry is the sum of two parts: the ex ante probability of successful entry times
expected profit of domestic production and ex ante probability of export times
expected profit from the export market. The expected value should be required to

equal to the sunk entry cost, which means:

 

_,\( , . ,
[ . . 71ml .. 71111 riff) 1’35— 1713-111.

1 _ (:193A111)1+11 _ G()An1.r)1—6— = fe'siymu’)‘ (2°19)

By using the two zero profit conditions and the specification of Pareto distri-

bution condition, we can write free entry condition in the following way:

A .
f(Ix —1)15L-A.,,,15F1 +f1( 1 ‘1)“1011111 — f16(2.20)

7721'

32

f(A' — 1)15LA,,,15 1+ f,(I \ —1)15’;I1.5F’* — _f,5 (2.21)

771.17

where [1— — k —+—011‘_ .The existence and uniqueness of {éHmr‘ ‘¢F111.r} is de-
rived by linking these two equations. In such a model with multiple industries, two
factors and heterogeneous firms, opening from autarky state to costly trade would
have different i111pacts on asymmetric countries and on different. industries. As a
result of asymmetry between two countries" factor intensities, the profits derived
from exporting also vary across different industries in different countries.

With other things being equal, we could compare the export varieties (number
of exporting firms or probability of exporting) between countries and industries.

The ratio of countries" export varieties in the same industry m is given by:

 

5 , .f..55(AF — f1')
1 _ G(¢5*H111.r) Z ( 1:L )A : ( m 7 f2 (2‘22)
OW r (1“ 1)(Af11A111Hf— 7E)

The te1m 1— G ( on”) not only stands for the share of firms that are involved in
exporting activities, but also represents the export variety of country H in industry
because each firm accounts for one variety.

It kil—U A (T

Assumption 2: ’7' ('91:) <7— > (:,IE)(7—I

H

This requires both variable and fixed trade cost are fairly large.

Under autarky, home country‘s relative skill abundance leads to a lower rela-
tive price of skilled labor and of the skill—intensive good. Under free trade without
trade cost, both goods and factors are mobile. As Samuelson (1949), Dixit and
Norman (1980) showed, we allow two countries‘ relative endowments of the two
factors lie in l_)etween of the integrated equilibrium factor intensity, thus the equi—
librium of free trade is characterized by F PE, which means the equilibrium wage
equals the value for the integrated world economy. However, the existence of fixed

and variable cost of trade in our model results in an intermediate relative wage.

33

From autarky to opening to trade. it leads to a decrease in the rewards for the
abundant factor. thus the unskilled relative wage 111A will fall in skill-abundance
country (Home) and the skill relative wage s/\will decrease in skill-scarce country

(Foreign).

where the superscripts A. CT. FT indicate autarky. costly trade and free
trade.

Proposition 1: For fixed country pairs, the cross-industry trade has the fol-
lowing patterns:

(a) the skill-abundant. countries will export more varieties in skilled-labor i11-
tensive industries ( bigger 13 )

(b) the skill-scarce countries will export more varieties in 1.111skilled-labor in-
tensive industries ( smaller 13 )

Proof: see appendix B

111 appendix B, we show that the absolute term of industry-lwel export variety
of home country increases with skill intensity of the industry. The reason why home
country could produce and export more varieties in skill intensive industries stems
from the fact that home country has more skilled labor. It could either use skilled
labor more intensively for each industry or have more skill-intensive varieties. Our
model assumes there is no technology difference within industries across countries;
therefore. home country capturing a larger share in terms of exported varieties in
skill intensive industries seems to be a natural result. The extensive margin of the
trade structure is relative factor abundance among countries.

Proposition 2: For fixed industries. the cross-country trade has the following

patterns. when relative factor price is controlled, the country that has cheaper cost,

34

in absolute terms export more varieties.

Proof: See appendix B

There is a bigger mass of firms in the skill—intensive industries in the skill-
abundant country (home country), and unskilled labor intensive industries in the
skill-scarce country (foreign country).

The asynnnetry of the proportion of exporting firms across industries stems
from the fact that comparative advantage industry features lower relative price.
Therefore, a more fierce competition exists in the comparative advantage industry.
but less fierce competition in the export market for comparative advantage indus-
try. This fact will result. in more firms entering comparative advantage industry,

and thus more export varieties in this industry.

3. The Data

Our empirical, framework proposed in the previous section requires data for
three variables: the number of product varieties in industries in the exports of
countries, production factor endowments in those countries, and input factor i11-
tensities in the industries.

For the product varieties in exports, we use the data on the US. imports
in 1990 from Feenstra, Romalis and Schott (2002). The data contain the infor-
mation on the US. imports of each commodities classified according to the very
disaggregated 10-digit Harmonized System (HS) from each exporting country. The
data also indicate product classification code according to the 4-digit U.S. Stan-
dard Industrial Classification (SIC, 1987 version) corresponding to each lO-digit
HS. This enables us to count the number of product varieties in each industry in
the f(_)llowi11g manner. Defining industries following the 4—Cligit SIC and product
varieties following the IO-digit HS, we measure the number of varieties in Industry
«i exported from Country 6, or 71 in, by the number of 10-digit HS commodities

included in the US. imports from Country c in each 4-digit SIC industry. Since

some 4-digit industries have more 10-digit varieties than others by nature, we ad-
just the number of varieties by the total number of 10-digit varieties that US.
imports from the world in each 4-digit industry; i.e., N,- = :"ic- Note that the
imports of the same 10—digit commodity from different coufitries are considered
as different varieties, as the theoretical model assumes that each firm produces
a unique product. Table 12 provides the number of exporters, total number of
varieties, and total import value in the US. imports, as well as those in the US.
manufacturing imports (imports in the 4—digit SIC 2011 through 3999) in the year
of 1990. Due to the availability of industry factor intensity data, we use the data
on manufacturing imports, which represent 94% of the total US. imports in 1990
in the number of varieties, and 83% in the value of imports.

The data. for factor endowment of countries are from Hall and Jones
(1999). Our theoretical model is in a two—factor framework with skilled labor (S)
and unskilled labor (U) and we use the data on the ratio of human capital to labor
as the measure of the abundance of skilled labor relative to unskilled labor (S / U )
In the source the data on human capital to labor ratio as of 1988 are available
for 127 countries. Since we consider the US. imports from other countries, we
calculate the exporters’ skilled~to-unskilled labor ratios relative to the ratio of the
US. (1' .e., M for each exporter r).

(5/ U )US

Our theoretical model assumes common production technologies across
countries, and we employ the data from the 1992 US. Census of Manufactures,
which covers 458 manufacturing industries classified by the 4-digit SIC (1987 ver-
sion; the codes 2011 through 3999) as the measure of the world common input
factor intensity in each industry. We measure the skilled-labor intensity of each
industry by the number of non-production workers as the share in the total number
of employees in each 4-digit SIC; and the unskilled-labor intensity by the number

of production workers as the share in the total en'iployi'nent.

36

The sample for our einlnrical analysis includes 115 countries, from which
the US. imports in any one or more manufacturing industries in 1990; and 394
111anufacturiug industries (4-digit SIC). in which the US. imports from any one or
more exporters in 1990. Table 13 lists these 115 countries in the sample; and Table
14 1*)1‘ovides the summary statistics of relative factor endowment (the skilled labor—
to-unskilled labor ratio. or S/ U) of the sample countries, as well as the lists of
ten most and least skilled labor-abundant countries. Table 15 shows the summary
statistics of the intensities of the two factors (S and U) of 3.9—1 sample industries,
and also lists ten most, and least. skilled labor-intensive industries. Figure 4 graphs
the number of countries from which the US. imports in each 4-digit industry in
1990. 011 which the industries are sorted (from the left to right) in the order of
skilled-labor intensity. Figure 5 and 6 plot the number of exporters and the total
number of varieties that the US. imports in each industry, respectively. against
industry skilled-labor intensity. These figures indicate that the U.S., the world
second most skilled-labor abundant. countries, tends to import. from more exporters,
and accordingly import. more varieties. in unskilled lalmr—intensive industries than

do in skilled-labor intensive industries.

4. Empirical Test
The key implication of our economic model presented in the second sec-

tion is that, as indicated by Equation (2.22), with the assumption that each firm
produces a unique variety of product, a country will export more varieties in indus-
tries in which the country has its comparative advantage in the HeekschenOhlin
sense than it will in other industries. In this section we empirically test this i111-
plication using the data described in the previous section.

4.1 Measuring Exported Varieties

Our model is to explain the munher of product varieties that each coun-

 

try exports to a common importer the US. in our empirical analysis—in each

37

industry by two elements: the exporter’s relative resource abundance and the i11-
dustry's relative factor use or intensity. As described in the previous section, we
define varieties by the Ill-digit HS connncxlities and industries by the 4-digit SIC,
and thus measure the number of a country’s exporting varieties in an industry by
the number of the 10-digit HS commodities that the country exports to the US.
in that 4-digit SIC. as follows:
"17c E No. of 10-digit HS commodities in a 4—(ligit SIC 1' exported by
country c
However. some 4-digit industries may contain by nature more 10-digit varieties
in its catalogue than others, and thus the US. may import more 10-digit varieties in.
those industries than in other industries. For a proper cross-industry comparison,
we use an adjusted measure of the number of varieties. which is constructed as

follows”:

72,}.

4' I

71_slz.a'reljc =

where N1 is the total number of varieties that US imports from the world in

' . . y .'. f ,7 _ . .
industry 2. A, — 2671“.

4.2 Regressions for Aggregate North and South:

\Ve first test our two-country. two—factor and nmlti-industry model with
the data for cmmtry aggregates. We divide our 115 sample countries into two
groups and construct two country aggregates, one of which consists of countries
that are relatively more skilled-labor abundant to unskilled (or with relatively high
S/U). and the other consists of countries that are relatively more unskilled-labor
abundant (or with low S / U). We call the former country group “North” and the
latter “South.” For North we select 51 countries with S/ U relative to the US.

above its sample mean, and other 64 countries for South“). Table 16 lists the

38

countries constructing the aggregates North and South. Table 17 con‘ipares the
within-group averages of relative factor abundance S / U .
The following equation is estimated by the OLS regression for North and

South”:

[Of/(n-9ll-(Il'ei‘A) = a + ,3 * skill, + 5, (2.23)

where 71.13/1a'rcljs A = ZeEA 1'1._slmxre,j. A for A =South, North, and skill, =skill
intensity of industry 17.18Our model suggests that relatively skilled-labor abun—
dant North exports more varieties in skilled-intensive industries than in unskilled-
intensive industries. and unskilled—abundant South exports more varieties in unskilled-
intensive industries; thus the expected sign of ,3 is positive for North and negative
for South. The estimation results are in fact consistent with this prediction, as

shown in Table 18.

4.3 Pooled Regression for Dependent Parameter Specification:
We. next use the pooled data for all the individual exporters to estimate
cross-industry patterns of the varieties in exports. We consider the following re—

gression model:

Iog(mshareic) = a + AC * skill, + 5-ic (2.24)

The slope coefficient for skilled—labor intensity. AC, would differ across exporter
countries. The tl'1eory predicts that the value of the slope coefficient is higher for
countries with greater relative skilled-labor endowment. and lower for exporters
with smaller relative skilled-labor endowment (or greater relative unskilled—labor
endowment). To capture this pattern, we impose the following structure on the

slope coefficient Ac:

39

[\(f : l/\((S//L’Y)(j) 2,1131 + {32 * (‘S/LY)( (2.25)

where (S/U)C is the skilled- to unskilled-labor ratio in exporter c relative to
the US. The theoretical prediction is that the sign of 1'31 will be negative (since
A, will be negative for countries with low skilled-labor abundance) and ,‘32 will be
positive (since AC will increase to be positive for countries with high skilled-labor
abundance). By substituting (2.24) into (2.25). we derive a specification for our

pooled regression as follows:

log(n5._slm.re,-(_.) 2,131 * skill, +132 >1< skilll- * (S/U), + [1,, + Sic (2.26)

\Ve include exporter-specific dummies. NC, to capture the effects of all other
factors than the relative skilled-unskilled abundance that differ across countries,
such as fixed and variable trade costs and the size of the exporterlg.

The result of the estimation of Equation (2.26) by the fixed-effect OLS is shown
in Table 19. The estimates of all the coefficients show the signs as expected from
the theory, and they are. all highly significant. In addition, using these estimates we
compute the "threshold" factor abundant (S / U)* that makes the slope coefficient
for skilled intensity Ac turn from negative to positive (i.e., A(:((S/U)*) = 0). The
value of the “threshold” S / U (relative to the US.) is 0.6620.

These results of the empirical tests suggest that the sen1i—Heckscher—Ohlin
prediction of our economic model on the exported varieties is supported by the data
on the US. imports.

5. Conclusion
I11 this paper. we have investigated the relationship between export. variety

and the exporter‘s comparative advantage in terms of relative resource abundance.

40

\Ve generalized the model by Bernard, Redding 85. Schott (2006) to the case with
continuum industries and derived a prediction that relates a country’s export va-
rieties in a certain industry to the industry’s “degree” of relative factor intensity.

