THREE ESSAYS IN INTERNATIONAL TRADE:
ANALYSIS OF EXPORT DECISIONS AND OFFSHORING
By
Xuan Wei

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Agricultural, Food and Resource Economics – Doctor of Philosophy
Economics – Doctor of Philosophy
2013

ABSTRACT
THREE ESSAYS IN INTERNATIONAL TRADE:
ANALYSIS OF EXPORT DECISIONS AND OFFSHORING
By
Xuan Wei
Trade liberalization through reductions in trade barriers by bilateral and multilateral
agreements boosts individual firm engagement in international trade. The importance of firm
decisions in exploring a new foreign market has been increasingly recognized in terms of
understanding the causes and consequences of aggregate trade flows. Meanwhile, strategic
participation in international trade in turn stimulates the process of global integration, and
impact the diversification of international trade from simple trading goods to more
complicated trade in service through channels such as offshoring. These new features of
international trade cannot be fully explained by pre-existing trade theories.
This dissertation, consisting of three essays focused on the reciprocal relationship
between international trade and firm behaviors, provides additional empirical evidence on the
interactions between these two and contributes to the development of theoretical research
along these lines.
In the first essay, a theoretical model is developed to analyze how an individual firm
may reduce or eliminate the uncertainty of trade compliance costs associated of entering a
foreign market by paying for information to making export decisions. I extend the
heterogeneous firm model of Melitz (2003) to show that in the presence of uncertain
compliance costs and non-zero information costs, average profits and productivity differences
between exporting and non-exporting firms are reduced.
The second essay investigates the effect of information costs and compliance costs
on firm decisions to export as well as how much to export through a hurdle model approach.
A bilateral trade flow data at SITC4-digit industry level from 1991-2000 are used to

approximate the export and value decisions of heterogeneous firms. Results show the effect
of fixed export costs is twofold: Information costs decreases the probability of export by
about five to six percentage points in the first stage. Once the export decision is made, firms
that paid information costs in the first stage tend to export more in the second stage to
compensate such costs. On the other hand, paying compliance costs decreases the probability
of export by about 36 percentage points. Compliance costs are more prohibitive in the
subsequent export value decision.
The third essay generalizes the Grossman and Rossi-Hansberg (2008) offshoring
model to include numerous tasks/skill levels. This generalization allows a possible and direct
linkage between the theoretical task offshoring model and occupational data that can be
aggregated from the CPSMORG (Current Population Survey Merged Outgoing Rotation
Groups) data from year 1983 to 2011. Empirical investigation of the effect of offshoirng on
occupational employment for the ten major occupational groups (at 2-digit SOC level) in the
U.S. labor market is conducted by estimating their offshoring cost functions using a nonparametric monotonic cubic spline interpolation method. Five relatively offshorable
occupational groups are identified from the estimated offshoring cost functions.
The number of jobs offshored and the offshoring percentage for the five relatively
offshorable occupational groups under three scenarios are calculated under NLS (non-linear
least squares) method by attaching a cubic offshoring cost functional form to all five groups.
Results show production occupations are the most offshorable while sales and related
occupations are the least offshorable among all five groups under all three scenarios.
Offshoring percentage for production occupations has been increasing in both pre- and post2000 periods while the offshoring percentages for professional and related occupations, and
management, business, and financial operations occupations have been decreasing over time.

Copyright by
XUAN WEI
2013

For
my mother Yuefen Wei and my father Jinlin Wu

v

ACKNOWLEDGEMENTS

This dissertation is an outcome of training myself to be an independent scholar
with abilities of identifying a question, proposing a method and resolving a problem.
During this process, I have been endowed with tremendous support and assistance from
many people. Without them this dissertation would have not been possible.
First of all, I would like to express the appreciation to my Major Professor and
Dissertation Committee Chair Dr. Suzanne Thornsbury. I especially appreciate her
willingness of continuously serving as my Dissertation Committee Chair even after she
left MSU in 2011. To me, she is not only an academic mentor throughout my doctoral
studies, but also a great friend to consult with concerns and problems as well as to share
moments in my life. She is particularly good at directing students to think through the
fundamental issues of a research. Under her guidance, I learned how to refine research
questions by sorting out different phenomena, and then turn a complicated issue into a
tangible problem.
I am grateful to Dr. David Schweikhardt (Co-chair), who is willing to serve as my
Dissertation Committee Co-chair from a middle point, and other two committee
members, Dr. Susan Chun Zhu and Dr. Jeffery Wooldridge for their deep engagement
and valuable comments to improve my dissertation. I am so fortunate to have such a
balanced committee, each with strengths in different aspects. In particular, I would like
to extend my gratitude to Dr. Zhu for her supervision of shaping my 3rd essay idea and
developing it to a dissertation chapter. I thank Dr. Wooldridge for his tremendous
contributions on choice of econometric methods and models, comparisons between model
vi