To test the prediction we have e1111ployed the disaggregated data on the. US.
imports. as well as the data. 011 countries’ human capital and labor endowments
from Hall 8: Jones and those 011 the industry-wise uses of skilled- and unskilled—
labor from the US. Census of l\/Ianufactures. The empirical tests support our
semi-Heckscher-Ohlin prediction. which shows that more unskilled-labor abundant
exporter tend to export more varieties of products in relatively unskilled labor-
intensive industries, and more skilled-abundant. exporters tend to export more

varieties in relatively skill-intensive industries.

41

Chapter 3: Globalized but Unhealthy? Feed-
back on Population Health from FDI

1. Introduction:

There has been numerous theoretical and empirical studies that underscore
the positive effects of Foreign Direct Investment (FDI) for economic growth. It
has been established that the economic benefits of F DI come in several channels.
First and foremost. FDI allows the transfer of the technology, which takes the form
of new varieties of capital input that cannot be achieved by trade in goods and
services (Feldstein 2000). Recipients of FDI also gain job training which helps to
develop the human capital in the host countries. Additionally, profits generated
by FDI contribute to corporate tax revenues in the target countries.

Since there is such widespread belief that FDI is beneficial to the host coun—
tries, F DI has become the pre-eminent source of capital flows into the developing
countries, and these governments have implemented a handful initiatives to attract
the foreign investment, such as tax incentives. As a result, net inflows of FDI in
the group of developing countries have increased almost five fold from an average
of 0.44% of GNP during the period 1970-1974 to 2.18% of GNP during the period
1993-1997.

Besides all the claims that FDI is a good thing. the hypothesis that F DI would
also improve the total living standard outside of the scope of economic growth
seems more controversial. Some critics demonstrate that F DI might increase wage
inequality (Driffield 85 Taylor, 2000), generate environmental degradation, increase
target countries’ exposure to international financial crises and deteriorate the prob-
lem of education inequality.

In this paper, we explore whether (and how) FDI affects the population health.
The population health is an important factor in achieving the. long-run economic

growth. Even so, the relationship between FDI and population health has received

42

rather scarce attention, both theoretically and empirically. What makes the em-
pirical test difficult is that a simple correlation study on FDI and health does not
help to reveal the true casual effect of FDI. The endogeneity of FDI usually causes
the identification problem that is hard to be addressed. The level of the govern-
ment effectiveness. quality of social infrastructure or international aid programs
that may attract (or discourage) more inward FDI would also influence the popu—
lation health. For example, countries with less stringent policies 011 the epidemical
diseases, such as HIV/ AIDS, malaria and tuberculosis would usually bear the con-
sequence of lower levels of human capital, labor productivity and likely higher
costs of operations to cope with the health-related expenditures, therefore, less
inward FDI would be a natural result. So the key question in our study would be
how we can control the endogeneity and ideally, 110w we can find the exogenously
calculated FDI level, then look at the effect of the variation of F DI 011 population
health.

Given the backdrop, this paper is the first of its kind to examine the link
between FDI and population health in a cross-country framework and untangle the
problem of endogeneity. Levine and Rothman (2006) looked at a similar question
whether openness of trade affects child health. They found that overall the trade
does little harm. Edmonds and Pavcnik (2006) examined the trade effect 011 child
labor and found that the openness elasticity of child labor is much smaller (-0.1)
and statistically insignificant. Both papers explore the import-export effect on a
country development indicator. A more related paper to ours is authored by Alsan,
Bloom and Canning (2006). who investigated whether population health affects
foreign direct investment inflows while we look at the causality in the reverse
direction. Their panel-data analysis on 74 countries during the period of 1980-
2000 showed that health has a positive and significant efiect 011 FDI inflows for

low— and middle—incmne countries. Though the finding is consistent with the view

43

that health is an essential component of human capital in developing countries,
their empirical testing more or less suffers from the identification problem.

Our empirical investigation on the FDI and health data suggests a negative
impact of F DI on the health status, however, the association is statistically in-
significant. Thus the cross-country data do not substantiate the assertions that
F DI per se plays a significant. role in perpetuating the low levels of health sta-
tus that pervade in low-income countries. Country-level inward and outward FDI
are from published and unpublished data prepared by UNCTAD. The merit of
UNCTAD data is that it is not only collected directly from the member countries,
but also is complemented by international organizations, such as IMF and world
bank. The population health indicators, including the measures of life expectancy,
mortality rates are from World Development Indicator Database. And the gravity
covariates are compiled from different sources.

The endogeneity problem is addressed with a semi—IV approach. We use the
framework proposed by Frankel and Romer (1999) and examine the relationship
between population health and FDI based on countries’ geographical character-
istics21. The identification depends on the assumption that geographical factors
do not. affect population health through other channels besides FDI. Then the
constructed measure of FDI volume is used to analyze the causality between glob-
alization and population health.

The rest of the paper is organized as follows. Section 2 presents a health
stock model and further discusses 110w FDI might affect the health outcome in
theory. Section 3 presents the econometric models to identify the effect and the
data used in the paper. Section 4 contains the result of empirical analysis. Section

5 concludes and provides discussions on further extension.

2. How does FDI affect health:

44

2.1 Theoretical Review:

There are several arguments put forth for exploring the link between F DI and
health. We seek to provide a comprehensive and systematic review of the evidence
concerning FDI and health service. We categorize the mechanisms through which
F DI affects the health condition into five groups: (1) the income effect; (2) gov-
ernment policy effect; (3) labor market effect; (4) effects through joint products of
economic activities and (5) direct result from higher FDI volume in medical sector.

Firstly, a higher level of F DI could overall increase the income level and eco-
nomic growth, which would improve health (Pritchett & Summers,1996). It may
come from several sources: improved nutrition (Fogel, 1994); improved access to
health care or higher government investment in public health funded by the tax rev-
enue. A larger income would also allow more extensive medical care: a frequency
of routine checkups, doctor visits and hospital episodes. However, transitory eco-
nomic growth is usually coupled with more intensive input of labor and health. As
Gustmann and \Nindmeijer (2004) showed, there is improved health status with
long-term higher income level but worse status with temporary wage raise for the
Germany case. Better macroeconomic states might be coupled with reductions
in the risky behaviors such as smoking (Ruhm, 2003) but in the meanwhile in-
creases obesity-inducing behavior and alcohol use (Ruhm, 2000; Freeman, 1999).
Therefore the results are rather mixed in this setting.

Secondly, a higher F DI may induce the government to provide a better safety
net, better regulations, and as a result, the better infrastructure would have a
positive impact on the health outcome. The data shows that there is a keen com-
petition among developed and developing countries to attract F DI for its many
economic benefits. This drive to lure investment makes different potential target.
countries to pursue their own strategies and assemble their own baskets of incen-

tives to attract a larger inward FDI. Various reforms and strategies have been

ili’1plemented. The improved general business environment, more skilled human
capital and the infrastructure of the higher quality, which would have been 011 a
smaller scale if it were not. for the incentive of attracting more FDI, all link to a
better p(‘)pulation health condition.

Thirdly, FDI can be linked to health through the labor market dynamics. The
link between FDI and labor market outcome (wage inequality and employment,
rate) has been previously explored. The motivation of FDI-either as horizontal for
111arket access, or as vertical to exploit the lower input cost in target developing
countries would yield different results for labor market. Though it has been shown
by various authors that heterogeneity exists across countries and over time for the
determinants of FDI. we contend the volatility of the workers’ wage, employment
status and the redistribution of wealth all have an impact on the population health.
Health is considered as an input of the production of goods and services. The
hazardous working conditions, the physical exertion and job-loss stress could all
have negative effects, but the condition could also deteriorate when job hunting
period is lengthened when market expands (Karasek & Theorell, 1990; Sokejima
& Kagamimori, 1998: Liu et al., 2002). Some sectors that boast higher intensity of
inward FDI, such as mining and construction also have higher accident rates as the
hire of inexperienced workers increases because of the lengthened working hours
to boost. the production (Catalano, 1979; Robinson, 1988). In the meanwhile, the
change of lifestyle during laid-off period also has impacts on the health status from
a (.lifferent channel. Evidence shows that there is reduction in alcohol use linked
with economic downturns (O’Neill, 1984; Evans 8; Graham, 1988). Severe obesity,
smoking. physical inactivity. consumption of fat decline as well. For a popular and
widely used health indicator, mortality, a. number of studies find a one percentage
point increase in the unen1pl(‘)yment rate. is typically associated with a 0.3 to 0.5

percent reduction in total mortality (.Iohansson, 2003; Gerdthanm & Ruhm. 2004;

46

Tapia Granados, 2004b). And infant and neonatal mortality in US declines by
0.6 percent when unemployment rises by one point (Ruhm, 2000). The change of
non-market "leisure“ time, gives people more or less health—producing activities,
such as exercise. To sum up, the evidence linking working hours and health is also
mixed (.Iohansson, 2003; Ruhm, 2004a).

Fourthly, FDI influences some joint products of economic activities, such as
pollution, also these present risks to health condition (Chay «$5 Greenstone, 2003).
The unpacts of these may be particularly pronounced for different. strata of the
population —workers who are. involved directly in certain sectors, such as mining, or
v1.1lnerable segments of the population. such as infants and seniors. The openness
in FDI induces a more fierce competition among firms, which may lead to a “race
to the bottom” that increases pollution and reduces govermnent expenditure for
investment in health.

The last momentum is from General Agreement on Trade in Services (GATS),
which aims to liberalize the service sector to a greater extent, and FDI is considered
as the most critical area for openness negotiation. GATS specifies “four modes
of supply" : cross-border supply. consumption abroad, commercial presence and
temporary movement of service providers. Among the four modes of supply, mode
3 (commercial presence) is considered as the most critical as an influence of FDI
011 the health sector, the benefit. being especially big where investment leads to
gains in basic, with additional resources and expertise improving the range, quality
and efficiency of the service offered (Chanda, 2001; Zhang 85 Felmingham, 2002).
For example, FDI could occur in building hospital facilities or be featured by
more medical professionals, in such cases the increased FDI would present more
opportunities and be linked with better health outcome.

2.2 A Simple Model:

To motivate the empirical specificaticm, we use a modified version of the in-

47

vestment model of health with FDI built in it. Health is considered as endogenous
and depends on both the time allocated to it and the medical care. And the
amount of these inputs is determined by the time and budget constraint.

We denote the representative individual (both agent and principle)"s utility
by U as a function of conteinporaneous consumption ct and leisure time It at given

time period t:

Ut = (.11 Iog(ct) + (12 Iog(lf) (3.1)

The production of health consists of time allocated to it (ht) and medical

care expenditure (771,). The Cobb-Douglas function would thus be:

Qt 2 Ct +131 Iog(ht) + ,8210g('mt) (3.2)

Ct is considered as exogenous human capital. With a depreciation rate (5, the

dynamics of the health stock could be described as:

D1 2 Ct +131 log(ltt) + ,1'3210g(mt(ff)) — (SDt (3.3)

The total medical care expenditure is funded by both the domestic government
and F D1 in medical health sector (the GATS channel) therefore is a function of the
current period FDI (ft). The gain from a higher health investment is represented
as a longer life duration, so that more future wage and leisure time could be
enjoyed. Death takes place when Dt is smaller than Dmin- Thus the duration of
life demuds on Dt. W'e normalize the length of each time period to 1, and wage
rate is ”ll-‘(ff). a function of FDI, which could either come from the income channel
or labor market channel for the reasons we previously argued. The time spent on
working is 1 — ht — lt for period t, assuming away the possibility of tax, the total

consumption can therefore be expressed as:

48

Ct. ='U’t(ft)(1-ht —lt) -Ptmt(ft) (3-4)

which presents a trade—off between time for health and time for work. The

dynamics of the model could be described as the current Hamiltonian H:

H = 011<>s(U‘t(ft)(1 — ht — It) — Pt'l'l'ldftll + 0'2 10120-1le
/\1(Ct +131 logfht) + 5321080711) — 5191.) (3-5)

The levels of control variables 11,, It and ft are determined by F DC as follows:

ht : Atlflfu’tfftlfl — ht — lt) — Ptmtfftll

 

 

a"1'U«’t(f1) (3'6)
_ a’2f“’t(ftl(1—ht‘ltl—Ptllltfftll
It — a"1U-‘t(f1) (3'7)
(1'1 I I

 

(111t(ft)(1—ht—lt)—ptmflft»["l'tfu’tfftlfl"h‘t — lt) _ Pt‘771¢(ft)) — ptmt]+

I
/\ 3 mt — 0 (3 8)
ti 2mt _ '
12,; 2 Ct +131 log(lzt) + ,13210g(mt(ft)) — 6Dt (3 9)
A, = (5 + p)A, (3.10)

p is the discount rate, A, is the Lagrange multiplier for the shadow value of

49

/ l . . .
health. In, and “’t are the first order condltlon of mt and wt w1th respect to ft
individually.
The system of the equations characterizes the change path of the health stock.

By collecting terms, the total health investment can be expressed by:

At1'31(u»‘t(ft) - Ptmtfftll +
“’t(ft)()‘t‘31 + 0'2 + a1)

 

Investment, 2 Q + 1'31 log(

0200101) — Ptmtfftl)
l"t(ft)()‘t1’31 + 0'2 + 0'1)

 

bet(1-210g( (3.11)

where the optimal level of ft is determined by equations of (3.6) to (3.8).

In order to determine the impact of F DI on health, we need to further look

. BInvestment . . . .
at the marginal effect 8ft t, and its Sign 18 determmed by the product

I I
513 + 0 ,—_],)t..— m w, — 111 m . The sign is undetermined because of the
l 1 1 at I’tmt t t t t -

 

uncertainty of the part (771..th — 1121:7711). When FDI allows higher foreign—aided
medical care expenditure without crowding out domestic share, 772.; is expected to
be positive. The feedback from FDI 011 the labor market is quite mixed as we
argued in the previous section. The impact. from joint economic activities, such
as pollution, comes into effect through the term At. in that higher At means a
faster depreciation of the health investment stock. Therefore the impact of FDI
on population health is ultimately an empirical question for us to explore.

3. Data and Specifications:

3.1 Data:

For empirical investigation, we focus 011 two basic measures of health: life
expectancy and probability to die (mortality), both measured in aggregate and by
gender, age range.

Life ex1’1ectaney is the average number of years that a person can expect to

live if they experience the current mortality rate of the population at each age.

Adult mortality risk is defined as the probability of dying between 15 and 59 years
old per 1000 population. Under-five mortality rate is defined as the probability
of a. child born in a specific year or period dying before reaching the age of five,
subject to age—specific mortality rates of that period.

FDI is becoming increasingly important in the global economy. According to
UN CTAD (2000) statistics, FDI from developed countries to developing countries
grew from 36 billion US dollars in 1992 to 155 billion US dollars in 1999, a level
more than three times that of the official development aid. Several sources provide
data. on net and gross FDI, including IMF22, OECD publications23and UNCTAD
FDIStat database. We chose the data that is the closest to the empirical spec-
ification. We. consider a F DI infiow (gross FDI) would influence target country
population health more directly than net FDI flow. The F DI data24and affiliate
enterprises are compiled from the unpublished data prepared by UNCTAD. The
data from 2000-2004 are used and a total of eighty countries have positive inward
FDI volume for any year during that period. The total FDI is the sum of three
components: equity, reinvested earning and other capital (intra-company loans)
on a current. US dollar basis. Other gravity measures, including distance, com-
mon language, common border, landlocked dummies, sizes are from Rose (2003).
The measure of medical expenditure as a share of GDP is from World Health
Organization. Real GDP per capita terms are from Penn World Table 6.1.

3.2 Specification:

The empirical work is to understand whether and how the inward F D1 in-
fluences the population health condition. For the regression, we first explore the
cross—sectional data, and the following linear model is used:

For each country i, at given time period t,

F DI ,-

” +8.,j (3.12)

health-.1: 2 ,Bl + (32 * GDP
.2

 

where the 5% is the volume of foreign direct investment in country i
weighted by the GDP, which could be considered as a FDI-induced index of open-
ness. The reason why we use FDI—GDP ratio instead of FDI volume is the large
variations in GDPs across country incomes. A GDP-weighted FDI measure elim-
inates the bias induced by heterogeneity in income levels and represents better
the country-level globalization degree. The dependent variable is from the series
of the health indicators including life expectancy and mortality (both by sex and
in total). Thus :32 would have an interpretation of the change of health measure
associated with the increase in the ratio of F D1 to total GDP. However this coef—
ficient should not be simply treated as the change of health measure induced by
the change in the FDI openness index because we have not dealt with endogeneity
issue yet.

Another factor we need to incorporate is the channel of per capita income level.
As we previously discussed in section 2, FDI influences the health condition through
the change of income. It is evident that a higher GDP per capita allows people
to spend more on medical care or consume hazardous products, such as cigarettes
and alcohol. To capture this effect, we also include the GDP per capital term
in the estimation of equation (1). We hypothesize that the relationship between
the income level and health might exhibit a non-linear property. Similar to the
(mvironmental Kuznets curve (Grossman and Krueger, 1993, 1995), higher income
level might be bad for population health because it, provides higher incentive for
working extra hours and worsens the health status, but later on it would benefit.
health, as countries become rich enough to pay more attention to the health quality.

Therefore we further estimate the following:

. _ FDI- . .
health, = {31+62*ln(5—D—Pz—_)+;33*111(GDPPC)+,134*(111(GDPPC))2+€Z- (3.13)
'I,

The term GDPPC stands for the GDP per capita. [32 therefore tells us the
degree of association between population health and FDI when controlling the
income level.

FDI enters each target country in different sectors. It is natural to expect that
FDI would enter different industries varying in the pollution extent in different
countries, therefore the impact of FDI would have different environmental, thus
health implications. Though it is impractical to fully capture all these impacts, we
would add more terms besides FDI term in the regression, as a practice, we further

include medical expenditure share of GDP for each country in the regression.

 

F DI: .
heal 1th,,- : ,81 + £32 * ln( G D PI ) + )33 * ln(medical_e.z:pe:ndit'u.7-eshare,-) + Si (3.14)
z

3.3 Control the Endogeneity and Include the Zeros:

In order to address the endogeneity issue, we construct measure of F DI-
induced openness based on geography, which is proposed by the original work
of Frankel and Homer (1999). In constructing these measures, we utilize a. pseudo
gravity model for FDI with covariates of geographical characteristics.

It yields the following regression:

1.,1FDIfl : [31 + {32 >1: lnsizq + {33 * lnsi:e-+

7

[34 * lll((ii.‘it(tll.(f(?.l‘j) + {35 * co-Imnon[(1712-3- + (36 * contiguity/U + ,‘37 >1: larl‘u'll()ck,-J- + E
(3.15)

The logarithm of inward FDI from country j to i is regressed on countries’

sizes, the distance between the importer and exporter, whether they share the

same language, same border and whether any of them is landlocked.

To predict FDI volume in this way has its own right. Ramondo (2006) used
simulated and actual data to justify the approach of estimating multinational pro-
duction function using geographical measures. The research shows that similar to
international trade flows, gravity governs the volume of F DI. Not only country sizes
matter, countries that are twice as distant face 56% higher cost than otherwise.

To determine the quality of instrument variables, we need to explore: whether
they are correlated with the actual FDI, whether they are not affected by popu-
lation health; and whether they are correlated with other factors that influence
health. As we just argued, FDI can be appropriately constructed by the gravity
measures and the empirical analysis also shows a high correlation between con-
structed measure and actual measure. On the other hand, it is hardly true that
countries’ geographical features could be influenced or influence the population
health. Therefore it is justifiable to say that the IVs affect the dependent variable
only through its effect on the endogenous variable of interest --—FDI. Different from
the conventional gravity model approach, we only use the country area sizes in-
stead of population or GDP to capture the size effect. We are concerned if GDP
or population is used, one might cast doubt on their suitableness as qualified in—
struments. The GDP itself could allow a higher proportion of income as medical
expenditure thus FDI is not its only channel to influence health. The population
measure is dubious as well, as population density and the initial stages of the
spread of disease are highly correlated (Tarwater, 1999).

Many countries actually witness a zero inward FDI. The FDI literature sug-
gests the underlying reason of the zero volumes is the high sunk cost of initiating
such F D1 in the target countries. Therefore we use a two-part model approach
to account for the existence of the zeros and thus we could include them into the

estimation. The choice of two-part model over sample—selection is justified by the

strand of the literature on a high entry barrier for FDI thus making it a corner-
solution case rather than potential missing value (l\rlullahy. 1997', VVooldridge,

2003).

4. Empirical Analysis:

The first step of the procedure is to estimate the fitted value of FDI based on
the geographical variables. First we use the positive FDI volumes only with the
specification of equation (3.15) and report the result in table 21. The the total
fitted value of FDI is calculated by exponentiating the predicted values for each
country pair then adding up for each target country. The correlation between the
fitted inward FDI volume and the actual measure for year 2002 is 0.61. In order
to account for the zero FDI volume, we invoke a two-part model approach so that
we can include all the possible country pairs and the result is reported in table 22.
A total of 170 countries are included in the large sample.

With the fitted values of F DI from both approaches, we could investigate the
causal effect of FDI on population health. The main regression results indicate
that after controlling the endogeneity problem, there is a significant and negative
correlation between population health and inward F DI. Tables 23 and 24 contain
the results of estimating equations 3.12 through 3.14 using the life expectancy as
the dependent variable. Table 23 presents the result of the regression of the level
variable of life expectancy on FDI measures while table 24 uses the logarithm of
life expectancy. Column 2 includes the result of the association between the actual
F DI / GDP ratio and health without controlling the endogeneity problem. A higher
FDI seems to link to a shorter life expectancy but the coefficient size is rather
smaller. Column 3 contains the result where the FDI measure is predicted from
gravity covariates. It can be argued that once the endogeneity issue is controlled,

the magnitude of the effect is even larger. A 1% increase in F DI/GDP ratio is

C}!
Cf!

associated with a 0.03 percent reduction in life expectancy. An interesting result
is that there exists gender difference in the degree of the impact of FDI on the
health condition. The general message is that male is more negatively influenced
by the globalization.

The analysis on mortality rate yields a similar result to life expectancy. The
higher FDI ratio induces higher probability to die for both adults and children
below five. A coefficient of 2.5 suggests that when FDI/ GDP increases by one

. . r
11111110112"

. the mortality rate would be increased by 2.5% for male infants dying
before five, and 8.5% for dying between 15 to 59 years old.

Additionally, we also notice that there is difference between using fitted values
from positive FDI volume only (78 country sample) and all country pairs (170
country sample). The negative impact. of FDI is relatively smaller in magnitude
when estimating FDI using all countries in the first step. Presumbly this is because
countries with actual zero FDI have a better health status than those with positive
FDI. But with an IV-aptn‘oach in the first step, there might be a positive estimated
FDI for these countries, thus offsetting the negative influence of FDI and making
it. smaller.

we also include the a log level of the GDP per capita term in the estimation,
positing that it might impact health in a non-linear way. Yet. the analysis shows
that the income impact on health is only significant and positive in a linear way, and
when income is controlled, we find no statistically significant association between
FDI and health. The drive behind this negative correlation might be that lower
income level countries are attractive targets therefore have higher FDI-indexed
openness.

Overall, the results from the regressions show a negative feedback from FDI
on population health. With the negative coefficients, one could go as far as the

implication that a higher FDI-induced globalization has a negative contribution,

but the coeflicients are hardly indicative of the importance of each mechanism

through which FDI influences health.

5. Conclusion

It is a quite a priori case to presume inward F DI benefits target. country in
many accounts. Since the later 19908, F DI has become especially welcome across
the world. in the developing countries in particular. Besides some other acclaimed
economic distortions caused by F DI, in this paper we look at. the linkage between
population health and F DI and address the endogeneity problem in a framework
proposed by Frankel and Romer (1999). Geographical vectors are used as IVs for
inward FDI. The cross-country evidence shows that there is an overall negative
association between population health and F DI. A 1% increase in FDI / GDP ratio
is associated with a 0.03 percent reduction in life expectancy. Once the income
difference is controlled, there is little evidence that FDI has impact on health.

The paper gives a systematic review on the mechanisms through which FDI
might affect health. And an investment stock of health model is used to motivate
the empirical specification. Though the view of positive economic benefit of FDI
has been established, the paper looks at the effect of the FDI on target countries
from another perspective.

The interaction of F DI and health has received considerable attention in the
literature, but the empirical evidence on this topic is scarce. The main problem
that limits the empirical investigation is the identification problem of FDI: the en-
dowment and policies that influence the FDI level also have impact on the health
outcome thus are difficult to fully control for in an empirical setting. In the re—
sults, although the coefficient on FDI is negative and statistically significant, its
confidence interval around zero is small in magnitude. Though the results suggest

a negative impact on health from F DI. we need to interpret the results cautiously.

The findings suggest that. a higher volume of inward F DI because of countries’ ge-
ographic features does harm to population health but its effect is not, substantive
when we control the income levels. It is not the foci of this paper to specifically
explore through which channel (and by what extent) FDI influences the health
condition though. However, it is definitely something worth investigating in the
future. Also, our paper does not concern the question whether better or worse
health conditions of the target country would lead to higher or lower levels of
the F DI. Also the evidence of minimal effect of FDI on health on average does
not imply that there are no circumstances that FDI will negatively impact health.
Identifying the atypical circumstances seems to be an importance avenue for future

research.

Footnote

1. Anderson (1979) was among the first to derive it from a theoretical trade
InodeL

2. Hummels and Levinsolm (1995) also poses new puzzles beside replicating
Helpman(1987)’s work and in the end, they found the empirical evidence less
overwhelming.

3. Data source is World Trade Flow documented by Feenstra etc. (2005).

4. Melitz (2003) (,liscussed the model in a general equilibrim‘n and explicitly
solved the number of firms endogenously.

5. As shown in Helpman, Melitz and Yeaple (2004), the parameter k represents
the dispersion of the productivities. Lower k means higher dispersion of the firms’
1;)roductivities.