specifications and interpretation of empirical results for both my 2nd and 3rd essays. I
would like to thank all my committee members for their tolerance of allowing me to work
from distance and maintain contact with them by emails and phone calls so that I can stay
with my family in Madison, WI.
I would like to thank the Elton R. Smith Endowment funding and the College of
Agriculture and Natural Recourses Dissertation Competition Fellowship for additional
financial support of my dissertation research.
A special thank goes to Dr. Scott Swinton for his strong support in many aspects
including funding arrangement and academic advice. I also thank Debbie Conway, the
Department Secretary for her always-ready-to-help attitude and effective assistance in
dealing with administrative process when I was off campus.
My acknowledgements would be incomplete if I don’t mention help and support
from my fellow colleagues, faculty and staff members at MSU as well as friends at UWMadison. I am especially grateful to, Dr. Sandra Batie, Mollie Woods, Lori Jean, Ge Bai,
Shan Ma, Fang Xia, Min Chen, Hui Wang, Li Cheng, Chenguang Wang, Chaoran Hu,
Tim Komarek, Christina Plerhoples, Yuliang Wang and Jinny Chen.
In addition, a thank you to Dr. Tadashi Yamada, Dr. Nobuyuki Hanaki and Dr.
Mamoru Kaneko at Univeristy of Tsukuba (Tsukuba Daigaku, Japan) who introduced me
to the world of economics and encouraged me to pursue advanced studies in economics.
My mother Yuefen Wei and father Jinlin Wu receive my deepest gratitude and
love for their unconditional support and respect of every choice I make for myself.
Without their constant encouragement, I wouldn’t be where I am today. As I am
struggling through the revision of my dissertation, my mother is fighting against her

vii

sickness. Her positive attitude inspires me and I believe that we together will win the
battle.
I also thank my sister Wei Wu and brother-in-law Weizhong Zhu for their
financial support, for helping me fulfill the responsibility of a dutiful daughter and granddaughter as I am not able to. Sadly, my grandma was not able to wait till the day coming
to share the joy with me, she passed away on February 10, 2013.
Finally, thanks to my husband Xianwei Meng, my daughter Elle Q. W. Meng and
my son William Q. W. Meng. I am so lucky to meet Xianwei, my soul mate at MSU. He
is also such a great partner to discuss my initial research ideas with, to debate about
different theories, and to provide a proofread of my mathematical calculation and
derivation. During the most stressful days of dissertation writing, he always has the
magic power to calm me down and rekindle my hope. Thank you for your faith in me and
always being there for me. Without your company and support, I truly cannot do this.
The first fruit of our marriage, Elle, my sweet daughter, was born on June 27,
2011. As a mom, I beg Elle’s forgiveness for not being able to stay with her and witness
every step of her growth. It was a difficult decision to send Elle back to China so that I
can focus on my dissertation and complete my degree. In this sense, I am deeply indebted
to my parents, especially my mother for taking care of Elle with love. For those days that
I have missed with Elle, I wish I could make up for it with the rest of my life.
My second little one William was born on April 9, 2013. Taking care of a
newborn and revising the dissertation at the same time are frustrating sometimes, but the
joy is overwhelming. I am glad that Xianwei shared the responsibility to looking after

viii

William. I would never forget the scenes that William sleeps comfortably on his daddy’s
chest in the first month.
My dear mom and my little sweetie pies, because of you, every day is full of
meaning to me. Because of you, I am stronger and will be even stronger.