6. See Duan et al. (1983, 1984, 1985) and Manning, Duan and Rogers (1987).
So far they have offered the strongest criticisms against sample selection model.
They argue the selection models are intrinsically flawed because they have to rely
on untestable assumptions and have poor statistical and numerical properties.

7. Hay, Leu, and Rohrer (1987) showed that two part model performs at least
as well as the sample selection model in terms of mean prediction bias and mean
squared prediction error, and significantly outperforms it in terms of parameter
squared error. In a different Monte Carlo investigation, Manning, Duan and Rogers
(1987) put tw -part model in a worst-case setting by assuming the true model is
sample selection. When there are no exclusion restrictions, they find that the
two part model outperforms in terms of mean squared prediction error and mean
prediction bias.

8. Sample selection model is also called adjusted Tobit model or Tobit 2
model.

9. Linear specification of the second part would be inamn'opriate because it

generates large standard errors due to the non-linear property of the second part.
And in order to use linear specification, we need to add unrealistic assumption of
fairly limited range of x and constant partial effects.

10. D21 does not. differentiate two way trade and one way trade. However,
the export from country j to i is treated differently from the export from country
i to country j; which is to say. two way trade between county i and country j will
correspond to two positive records.

11. 1. Prol_)it+Lognormal2. Logit+Lognormal3.Probit+Exponential4.Logit+
Exponential

12. The dummy for contiguity takes the value of one if two countries are
adjacent.

13. The colonial ties are defined in the following way: two countries have had
a. common colonizer after 1945, have ever had a colonial link, have had a colonial
relationship after 1945, or are currei'itly in a colonial relationship.

14. The distance formula used is a generalized mean of city-to-city bilateral
distances developed by Head and Mayer (2002), which takes the arithmetic mean
and the harmonic means as special cases.

15. The variable to be explained in Equation (2.22) is the probability that an
entree firm becomes an exporter. This can be interpreted as the adjusted number of
exported varieties, in the following case: The relative number of potential entrees,
or the relative size of the mass of potential entrees in each industry is the same
across exporters, which is also the same as the relative number of total varieties in
the US. imports. (For an illustrative example; the size of the potential entree mass
in book publishing industry is twice as large as that. in men’s footwear industry
in all exporters, and the US. imports twice more varieties in the former industry
than does in the latter.

16. we also attempt two other cutoffs of S/ U to divide the sample countries

60

into North and South; above or below the 75 percentile (29 countries for North,
86 for South), and above or below 0.7 of the value of S/ U relative to the US.
('25 for North, 90 for South). These groupings are also indicated in Table 16.
The qualitative results of estimation (the sign and significance of the coefficient
estimate) are the same as shown in Table 18 regardless of the cutoffs.

17. nshure is skewed in distribution, and so log—scaled in regressions to adjust
for heteroskedasticity.

18. As described in Section 3, skill is measured by the share of non-production
workers in the total number of employees. Unskilled-labor intensity is defined by
the share. of production workers in the total employment.

19. Recall that we develop the theoretical model in the two-country frame-
work. The values of 1*)arameters in the model are likely to be different across
country pairs.

20. The mean of S/ U, which is used as the cutoff of North-South in the
previous subsection, is 0.59, which is a little lower than this value.

21. The same approach has been used by Frankel and Rose (2003) for analysis
on the effect of trade on the environment; Edmonds and Pavcnik (2006) for effect
of trade on child labor.

22. Publications include International Financial Statistics and Balance of
Payment Statistics.

23. Geographical Distribution of Financial Flows to Developing Countries.
The data include gross FDI origninated in OECD countries into developing coun-
tries.

24. FDI inflows and outflows comprise capital provided (either directly or
through other related enterprises) by a foreign direct investor to a FDI enterprise,
or capital received by a foreign direct. investor from a FDI enterprise. FDI includes

the three following components: equity capital. reinvested earnings and intra-

61

company loans. - Equity capital is the foreign direct. investor’s purchase of shares
of an enterprise in a country other than that of its residence. — Reinvested earnings
comprise the direct investor’s share (in proportion to direct. equity 1')articipation)
of earnings not distributed as dividends by affiliates or earnings not remitted to the
direct investor. Such retained profits by affiliates are reinvested. - Infra—company
loans or intra—company debt transactions refer to short- or long-term borrowing
and lending of funds between direct investors (parent enterprises).

25. The GDP measure from Penn Table is in thousands US current dollars

and FDI in million dollars, so each unit of the FDI/GDP is in thousands.

APPENDICES

63

Appendix A
A.1: Proof of Lemma. 1
Recall that revenue term

 

 

 

 

(.I.
, J 1—0
7... : v. \
J J Willi?)
._ 1-0. .
ll] — 7' r]
5(6) = 0f
*
Hoary) Ufj‘i
Then
TC,‘
y'. ( )1—0’
.,1.* 2 .-* . A .,
Dir-1:11) {A = p".1'ji(l)Pl _( 9v? )1—0szTl—a
rear) f y 1..., set-- Y-
1 1 J (Pfifillpj) .sz J
.. r] (——...—"J——)1—0 *
“(TN-.1.) _f1] _ f)¢:rij(l)Pj __ 99]. 1_0Yj 1—0
W‘T“ . g 1_ 4...») W
J Y.) 33, ( W45 ,) " *zzr—z-J "
Par/Jinpz
Thus
l"*.. .. /'
_ [Lo-1(2)”: = 9
par — f c- Y- 1
J ”J .7

 

'fi* .- .p *o -
Val-J1 : @Tg_1((l)g ] (22

e? f

A.2: Exponential Distribution

(A.6)

The exponential distribution may be. viewed as a continuous etmnterpart of
the geometric distribution. The log likelihood function specifications are similar

to lognormal distribution.

With the first. setting of probit. for hurdle part and expmiential distribution

as the level part:

P(D,- =1

 

64

I . ,’ ,

(A9)

I , .
E(Tilsz = LX223 = €XP(X2,-f32) (A10)

The unconditional expected value is:

. r I r, /
Esz'leiaXZi) = Ethlei = LX223 P'T'fDi = 1pm) = exp(X2,:x’32)‘P(A1z-fii)

 

(A.11)
The log-likelihood functions will change accordingly to
I
111(0) : 171(1 — CI)(X1’:,131)) (A12)
J .. . ’ ,, Ti
(X21432)

and marginal effect of distance is:

0E(T2-)
BIA.
In estimation part. when we use total trade volume for level of Ti and I}; is
the log form of distance, the elasticity of distance on trade volume is:

 

I , ’I I 4. , rI
= (I)(.X'1?-_,131);32k exp(X2i;32) + 6(X1ii31);31k exp()&2i,{32) (A.14)

 

 

6E(ln trade) I _, ,. , ’ , ’
. = (I) X 3 . r X :13 ,3 . X -3 A.1"
0(lndistance) ( W31)“ 2”“ 11,1) 1“ 2” 2) ( ")
0E(lntrade) I I I , I
= X :13 (I) X -;3 , _ — X d3 (I) X d3 . _
0(RTA) e\1)( 2:... 2) ( 1'Is1llRTA—1 ekpt 2,, 2) ( 11,1)lRTa—f6)

When part one is logit and second part lognormal, the unconditional expected
value would be:

 

 

E('Ti|X127aX2-zt) = Emmy: = 11X27i) P7'(Di:1lX1i) (A17)
The log-likelihood functions will change accordingly to
I
ex X .213
111(0) =ln(1— p(, 1” 1) (A.18)
exp(X1i/31) + 1
(X, '3 ) T
eX) .1 I ' -
ln(Ti) = ln(l — I I 12" 1 ) + ln(XQi-flg) - -,—"’ (A.19)
exp(X1il31) +1 (X21332)
and marginal effect of distance is:
I ,_ ,I
0E(Ti) exp(X1?-/31) exp(X1i/31)

 

 

 

/ , ’ .
2 [32k exp( X 2i ,{3 2 ) + [31 k exp ( X21132)

' .. I I
0" k exp(X1i1_l31) + 1 (GXMXUf/gl) 'i’ 1)2

(A20)

The elasticity of distance on trade volume is:

I , I ,
0E(lntrade) __ exprlifll) $3 + exprh-I'31)

I
., . . . — . 2A. 1311.:(X 332) (A21)
0(1n dlstance) 9X1)(Xii/31) + 1 (avail-I31) + U2 27

 

 

 

“’hen part one is logit and second part lognormal, the unconditional expected
value would be:

 

 

 

 

 

 

 

I
p . I , r 2 exp(X1i,{31)
E(TilDi = 1.}(22') P'l‘(D.Ij = IIJYU) = CXp(JY2i,I32 + 0.00 ) I
exp(X1,i_/)’1)+1
(A22)
The log-likelihood functions will change accordingly to
I
ex ) X J3
111(0) 2 ln(l — M, 1" 1) (A23)
exp(X1,l-,’31) + 1
(X' '3 ) 1 T X' 3
ex n ' —
ln(Ti) : ln( 1) ', 12 1 )+ 111(6)( 2 2M 2)) — 1n 0T2- (A24)
exp(X1i/31) +1 0
and marginal effect of distance is:
I I
a}: T- exp<X .51) , , ,. exth -..r31) . , .
( l) = 1" (32k exp(X2,l-,z32) + 1" (31k exp(X2i;32)

'1‘ _ I . I ’
()J'k ex1)(1’(1i,i'31) + 1 (9X1)(X1i,i31) + 1)?

 

 

 

 

 

 

(A25)
The elasticity of distance on trade volume is:
(El 1 (‘53) (XI?)
. t, . _ eX) 1 , em :1 I
a 1 (Slim?) = I , 11 1 #3211: + I , 1' 1 ,.r'31k(x2i,32) (A26)
( n 1stance) exp(Xll./31) + 1 (exp(X1i/31) + 1)2
0E(ln trade) _
0(RTA)
(x’ 3 ) (X' 3 )
Pl exp . 1 fl exp ,7; 1
ex1)(.\21,i“32) I h lRTAzl _ CXI)(4X2i,i'.32) ,I 1.! lRTA=0
exp(X1i/31) + 1 exp(X1i,31) + 1
(A27)

66

A3: Vuong’s Test:

For the first part of the estimation, we choose from Probit vs. Logit esti-
mations, then exponential vs. Lognormal for the second part. These two sets
of models are strictly non-nested because we cannot get one model by putting
restrictions on the other model. Our Vuong’s test statistic is

f(yll (21112 ’6‘) 1 III-I26)
TLR1VV’_\/-Zl I I {120121 I ——))2—(;Zl11%)2}"6

(.Uil—l—tj') 9(1/1l711’7)i___1 g(.(/zl111’l)
(A28)
\‘Ve spet ifv log likelihood fun< tions for exponential and lognormal distribution
individually by f( (ll; |.1i,) a)nd (1(1)) I 1,, '7). For 132, we use

:2:{_ :Zml( _f_____(?/jl-1i16) _ (1:111 f(yzlrlfifé ___’;%)2} (A29)

lelzv') 91"

as an estimate of the variance of WLR“), :7‘).

The hypothesis is:

—_ _ 1k.

Eh denotes the expectation with respect to the density of the true data-
generating process, h,(yZ-|;1?.z-). We reject at the significance level 5% the null hy-
pothesis of equivalence of the model in favor of F9 being better (or worse) than
072 if TL R, N N > 30.05 ( or if TL 3. N N < —:0.05 ). The null hypothesis is not
rejected if lTLRNNl S «30.025 -

The Vuonng test is applied on the full sample for the first part while only
sub-sample with positive values for the second part. The test output is shown in
Table 11.

67

Appendix B
B.1. Derivation of equations (2.17),(2.18):
Use equations 2.6. 2.15 and 2.16, we can get

H
H {H A 1_0
aft-A 2(1)} (———H H) (B.1)
663‘ P,-
F
. F_ . F A 1—0
asz —a,Y (pc‘fi‘FP-F) (B2)
1. 2
H
H F T14 1_0. ‘
0 -.A 202:)” —— 8.3
.1. . (pet! Pf) ( 1
F
F -H TA 1—0
OfiIA :OiY ( , F H) (8'4)
p69: P2.
and A)‘ = (s’\)282'-(2w)‘)1_1‘3i
Divide (B.1) by (BA), and (B2) by (BB),
2H , H
(’0' — f A 1_
(4)01:_l_._0.7 0 (3.5)
(3),}: fix (AF)

IF . F
01'. )0—1 _ _fz_ A 1—0

 

(— — -(—)U'T (13-6)
9’95 fir AH
B2. Derivation of equations (2.20), (2.21)
we can write
7 4.1
2 A . . (1”(‘2’94 ) —1 H
1212»: 1:11— 6(22:A)II<—;\—1" — 11 (3..)
‘32:
and (3(0)?) is the weighted average productivity which is expressed as:
271w“) = 1 [/00 ¢0_lg((‘9)d<f>]0l1 (B.8)
* " 11 — G16?» 22:1 '

In Pareto specification,

 

~‘ 51A 1 .
Q1222“) = [sac/221110;; )10—1—11 = 11—G<22;1*>III«’—1I = (Ix'—1)<%%;1k
" z(13.9)

68

Therefore. the free entry conditions under costly trade would become:

f,3(1{— 1)€>i..i..(<.°f‘)_"+ f,,.(K — 1)..,1;,,,,(., .H ,H =f,-,,21 (13.10)
f,-(I\'—1)o(‘nin(oF) +f,-,.I( 1 —1)2f;,in(2£)’k = f...) (B.11)
Combining (B.10). (8.11) with equations (2.17), (2.18), we can derive a system

of two equations for two unknowns: (oz-H )‘k and (of)-

linear in these two variables.

1“. Both equations are

fj(1\‘ _ IO) 111in\1H((-"1Fit—) k + firiK _ 11‘2‘)‘i(C‘)g)—k Z fie‘s (13.12)
F A ,
"f(1"‘—1)©111111Ai ((2’) 1H.1'—) k+fi.r(I\ ‘ 1)"L( 13)-," Zfic‘s (13.1.3)

B.3. Proof of proposition 1:

fieMAgi ’ I11)

 

 

 

 

1.022 7.”) = (—;‘.:‘}}‘>’: = f‘ 2 (13-141
"Di”? I —1 AiFAz-H
( \ )( fi- if?)
Then
f —k —Im 4H l—A'a—cr
0(1— C(cvj‘JHH: fied‘rl" (fi)fl(affok)(fl*) 0—]‘ (Bl )
. 5
, AH f2
d<71T)(I\ —1) [221(91) ‘Tfi— 41.11)]
291— Ga *H
By assumption 2 the sign of ‘ A(H*‘L‘ )] =the sign of (3LT)- which is
. (9"?)
negative.
Therefore,

a[1—G(ejfif’)l _ 011*G(<°i.“)l < 0 (B 16)
,1_f}i — :11. ‘

 

 

 

7.1} F

:H_
. 11’ “1 [,2 .
Tlns means when —fl is fix<d we can inspe (t the effect ofchange of (i2 )"32
UV

on 1 -— ("(031111

60

. . . . S 8 .
\\ hen exporter 1s more sklll abundant. than the unporter, 17% > E111?" then

the increase of 13 (more skill intensive industrv) would result in smaller export
varieties. \\ hen home eonntrv is more skilled labor intensive than the f01eign

S
country. 5% < ”5-11 then the increase of 3 (mole skill- intensive industrv) would

result in larger expolt varieties.
_11_
. . u. . .
[3.4. Proof of propostion 2: VV hen we fixed (W— 2 term. the. negative der1va-

(35')

tive of equation A. 15 would simplv mean highel unskilled-1211101 cost in home coun-
try result in s1nalle1 export variety.

Figure 1: Cross-Country Distribution of Trade

 

 

 

90%.. HH__~H_H.2._112._

m%—~~e~~~~wee~ar

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

4m2~~~e~~~~~~~~~~
w%~e~~~~~e~e~~e~
20%LHH.~_12___M_._HH~~

m%w~aA~~e~e

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mm2~H 1+ »»~

 

70% ’“ ”HH“*~*~~~~~
60% I] I loll] I '1'" II |

50% ~24 H H .2 2 2+ b 22 2.-.- 2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0W0 1L1 .Jdt

 

 

 

[31 no trade

C] two-way trade

2 I one-way trade

 

 

Years 1980-1997

Note: Years are graphed on x-axis from 1980-1997

All possible country pairs are 36290

Data is from World trade flow, by Feenstra ect. (1995)

71

 

Figure 2: Geographical Distribution of RTA (Both in force or under negotiation)

 

 

 

 

120 ~ —————---—-— -—
El FTAs under negotiation
100 1-- _____-__xm2,- -21x% IFTAs in force
ECU in force

 

 

 

 

 

 

Number of RTAs

 

 

 

 

 

 

 

(I) O C C -
(U (U
8 ’5 23:17: 8 ‘3 c
C m '-< c m .9
‘1’ n. 3— 6m '5 g
E m “‘9 '5': a) a:
< “’0’ '0 3 3
‘60 0 m o
co 2 -
Lu°5 0

Region

Source: World Trade Organization 2000 report

72

Figure 3: Geographical Distribution of RTA as of year 2000

 

no RTAs

1 to 3 RTAs

4 to 7 RTAs

8 to 12 RTAs
12 to 15 RTAs
>16 RTAs

HDDDDD

Source: World Trade Organization year 2000 Report

73

Table 3: List of Countries:

Afghanistan
Albania
Algeria
Amer. samoa
Angola

Anti gua&Barbuda

Argentina
Armenia
Aruba
Australia
Austria
Azerbaijan
Bahamas
Bahrain
Bangladesh
Barbados
Belarus
Belgium
Belize
Benin
Bermuda
Bhutan
Bolivia

Bosnia&Herzego.

Botswana
Brazil

Brunei Darussalam

Bulgaria
Burkina Faso
Burundi
Cote d'Ivoire
Cambodia
Cameroon
Canada

Cape Verde
Central African
Chad

Chile

China Hungary
Colombia Iceland
Comoros India

Congo Indonesia
Costa Rica Iran
Croatia Iraq

Cuba Ireland
Cyprus Israel

Czech Republic Italy

Korea Jamaica
Congo Japan
Denmark Jordan
Djibouti Kazakhstan
Dominica Kenya
Dominican Rep. Kiribati
Ecuador Kuwait
Egypt Kyrgyzstan
El Salvador Lao
Equat.&Guinea Latvia
Eritrea Lebanon
Estonia Lesotho
Ethiopia Liberia

Fiji Libyan Arab
Finland Lithuania
France Luxembourg
French Polynesia Macao
Gabon Madagascar
Gambia Malawi
Georgia Malaysia
Germany Maldives
Ghana Mali
Greece Malta
Grenada Marshall Is.
Guatemala Mauritania
Guinea Mauritius
Guinea-Bissau Mexico
Guyana Micronesia
Haiti Mongolia
Honduras HongKong

74

Morocco Slovakia
Mozambique Slovenia
Namibia Solomon Islands
Nepal Somalia
Netherlands South Africa
Net. Antilles Spain

New Caledonia Sri Lanka
New Zealand Sudan
Nicaragua Suriname
Niger Swaziland
Nigeria Sweden
Norway Switzerland
Oman Syrian Arab
Pakistan Taiwan
Palau Tajikistan
Panama Thailand
Papua N.Guinea Yugoslav
Paraguay Togo

Peru Tonga
Philippines Trinidad
Poland Tunisia
Portugal Turkey
Puerto Rico Turkmenistan
Qatar Uganda
Korea Ukraine
Moldova United Arab
Romania UK

Russian Fed. Tanzania
Rwanda US

St. Kitts&Nevis Uruguay
Saint Lucia Uzbekistan

St Vt.&Grenadines Vanuatu
Sao tome&principe Venezuela

Saudi Arabia
Senegal
Seychenes
Sierra Leone
Singapore

Viet Nam
Yemen
Zambia
Zimbabwe

Table 4: Summary Statistics

Variable

Trade

Log of trade

Log of importer GDP
Log of exporter GDP
Log of distances
Contiguity

Common Language
Colonial Ties
Landlocked Exporter
Landlocked Importer
Major cities in exporter
Major cites in importer
RTA

Full

sample

Export>0

 

Mean

165511 2534012

10.1249
10.1249
8.77504
0.01565
0.16269

0.0108
0.18848
0.18848
21.8684
21.8684

0.0652

2.2714
2.2714
0.77531
0.12413
0.36909
0.10337
0.3911
0.3911
6.34529
6.34529
0.24688

75

Std.Dev. Mean

562450
9.52777
1 1.6104
1 1.7063
8.60933
0.02585

0.1222
0.02837
0.13981

0.1355

23.825

23.825
0.06742

Std .

4647484

2.75813
1 .90957
1.81652
0.83267
0.15868
0.32753
0.16605

0.3468
0.34227

4.3851

4.3851
0.25076

Table 5: Estimation of Gravity Equations by Conventional Methods

 

Estimator OLS OLS PML PML

Dependent variable ln(Tij) ln(1+Tij) Tij>0 Tij

Log exporter GDP 1001*" 1.030*** .102*** .360***
(0.011) (0.009) (0.001) (0.004)

Log importer GDP .910*** .965*** .092*** .334***
(0.010) (0.009) (0.001) (0.004)

Log distance -1 019*” -.647*** 2104*" -.262***
(0.022) (0.024) (0.002) (0.008)
Contiguity .473*** .343** 001 _.290***

(0.116) (0.151) (0.009) (0.048)
Common-language .565*** .196*** .056*** —0.001
(0.054) (0.049) (0.005) (0.024)
Colonial-tie 1.141*** 2866*" .099*** .232***
(0.106) (0.173) (0.008) (0.050)
Landlocked-exporter -.376*** -.642*** -.038*** -.096***
(0.051) (0.046) (0.005) (0.020)
Landlocked-importer -.677*** -.788*** -.072*** -.193***
(0.051) (0.046) (0.005) (0.020)
Num. of cities in expor-.052*** -.057*** -.005*** 0.002
(0.004) (0.003) 0.000 (0.002)

Num. of cities in impor-.057*** -.052*** -.005*** 0
(0.003) (0.003) 0.000 (0.001)
RTA 0.61*** -0.031 0593*" 0.654***

(0.147) (0.081) 0.000 0.000
Observations 10607 36290 10607 36290

Note: ***, **, * each indicates significance at the level of 99%, 95%,90%
OLS stands for ordinary least squares estimation
PML stands for Poisson Maximum Likelihood estimation

76

Table 6: Estimation Result: Two-Part Model Without Sampling Weight

 

Probit Probit Legit Logit Probit Probit Logit Logit
Lnormal Lnormal Lnormal Lnormal Exp. Exp. Exp. Exp.
Part 1 Part 2 Part 1 Part 2 Part 1 Part 2 Part 1 Part 2
Log (ex_GDP) .45*** 100*" .82*** 100*” .45*** .74*** .82*** .74***

 

 

 

 

 

 

 

 

 

 

 

 

(0.010 0.010 0.010A 0.000| (0.005A 0.004 0.010 0.004
Log(im_GDP).42*** “91*" .42*** . .73***
(0.010A 0.010 0.010 0.010A 0.005A 0.004

”704:4!!!

  
  

 

 

 

(0.010A 0.020A 0.020A 0.020A 0.010 . A . A

(0.040 0.070 0.070 0.070 0.040 0.040A 0.070 0.040A
Colonial .46*** 1.24***
Tie (0.080A 0.100A 0.150A 0.100A A 0.150A 0.060A
‘Lan ua e (0030 0.050 0.050 0.060A . . A 0.050 0.030
Contiguity -.38*** 039*" -.71*** 0.40*** -.38*** 070*" -.71*** 0.70***
ex orter (0.020A 0.050 0.040 0.050A 0.020 0.020A 0.140A 0.520
Landlocked -.21*** -.65*** -.35*** .65*** -.20*** -.82*** -.35*** -.86***

 

 

 

 

 

Exporter -.004** m-.004** -.05*** -.01** 05***
City number (0.001A 0.004A 0.003A 0.010A 0.001(0002A 0.003A 0.002
Importer -.006** -.O6*** -.01*** -.05*** -.006*** -.O8*** -.01*** -.08***
City number (0.002 (0.003 (0.003 (0.003 (0.002 (0.002 (0.003 (0.002)

 

 

Log
Likelihood -l 34563.2 I -l 34519.4 I -I 39782. 14' -l 39738.33

 

 

 

 

77

Table 7: Estimation Result: 2-palt Model With Sampling Weight

 

 

 

 

 

 

 

 

 

 

 

Probit Probit Logit Logit Probit Probit Logit Logit
Lnormal Lnormal Lnormal Lnormal Exp. Exp. Exp. Exp.
P1 P2 P1 P2 P1 P2 P1 P2
Log .10“ 1.13“” .17‘“ 1.13“" .19“ 93*“ .17“ 93*“
(ex cop) (0.040) (0.040) (0.100) (0.010) (0.040) (0.030) (0.090) (0.030)
Log .18*** 1.12"" .32'“ 1.12"" .18*** 1.03“" .32*** 1.03“”
(im GDP) (0.030) (0.040) (0.010) (0.010) (0.030) (0.040) (0.060) (0.040)
Ln(dist.) -0.07 -.81*"’* -.15*** -.81*"’* -0.07 -.65"** -0.15 -.65***
(0.070) (0.060) (0.260) (0.010) (0.070) (0.040) (0.140) (0.040)
RTA 1.05““r 1.32““ 196*“ 1.32'“ 1.05“" .87*** 1.96“ .87***
(0.370) (0.270) (0.130) (0.040) (0.370) (0.190) (0.740) (0.890)
Colonial 0.64 .43" 1.27m 43*" 0.64 0.17 1.26 0.17'
(0.400) (0.180) (0.090) (0.030) (0.400) (0.150) (1.790) (0.150)
Common -.85** 0.29 -1.5*** .29*** -.85** 0.23 -1.53** 0.23
(0.360) (0.230) (0.460) (0.020) (0.360) (0.190) (0.640) (0.190)
Contiguity -0.56 -0.52 —.99*** -.52W -0.56 0.09 -0.98 0.09
(0.600) (0.480) (0.080) (0.030) (0.600) (0.290) (1.180) (0.290)
landlocked -0.11 -0.12 -.23** -.12*"’ -0.11 -.22** -0.23 —.22**
(0.120) (0.860) (0.080) (0.040) (0.120) (0.120) (0.230) (0.120)
landlocked 0.01 -0.04 0 -0.04 0.01 -0.06 0.001 -0.06
(0.100) (0.130) (0.070) (0.040) (0.100) (0.120) (0.180) (0.120)
Exporter -0.01* -.07*** -.03*** -.07*** -02" -.06*** -0.02 -.06***
(0.010) (0.010) (0.010) 0.000 (0.010) (0.010) (0.010) (0.010)
Importer -0.03** -.11*** -.06*** -.11*** -.03*** -.09*** -.06** -.10***
(0.010) (0.010) (0.010) 0.000 (0.010) (0.010) (0.020) (0.010)

 

78

Table 8: Marginal Effect Derived From Other Methods

 

 

 

 

 

 

Estimation Dependant Mar. effect of Mar. effect of
Variable ln(dist.) RTA
OLS ln(Tij) -1.0l9*** 0.61 ***
positive trade only (.022) (0.147)
OLS ln(. 1 +Tij) -.647* * * -0.03l
all possible (.024) (0.081)
country pairs
PML Tij>0 -.104*** O.593***
positive trade only (.002) (0.000)
PML Tij -.262*** 0654*”
all possible (0.008) (0.000)

country pairs

 

 

 

 

Note: ***, **, * each indicates significance at the level of 99%, 95%, 90%
OLS stands for ordinary least squares estimation

PML stands for Poisson Maximum Likelihood estimation

79

 

Table 9: Marginal Effect of Iog(distance) and RTA on Iog(trade volume)(with the
Trading Partner’s Sizes as Weight Factor)

 

 

 

 

 

 

 

 

 

 

Marginal effect of Marginal effect of RTA
Log(distance)

Pl: Probit -1.05*** 161*“

P2: Lognormal (0.24) (0.04)

P1: Logit -0.85*** 1.61***

P2: Lognormal (0.01) (0.32)

P1: Probit -0.88*** 121*“

P2: Exponential (0.24) (0.22)

Pl: Logit -.69*** 1.16***

P2: Exponential (0.08) (0-25)

 

 

 

80

 

 

Table 10: Marginal effect of log(distance) and RTA on log(trade volume)(without any

 

 

 

 

 

 

 

 

 

Weight Factor)
Marginal effect of Marginal effect of RTA
Log(distance)
Pl: Probit -1.20*** 026*”
P2: Lognormal (0.025) (0.098)
P1: Logit -1.34*** 035*”
P2: Lognormal (0.089) (0.096)
P1: Probit -0.92*** 0.18***
P2: Exponential (0.016) (0.075)
P1:Logit -l.07*** 028*“
P2: Exponential (0.022) (0.074)

 

 

 

81

 

Table 1 1: Vuong (1989)”5 Test for Model Selection

 

LR Test Statistics

Result

 

Probit (Model 1)

 

 

Vs. -4.193e-O7 H0 is not rejected
Logit (Model g)

Exponential (Model f)

Vs. -8.32e-10 H0 is not rejected

Lognormal (Model g)

 

 

 

82

 

Table 12: US. Import and Varieties in 1990

1111111 Import h‘lillltllilcttll'illg Import

 

 

 

Number of Exporting 153 153
Countries

Number of Varieties 182,230 171,322
Total Import Value 495,260 409,953 '
(in million S)

 

 

 

 

 

Notes:

1. The data are from Feenstra, Romalis, and Schott (2002).

2. Manufacturing import is the import in the industries classified as the 4-digit
U.S. SIC (1987 version) 2011 through 3999.

3. Exporters in this table include overseas territories of countries.

4. The number of varieties is defined as the number of commodities classified by
the 10-digit Harmonization System (HS) that the US. imports fiom each

exporter. (i.e., the same IO-digit HS commodities imported from
different exporters are counted as different varieties.)

5. Import value is the customs value of general imports. General Imports measure
the total physical arrivals of merchandise from foreign countries, whether such
merchandise enters consumption channels immediately or is entered into bonded
warehouses or Foreign Trade Zones under Customs custody.

83

Algeria
Angola
Argentina
Australia
Austria
Bangladesh
Barbados
Belgium
Benin
Bolivia
Brazil
Burkina Faso
Burundi
Cameroon
Canada
Central African Republic
Chad

Chile

China
Colombia
Congo

Costa Rica
Cote d'Ivoire
Cyprus
Czechoslovakia
Denmark
Dominican Republic
Ecuador
Egypt

El Salvador
Fiji

Finland
France
Gabon
Gambia
Germany
Ghana
Greece
Guatemala
Guinea

Table 13: Country List (as of 1990, 115 countries)

Guinea-Bissau

Guyana
Haiti
Honduras
Hong Kong
Hungary
Iceland
India
Indonesia
Iran

Ireland
Israel

Italy
Jamaica
Japan
Jordan
Kenya
Madagascar
Malawi
Malaysia
Mali

Malta
Mauritania
Mauritius
Mexico
Morocco
Mozambique
Netherlands
New Zealand
Nicaragua
Niger
Nigeria
Norway
Oman
Pakistan
Panama

Papua New Guinea

Paraguay
Peru
Philippines

84

Poland
Portugal
Reunion
Rwanda
Saudi Arabia
Senegal
SeycheHes
Sierra Leone
Singapore
Somalia
South Africa
South Korea
Spain

Sri Lanka
Sudan
Suriname
Sweden
Switzerland
Syria
Taiwan
Tanzania
Thailand
Togo
Trinidad and Tobago
Tunisia
Turkey
U.S.S.R.
Uganda
United Kingdom
Uruguay
Venezuela
Yugoslavia
Zaire
Zambia
Zimbabwe

Table 14: Factor Abundance of Countries: Skilled Labor (S) to Unskilled Labor (U)

 

Variables Mean Std. Min. Max.
Dev.

S/U ratio 1.879 0.553 1.075 3.369

S/U ratio 0.567 0.167 0.325 1.017
relative
to US

 

Number of countries: 115

10 most skilled labor-abundant countries:

 

 

Country S/ U ratio S/ U relative to U.SA
New Zealand 3.369 1.017
Hungary 3.086 0.932
Norway 3.010 0.909
Canada 3.008 0.908
Denmark 2.999 0.905
Australia 2.981 0.900
Finland 2.833 0.855
Sweden 2.825 0.853
Israel 2.818 0.851
Belgium 2.768 0.836
10 most unskilled labor-abundant countries:
Country Name S/ U ratio S/ U relative to U.S.A
Niger 1.075 0.325
Guinea-Bissau 1.078 0.325
Benin 1.098 0.332
Mali 1.116 0.337
Rwanda 1.119 0.338
Gambia 1.119 0.338
Sudan 1.130 0.341
Mozambique 1.156 0.349
Central African Republic 1.184 0.357
Nigeria 1.217 0.367

Note: The relative skilled-labor abundance to unskilled labor (S/ U) is measured by
the human capital-to-labor ratio provided by Hall and Jones (1999).

85

Table 15: Input Factor Intensity of Industries: Skilled-labor (S) to Unskilled-labor (U)

 

 

 

 

 

 

 

 

 

 

Variables Mean Std. Dev. Min. Max.
S -intensity 0.296 0.124 0.078 0.827
U -intensity 0.704 0.124 0.173 0.922
10 Most Skilled-labor intensive industries
SIC Industry Description S—intensity U-intensity
2721 Periodicals 0.827 0.173
2731 Book Publishing 0.766 0.234
3571 Electronic Computers 0.718 0.282
3761 Guided Missiles & Space Vehicles 0.685 0.315
271 1 Newspapers 0.676 0.324
2741 Miscellaneous Publishing 0.638 0.362
2835 Diagnostic Substances 0.633 0.367
3572 Computer Storage Devices 0.627 0.373
3826 Analytical Instruments 0.617 0.383
2086 Bottled and Canned Soft Drinks 0.604 0.396
10 Most Unskilled-labor intensive industries
SIC Industry Description S-intensity U-intensity
2322 Men's & Boys' underwear & 0.078 0.922
nghtwear
2281 Yarn Spinning Mills 0.089 0.911
2284 Thread Mills 0.097 0.903
2211 Weaving Mills, Cotton 0.102 0.898
2436 Softwood Veneer and Plywood 0.105 0.895
2015 Poultry and Egg Processing 0.108 0.892
3263 Fine Earthenware Food Utensils 0.111 0.889
2325 Men's & Boys' Trousers & Slacks 0.116 0.884
2321 Shirts, Men's and Boys' 0.120 0.880
3144 Women's Footwear, Except Athletic 0.120 0.880
Notes:

1. The source of the data for factor intensity is 1992 US. Census of Manufactures.

2. Industries are classified according to the 4—digit US. Standard Industrial Classification

(SIC; 1987 version).

3. Skilled-labor (S) intensity is defined as the share of non-production workers in the total

number of employees; and unskilled-worker (U) intensity is defined as the share of
production workers. The sum of S-intensity and U-intensity is thus one for each

industry.

86

Table 16: List of Countries in Aggregate North and South

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

North (51 countries) South (64 Countries)
Argentina Sri Lanka Algeria Oman
Australia Sweden Angola Pakistan

Papua New
Austria Switzerland Bangladesh Guinea
Barbados Taiwan Benin Paraguay
Belgium Thailan Bolivia Portugal
Trinidad and
Canada Tobago Brazil Reunion
Chile United Kingdom Burkina Faso Rwanda
China Uruguay Burundi Saudi Arabia
Costa Rica U.S.S.R. Cote d'Ivoire Senegal
Cyprus Venezuela Cameroon Seychelles
Central African
Czechoslovakia Yugoslavia Republic Sierra Leone
Denmark Malaysia Chad Singapore
Ecuador Malta Colombia Somalia
Egypt Morocco Congo Sudan
Dominican
Fiji Netherlands Republic Suriname
Finland New Zealand El Salvador Syria
France Norway Gabon Togo
Germany Panama Gambia Tunisia
Greece Peru Ghana Turkey
Guyana Philippines Guatemala Uganda
Hong Kong Poland Guinea Tanzania
Hungary South Korea Guinea-Bissau Zaire
Iceland South Africa Haiti Zambia
Ireland Spain Honduras Zimbabwe
Israel Japan India Mauritania
Italy Indonesia Mauritius
Iran Mexico
Jamaica Mozambique
Jordan Nicaragua
Kenya Niger
Madagascar Nigeria
Malawi Mali
Mali

 

 

 

 

87

 

Table 17: Skilled-to-Unskilled Labor Ratios (S/ U) of North and South

 

 

S/ U S/ U relative to
US.
(group average) (group average)
North 2.4 0.72
South 1.47 0.44

 

 

 

 

Note: The relative factor abundance (S/ U) is measured by the human capital to labor
ratio in Hall & Jones (1999).

88

Table 18: Regressions for Aggregate North and South

Dependent Variable: Log of aggregate no. of varieties as the share in the total no. of
varieties imported by the US.

 

 

North South
skill 0256‘" -1.21'"
-0.041 -0.208
constant -0.256m -1.54m
-0.014 -0.063
Observations 394 385
R2 0.1 0.12

 

Notes:

1. Regression equation is (4.1).

2. skill is skilled-labor intensity of each industry.

3. Robust standard errors are in parentheses.

4. *, **, and *** indicate that the coefficient estimate is significant at the 1%-level,
5%-level, and 10%-level, respectively.

Table 19: Pooled Regression for Individual Exporters

Dependent Variable: Log of no. of exported varieties in each industry as the share in
the total no. of varieties imported by the US.

 

 

skill_i -272”

-O.566

skill_i * (S / U )_c 4.14”

-0.802

Observations 17,050
it2 0.13

 

Notes:

1. Regression equation is (4.4). Country-specific dummies are included.
2. skill_i is skilled-labor intensity of each industry; and (S/U)_c is skilled-to-unskilled
labor endowment ratio in each exporter, relative to the US.
3. Standard errors in parentheses are clustered by country.
4. *, **, and *** indicate that the coefficient estimate is significant at the 1%-level,
5%-level. and 10%-level, respectively.

89

6

EmEd A” 25869 c». mxvonoa 8 90 cm. 5 one: 55360815 59,654“ 5 Eco

88888388

208%

_. 59538 2.0 58m. 5 came ow mE=oa-_mco_. 38.55. .28 a: mm 9o 30% arm—E9553 383?? Ba 30 1%: mm Ea Ema.

N. 92:8

H.63— Zsaus. on 562.8: 8 Cm E 53" 3. =5:me

 

Facade”

  

3333:2233;3m: $23,332,:

ram: :38

Q: :5 62—2. 3. wE__oa-_a—.S. Ema—Ea: 5: W Home arm—.oaiiaew ”58533

:59. 58:me mm cannon mm :8 £58 ow =0:

Bonanza: €9.52 5 =5 88— 2552 cm oat—8.8m.

   

90

Figure 5: Scatterplot of Number of Exporters v.s. Industry Skilled-labor Intensity
(U.S. Manufacturing Imports in 1990)

 

 

8

888858

B

 

0 0. 1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

 

 

 

Plot: Number of Exporters vs Industry Skilled-labor Intensity (Solid Line=trend
line)

x-axis is the Skilled-labor Intensity; y-axis is the number of Exporters

Note: Skilled-labor intensity is defined as the share of non-production workers in the
total number of employees in each industry.

91

Figure 6: Scatterplot of Number of Varieties v.5. Industry Skilled-labor Intensity
(U.S. Manufacturing Imports in 1990)

10.000 p ...

 

 

 

 

 

1 n

O 0. 1 0.2 0.3 0.4 0. 5

Plot: Number of US-Importing Varieties v.
Manufacturing Imports in 1990)

0.6 (17 0.8 0.2
5. Industry Skilled-labor Intensity (US
X-axis is the Skilled-labor Intensity; y-axis is the total number of Varieties.

Notes:

1.The number of varieties in each industry is defined as the number of lO-digit HS
commodities that the US. imports fiom each exporter in each 4-digit SIC industry (i.e.,
the same HS commodity imported from different countries are counted as different
varieties). The mapping between the lO-digit HTS and the 4-digit SIC is according to
F eenstra, Romalis, and Schott (2002).

2. Skilled-labor intensity is defined as the share of non-production workers in the total
number of employees in each industry.

92

Figure 7: Individual Exporter Regression:
Scatterplot of Slope Coefficient v.s. Skill Abundance of the Country

 

Fltted Line (weighted)

 

 

 

 

 

 

 

 

 

 

 

 

..+‘.___ _._ —?_._ _

Coefficient of Individual Reg
O 01

 

I
U"
l
l
l
1
l
l

 

0.4 0.6 0.8 1

 

SIU relative to LB

 

 

93

 

Figure 8: Exporter’s Relative Factor Abundance, Industry Factor Intensity, and Number
of Varieties in US. Manufacturing Imports in 1990 (1): Selected Skilled
Labor-abundant Countries (relative to unskilled: S/ U)

 

 

O

rank14
rank1 5

ii
N9
‘_

x
‘c‘:
99

 

 

 

 

 

 

 

 

 

 

 

 

 

Average variety Average variety
Rank in Exporter number share in 20 number share in 20
S/U ratio most skill-intensive most unskill—
industries intensive industries
1 New Zealand 0.01 13 0.0084
2 Hungary 0.0065 0.0076
3 Norway 0.0161 0.0046
4 Canada 0.0758 0.0780
5 Denmark 0.0226 0.0099
6 Australia 0.0259 0.0139
7 Finland 0.0160 0.0057
8 Sweden 0.0308 0.0176
9 Israel 0.0273 0.0205
10 Belgium 0.0249 0.0200

 

 

 

 

 

94

Figure 9: Exporter’s Relative Factor Abundance, Industry Factor Intensity, and Number

of Varieties in US. Manufacturing Imports in 1990 (2):
Selected Unskilled Labor-abundant Countries (relative to skilled: U/S)

 

 

 

 

Average variety

Average variety

 

 

 

 

 

 

 

 

 

 

 

 

Rank in Exporter number share in 20 number share in 20
S/ U ratio most skill-intensive most unskill-

industries intensive industries

106 Nigeria 0.0006 0.001 1

105 Haiti 0.0014 0.0042

101 Pakistan 0.0033 0.0090

84 Guatemala 0.0028 0.0101

81 Cote d’Ivoire 0.0010 0.0014

80 India 0.0085 0.0155

79 Kenya 0.0017 0.0015

74 Turkey 0.0035 0.0078

73 Brazil 0.0147 0.0244

72 Honduras 0.0012 0.0057

71 El Salvador 0.0022 0.0051

 

 

 

 

95

 

Table 20: Summary Statistics:

 

 

 

 

 

 

 

 

 

 

 

Variable N Mean Standard Min. Max.
Deviation

Life expectancy 170 65.11 11.96 34 81.9

Male life expectancy 170 62.75 1 1.41 32.4 78.4

Female life expectancy 170 67.54 12.62 35.7 85.3

Prob. of dying (per 170 67.25 70.99 4 332

1000)

Under five, male

Prob. of dying (per 170 61.13 66.54 3 303

1000)

Under five, female

Prob. of dying (per 170 288.67 172.52 81 902

1000)

15-59, male

Prob. of dying (per 170 208.54 166.95 46 789

1000)

15-59, female

General medical 170 1.73 0.39 .40 2.68

expenditure as % of

GDP

GDP per capita 170 9153.43 9607.07 515.50 49367.77

 

 

 

 

 

 

96

 

Table 21: Countries List (78 member countries reporting positive inward F DI during
2000-2004, UNCTAD)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ARE United Arab Emirates LTU Lithuania
ARG Argentina LUX Luxembourg
ARM Armenia LVA Latvia
AUS Australia MAR Morocco
AUT Austria MDG Madagascar
AZE Azerbaijan MYS Malaysia
BGD Bangladesh NGA Nigeria
BGR Bulgaria NLD Netherlands
BOL Bolivia NOR Norway
BRA Brazil NZL New Zealand
BWA Botswana OMN Oman
CAN Canada PAK Pakistan
CHE Switzerland PER Peru
CHL Chile PHL Philippines
CHN China PNG Papua New Guinea
COL Colombia POL Poland
CPV Cape Verde PRT Portugal
CYP Cyprus PRY Paraguay
CZE Czech Republic QAT Qatar
DNK Denmark RUS Russian Federation
EST Estonia SAU Saudi Arabia
ETH Ethiopia SGP Singapore
FIN Finland SLV El Salvador
FRA France SVK Slovakia
GBR United Kingdom SVN Slovenia
GEO Georgia SWE Sweden
HRV Croatia SWZ Swaziland
HUN Hungary SYR Syrian Arab Republic
IDN Indonesia THA Thailand
IND India TUN Tunisia
IRL Ireland TUR Turkey
IRN Iran, Islamic Republic of TZA United Republic of Tanzania
ISL Iceland UGA Uganda
ITA Italy USA United States
JPN Japan VEN Venezuela
KAZ Kazakhstan VNM Viet Nam
KHM Cambodia YEM Yemen
KOR Republic of Korea ZAF South Africa
LBN Lebanon ZMB Zambia

 

 

97

 

Table 22: Predicting F DI: Positive Volume Only

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Variable: inward FDI
ln(distance) -.46***
(0.06)
ln(target country size) 0.02
(0.02)
In( investing country size) 0.04*
(0.02)
Landlocked, target country -0.25*
(0.15)
Landlocked, investing country -0.27*
(0.16)
Common language 082*”
(0.16)
Common Border 0.17
(0.26)
Colonial-tie 0.99***
(0.26)
Observations 3 1 89
R-squared 0.04

 

 

98

 

Table 23: Predicting FDI: Using Two-part Model Approach to Account for the Zeros

 

Dependent Variable: inward F DI

 

 

 

 

 

 

 

 

 

 

ln(distance) _3.95***
(0.21)
ln(target country size) 1.06***
(0.07)
ln(investing country size) 1.06***
(0.07)
Landlocked, target country -0994”
(0.39)
Landlocked, investing country -3. 351:”
(0.45)
Common language -.53
(0.42)
Common Border __15
(0.82)
Colonial-tie 8.85M...
(0.88)
Observations 3 5 532

 

 

99

 

Table 24: Life Expectancy (level) and FDI Ratio (log), Year 2002

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent lexp lexp lexpm lexpf lexp lexp lexp
Variable
ln(ActuaIFDI/GDP) -0.33"
(0.14
ln(fittedFDI/GDP) -1.90m -1.98m -1.85m -0.32 -0.36 -1.66m
(0.63 (0.59)] (0.67)] (0.47)| (0.48) (0.62)
ln(GDPPC) 6.85m)| 14.7)I
(0.76 (10.37
[1n(GDPPC)]_sq 045'
(0.59
In(med_exp_ratio) 6.84”
(2.81)]
Expenditure no no no no no no yes
indicator
Income indicators no no no no no es es
Constructed FDI no es es es yes m es
Observations 7. 78 78 78 7 78 78|
R-squared 0.06 0.1] 0.12] 0.09] 0.56] 0.56I 0.17|

 

Robust standard errors in brackets.
GDPPC is log PPP GDP per capita.
The instruments for constructed FDI measure are the geographical variables.
Lexp: life expectancy at birth
Lexpm: life expectancy at birth for male
Lexpf: life expectancy at birth for female

100

Table 25: Life Expectancy (log) and FDI Ratio (log), Year 2002

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

 

    

 

 

 

Dependent Variable lexp lexp Iexpm lexpf lexp lexp lexp
ln(ActuaIFDI/GDP) -0.005"
(0.00
ln(fittedFDI/GDP) -0_03... -0.03m -0.02m -0.00 -0.007 -0.03u
(0.01)l (0.01)I (0.01)I (0.01 (0.01)] (0.01)I
ln(GDPPC) 0.11m -0.3Zt
(0.01 (0.18)
[ln(GDPPC)]_sq 001'
(0.01
ln(med_exp_ratio) 0.09"
(0.04)]
Expenditure indicator no no no no no no yes
Income indicators no no no no no yes yes
Constructed FDI no yes
Observations 78 z i 78 7 :1 7: 78|
R-squared 0.05] 0.1] 0.1] 0.09L 0.49] 0.5 0.14]

 

Robust standard errors in brackets.
GDPPC is log PPP GDP per capita.
The instruments for constructed FDI measure are the geographical variables.
Lexp: life expectancy at birth
Lexpm: life expectancy at birth for male
Lexpf: life expectancy at birth for female

101

Table26: Mortality (level) and F DI ratio (log), Year 2002

Variable Mortality Mortality Mortality Mortality Mortality Mortality_ Mortality_ Mortality
_young _adult adult young

ln(ActuaIFDl/GDP) 1.49"
0.72
ln( fittedFDl/GDP) 7.90" 646*" 2966*" 22.01 *** 0. 0.3
3. I 3 2. 9.70 9.44 0.44 0.3
GDPPC) -37.46*** -218.72*“

3.08 38. I 3
GDPPC_sq) 10431:"

2.1
med_exp_ratio) -28.37**

14.5
xpenditure no

Income indicators no
FDI no
lations

 

Robust standard errors in brackets.

GDPPC is log PPP GDP per capita.

The instruments for constructed F DI measure are the geographical variables.

Mortality _young: the probability of a child born in a specific year or period dying before
reaching the age of five.

Mortality_adult: the probability of dying between 15 and 59 years.

ln(med_exp_ratio): log level of the total expenditure on health as % of gross domestic
product.

102

Table 27: Life Expectancy (level) and FDI Ratio (10g), Year 2002, With Two-part Model
Including All 170 Countries

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Variable lexp Iexpm lexpf lexp lexp lexp
ln(ActualFDI/GDP) -058...
(0.12)
ln(fittedFDl/GDP) -0.53m -0.38m -0.21 -0.11 -0.46"
(0.23) (0.22) (0.24)l (0.20) (0.16)] (0.62)

ln(GDPPC) 3,41... -6.56m |

(0.42) (1 .01)l
ln(GDPPC_sq) .34...

(0.08)]

 

 

 

Robust standard errors in brackets.
GDPPC is log PPP GDP per capita.

The instruments for constructed FDI measure are the geographical variables.
Lexp: life expectancy at birth
Lexpm: life expectancy at birth for male

Lexpf: life expectancy at birth for female

103

Tab1628: Mortality (level) and FDI Ratio (log), Year 2002, 170 Countries

Variable Mortality _y Mortality 1' Mortality _y Mortality_a Mortality_a Mortality _y Mortality _y Mortality _y
male male female male female male male oung male

ActualFDl/GDP) 3.1 I...

0.76

ln(fittedFDl/GDP) 2.50« 2.24.- 8.05-- 3. 0. 0.41
1.39 1.30 3.36 3.30 1.19 0.96

ln(GDPPC) -2059... 352)...

2.50 6.1
ln(GDPPC_sq) 4.7]...

0.49

ln(med_exp_ratio)

Indicator no
Income indicators n0
FDI no

 

GDPPC is log PPP GDP per capita. The instruments for constructed FDI measure are the
geographical variables. Mortality _young: the probability of a child born in a specific year
or period dying before reaching the age of five. Mortality_adult: the probability of dying
between 15 and 59 years. ln(med_exp_ratio): log level of the total expenditure on health
as % of gross domestic product.

104

BIBLIOGRAPHY

105

BIBLIOGRAPHY

Alsan, M., Bloom, D. & Canning, D.: The Effect of Population Health on
Foreign Direct Investment", NBER Working Papers 10596

Anderson, J. and Marcouiller, D.: "Insecurity and the Pattern of Trade: An
Empirical Investigation", Review of Economics and Statistics, Vol. 84, 2002,
pp.345-352

Anderson, J. and van Wincoop, E.: "Gravity with Gravitas: A solution to the
Border Puzzle", American Economic Review, Vol. 93, 2003, pp.170-192

Anderstan, J. and van Wincoop, E.: "Trade Costs", Journal of Economic
Literature, Vol. 42, 2004, pp.691-751

Arrellano, M. and Bond, S.: "Some Tests of Specification for Panel Data: Monte
Carlo Evidence and an Application to Employment Equations", Review of
Economic Studies, Vol. 58, 1991, pp.277-297

Arrellano, M. and Rover, 0.: "Another Look at the Instrumental Variable
Estimation of Error-Components Models", Journal of Econometrics, 68, 1995, pp.
29-51

Baier, S. and Bergstrand, J: "Do Free Trade Agreements Actually Increase
Members' International Trade?", Journal of International Economics, forthcoming

Baltgagi, B. 11.: Econometric Analysis of Panel Data, Second Edition, New York,
John Wiley

Bernard, A.B., Eaton, J., Bradford, Jensen J., and Kortum, S.: "Plants and
Productivity in International Trade", American Economic Review, Vol. 93, 2003,

pp.1268-1290

Bernard, A.B., Redding, S and Schott, P.: Comparative advantage and
heterogeneous firms, NBER working paper 10668

Biddle, Jeff E. and Hamermesh, D.: "Sleep and the Allocation of Time", Journal
of political Economy, Vol. 95, 1990, pp.922-943

Blundell, R. and Bond, S.: "Initial Conditions and Moment Restrictions in
Dynamic panel data models", Journal of Econometrics, Vol. 87, 1998, pp. 115-143

106

Borensztein, E., De Gregorio, J. and Lee, J.: "How does Foreign Direct
Investment affect economic growth?", Journal of lntemational Economics, Vol. 45,
1998, pp. 115-135

Broda, C. and Weinstein, D: “Variety Growth and World Welfare”, American
Economic Review, Vol. 94, 2004, pp. 139-144

Chaney, T.: "Distorted Gravity: Heterogeneous Firms, Market Structure and
Geography of lntemational Trade", working paper, MIT, 2004

Choi, Y.C., Hummels, D. and Xiang, C: “Explaining import variety and Quality:
the Role of the Income Distribution”, working paper, Purdue University, 2006

Davis, DR. and Weinstein, D.E.: An Account of Global Factor Trade, American
Economic Review, Vol. 91, 2001, pp. 1423-1453

Davis, DR. and Weinstein, D.E.: "Market Access, Economic Geography and
Comparative Advantage: An Empirical Test", Journal of lntemational Economics,
Vol. 59. 2003, pp.1-23

Deardorff, A.: "Determinants of Bilateral Trade: Does Gravity Work in a
Neoclassical World?", in Jeffrey A. Frankel, ed., The Regionalization of the
World Economy, University of Chicago Press, 1998

Dehejia, R. and Lleras-Muney, A.: "Booms, Busts, and Babies' Health",
Quarterly Journal of Economics, Vol. 119, 2004, pp.1091-1130

Dixit, A. K. and Norman, V.: “Theory of lntemational Trade”, Cambridge
University Press, Cambridge, UK.

Dornbusch, R., Fisher, S. and Samuelson, P.: “Comparative advantage, Trade,
and Payments in a Ricardian Model with a Continuum of Goods”, American
Economic Review, Vol. 67, 1997, pp. 823-839

Dow, W. and Norton, E.C.: "Choosing Between and Interpreting the Heckit and
Two-Part Models for Comer Solutions", Health Services & Outcomes Research
Methodology, Vol. 4, 2003, pp.5-18

Duan, N., Manning, W.G., Morris, C.N. and Newhouse, J.P.: "Choosing
Between the Sample-Selection Model and the Multi-Part Model", Journal of
Business and Economic Statistics, Vol. 2, 1984, pp.283-289

Eaton, J. and Kortum, S.: "Technology, Geography and Trade", Econometrica,
Vol. 70, 2002, pp. 1 741-1 779

107

Economou, A., Nikolau, A. and Theodossiou, 1.: "Are Recessions Harmful to
Health After All? Evidence from the European Union", working paper, University
of Macedonia, 2004

Edmonds, E. and Pavcnik, N.: "International Trade and Child Labor: Cross-
Country Evidence", Journal of lntemational Economics, Vol. 68, 2006, pp.115-
140

Edmonds, E. and Pavcnik, N.: "The Effect of Trade Liberalization on Child
Labor", Journal of lntemational Economics, Vol. 65, 2005, pp.401-419

Ettner, S. L.: "Measuring the Human Cost of a Weak Economy: Does
Unemployment Lead to Alcohol Abuse?", Social Science and Medicine, Vol. 44,
1997, pp.251-260

Falvey, R. E. and Kierzkowski, H: “Product Quality, Intra-industry Trade and
imperfect Competition”, in Kierzkowski, H. (ed.), Protection and Competition in
lntemational Trade: Essays in Honor of W. M. Corden, Basil Blackwell, Oxford
and London

Feenstra, R. and Kee, H.L.: “Export Variety and Country Productivity”, World
Bank Policy Research Working Paper 3412, 2004

Feenstra, R., Romalis, J. and Schott, P.K.: “U.S. Imports, Exports, and Tariff
Data, 1989-2001”, NBER Working Paper No. 9387, 2002

Feenstra, R., Yang, T. and Hamilton, G.: “Business groups and product variety
in trade: evidence from South Korea, Taiwan and Japan”, Journal of lntemational
Economics, Vol. 48, 1999, pp.7l-100

Feenstra, R.: "Advanced lntemational Trade: Theory and Evidence", The MIT
Press, 2003

Fobes, J. F. and McGregor, A.: "Unemployment and Mortality in Post-War
Scotland", Journal of Health Economics, Vol. 3, 1984, pp.239-257

Fogel, R.: "The Relevance of Malthus for the Study of Mortality Today: Long-run
Influences on Health, Mortality, Labor Force Participation, and Population
Growth. NBER Working Paper 54, 1994

Frankel, J. and Romer, D.: "Does Trade Cause Growth?", American Economic
Review, Vol. 89, 1999, pp.279-399

Frankel, J. and Rose, A. : "Is Trade Good or Bad For the Environment? Sorting
Out the Causality", the Review of Economics and Statistics, Vol. 87, 2005, pp. 85-
91

108

Frankel, J. and Romer, D.: "Does Trade Cause Growth?”, American Economic
Review, Vol. 89, 1999, pp. 379-399

Funke, M. and Ruwedel, R.: Product Variety and Economic Growth: Empirical
Evidence for the OECD Countries, IMF Staff Papers, Vol. 48, 2001

Gerdtham, U. and Johannesson, M.: "A Note on the Effect of Unemployment
on Mortality", Journal of Health Economics, Vol. 22, 2003, pp.505-518

Hall, R. and Jones, C.: “Why Do Some Countries Produce So Much More
Output Per Worker Than Others?”, Quarterly Journal of Economics, Vol.114,
1999, pp. 83-116

Hallak, J. C.: “Product Quality and the Direction of Trade”, Journal of
lntemational Economics, Vol. 68, 2006, pp. 238-265

Hallak, J. C.: “A Product-Quality View of the Linder Hypothesis”, working paper,
2005

Harrison, A.: "Openness and growth: a time series, cross-country analysis for
developing countries", Journal of Development Economics, Vol. 48, 1996, pp.
419-447.

Helpman, E. and Krugman, P.: Market Structure and Foreign Trade, MIT Press,
Cambridge, 1985

Helpman, E, Melitz, M.J., and Rubinstein, Y.: "Trading Partners and Trading
Volumes", working paper, 2006

Helpman, E, Melitz, M.J., and Yeaple, S.R.: "Export versus FDI with
Heterogeneous F irms", American Economic Review, Vol. 94, 2004, pp.300-316

Holtz-Eakin, D., Newey, W. and Rosen, H. S.: "Estimating Vector
Autoregressions with Panel Data" Econometrica, Vol. 56, 1986, pp. 1 371-95

Hsiao, C. & Shen, Y.: "Foreign Direct Investment and Economic Growth: The
Importance of Institutions and Urbanization", Economic Development and
Cultural Change, Vol. 51, 2003, pp.883—896

Hummels, D. and Klenow, P. J.: “The variety and Quality of a Nation's Trade”,
American Economics Review Vol. 95, 2005, p. 704-723

Hummels, D. and Levinsohn, J.: "Monopolistic Competition and lntemational
Trade: Reconsidering the Evidence", the Quarterly Journal of Economics, Vol.
110, 1995, Pp-799-836

109

Krugman, P. R.: “Increasing Returns, Monoplistic Competition, and International
Trade”, Journal of lntemational Economics, Vol. 7, 1979, pp. 469-479

Krugman, P: "Scale Economics, Product Differentiation, and the Pattern of
Trade", American Economic Review, Vol. 70, 1980, pp.950-959

Krugman, P.: "Increasing Returns and Economic Geography," Journal of
Political Economy, Vol. 99, 1991, pp.483-99,

Learner, E.E.: "The Commodity Composition of lntemational Trade in
Manufactures: An Empirical Analysis", Oxford Economic Papers, Vol. 26, 1974,
pp.350-374

Leung S.F. and Yu S.: "On the Choice between Sample Selection and two-part
Model", Journal of Econometrics, Vol. 72, 1996, pp.197-229

Levine, D. and Rothman, D.: "Does Trade Affect Child Health?", Journal of
Health Economics, Vol. 25, 2006, pp. 538-554.

Liu, Z., Dow, W. and Norton, E.: "Effects of Drive-through Delivery Laws on
Postpartum Length of Stay and Hospital Charges", Journal of Health Economics,
Vol. 23, 2004, pp.129-155

Manning, W.G., Duan, N. and Rogers, W.H.: "Monte Carlo Evidence on the
Choice between Sample Selection and Two-Part Model", Journal of Econometrics,
Vol. 35, 1987, pp.59-82 -

Melitz, M.J.: "The impact of Trade on Intra-Industry Reallocations and
Aggregate Industry Productivity", Econometrica, Vol. 71, 2003, pp.1695-2003

Mullahy, J.: "Much ado about two: Reconsidering retransformation and the two-
part model in health econometrics", Journal of Health Economics, Vol. 17, 1998,
pp.247-282

Olsen M.K. and Schafer J.L.: "A Two-Part Random-Effects Model for
Semicontinuous Longitudinal Data", Journal of the American Statistical

Association, Vol. 96, 2001, pp.730-743

Rauch, J.: "Networks Versus Markets in lntemational Markets", Journal of
lntemational Economics, Vol. 48, 1999, pp.7-35

Rose, A.: "Do We Really Know that the WTO Increases Trade?" American
Economic Review, , Vol. 94, 2004, pp.98-114

Ramondo, N.: "Size, Geography, and Multinational Production", working paper,
University of Texas, 2006

110

Romalis, J.: “Factor Proportions and the Structure of Commodity Trade,
American Economic Review, Vol. 94, 2004, pp. 67-97

Rose, A.: "Do We Really Know That the WTO Increases Trade?", American
Economic Review, Vol. 94, 2004, pp. 98-1 14

Ruhm, C.: "Economic Conditions and Alcohol Problems", Journal of Health
Economics, Vol. 14, 1995, pp. 583-603

Ruhm, C.: "Are Recessions Good for Your Health?", Quarterly Journal of
Economics, Vol. 115, 2000, pp. 617-650

Ruhm, C.: "Good Times Make You Sick", Journal of Health Economics, Vol. 24,
2003, pp. 637-658

Ruhm, C.: "Healthy Living in Hard Times", Journal of Health Economics,
forthcoming

Schott, P.K.: “One size fits all? Heckscher-Olin Specialization in Global
Production”, American Economic Review, Vol. 93, 2003, pp. 686-708

Schott, P.K.: “Across-Product versus Within-Product Specialization in
lntemational Trade”, Quarterly Journal of Economics, Vol. 119, 2004, pp. 647—
678

Smith, J.P.: "Healthy Bodies and Thick Wallets: the Dual Relationship Between
Health and Economic Status", Journal of Economic Perspectives, Vol. 13, 1999,
pp. 145-166

Smith, R. D.: "Foreign direct investment and trade in health services: a review of
the literature", Social Science & Medicine, Vol. 59, 2004, pp. 2313-2323

Stewart, J. M.: "The Impact of Health Status on the Duration of Unemployment
Spells and the Implications for Studies of the Impact of Unemployment on Health
Status", Journal of Health Economics, Vol. 20, 2001, pp. 781-796

Soloaga, Land Winters L. A.: "Regionalism in the nineties. What effect of
trade?" The North American Journal of Economics and Finance, Vol. 12, 2001,
pp.1-29

Tinbergen, J.: "Shaping the world economy: Suggestions for an international
economic policy", The Twentieth Century Fund, 1962

Wooldridge, J.M.: "Econometric Analysis of Cross-Section and Panel Data",
MIT press, Cambridge, MA, 2002

111

   

1liltiljttttttj1111(1))