ix

TABLE OF CONTENTS

LIST OF TABLES............................................................................................................. xii
LIST OF FIGURES .......................................................................................................... xiv
INTRODUCTION ............................................................................................................... 1
REFERENCES .................................................................................................................... 6
CHAPTER 1: FIRM LEVEL EXPORT DECISIONS: THE ROLE OF INFORMATION
COST1 ................................................................................................................................. 8
1.1 Introduction ................................................................................................................ 8
1.2. A Conceptual Model ................................................................................................. 9
1.3. Analytical Results ................................................................................................... 15
1.4. Conclusions ............................................................................................................. 20
APPENDIX ....................................................................................................................... 21
REFERENCES .................................................................................................................. 25
CHAPTER 2: AN EMPIRICAL EXAMINATION OF INFORMATION COSTS AND
COMPLIANCE COSTS USING A HURDLE MODEL APPROACH ............................ 27
2. 1. Introduction ............................................................................................................ 27
2.2. Empirical Framework ............................................................................................. 31
2.2.1 Firm Level Decision Process in the Presence of Information and Compliance
Costs ........................................................................................................................... 31
2.2.2 Econometric Model ........................................................................................... 31
2.2.3 The Data ............................................................................................................ 37
2.2.4 Theoretical Assumptions and Restrictions ........................................................ 39
2.2.5. Specification of Estimation Equations for a Representative Firm ................... 43
2.3 Results and Discussion ............................................................................................ 46
2.4 Conclusion ............................................................................................................... 51
APPENDIX ....................................................................................................................... 53
REFERENCES .................................................................................................................. 67
CHAPTER 3: A STRUCTURAL ESTIMATION OF THE EMPLOYMENT EFFECTS
OF OFFSHORING IN THE U.S. LABOR MARKET ..................................................... 70
3.1 Introduction .............................................................................................................. 70
3.2 A Simple Structural Model of Offshoring ............................................................... 74
3.2.1 Model Specification .......................................................................................... 74
3.2.2 Model Derivation .............................................................................................. 75
3.2.3 Model Interpretation.......................................................................................... 78
3.3 Estimation Framework and Method ......................................................................... 79
3.3.1 The Empirical Framework ................................................................................ 79
3.3.2 Application of Monotonic Cubic Spline Interpolation Method ........................ 80

x

3.3.3 Estimating Offshoring Cost Functions for the Ten Major Occupational
Groups ........................................................................................................................ 83
3.3.4 Estimating Number of Jobs Offshored and Offshoring Percentage for the Five
Relatively Offshorable Occupational Groups ............................................................ 86
3.4 Data Description and Adjustment ............................................................................ 87
3.5 Results and Discussion ............................................................................................ 89
3.5.1 Offshoring Costs for the Ten Major Occupational Groups ............................... 89
3.5.2 NLS Results for the Five Relatively Offshorable Occupational Groups .......... 91
3.6 Conclusion ............................................................................................................... 94
APPENDIX ....................................................................................................................... 96
REFERENCES ................................................................................................................ 146

xi

LIST OF TABLES

Table 2.1: Costs of Export Decision at Period t Contingent on Period t-1 and t-2
Status ................................................................................................................................. 42

Table 2.2: Marginal Effects of Compliance Costs and Combined Information and
Compliance Costs .............................................................................................................. 48

Table 2.3: List of 117 Exporting Countries and Areas ...................................................... 54
Table 2.4: List of Top 30 Importing Countries* ................................................................ 55

Table 2.5: Descriptive Statistics of Variables ................................................................... 56

Table 2.6.1: Compliance Costs Effect in Export Decision Equation ................................ 58

Table 2.6.2: Compliance Costs Effect in Export Value Equation .................................... 60

Table 2.7.1: Combined Information Costs and Compliance Costs Effect in Export
Decision Equation.............................................................................................................. 62

Table 2.7.2: Combined Information Costs and Compliance Costs Effect in Export
Value Equation .................................................................................................................. 64

Table 3.1: Offshorablility in Major Occupational Groups ............................................... 84

Table 3.2.1: Major Occupational Groups in Pre-2000 Period (1983-1999) ...................... 97

Table 3.2.2: Major Occupational Group in Post-2000 Period (2000-2011) ...................... 97
Table 3.3: Occupational Employment Size1 Variation ..................................................... 98

xii

Table 3.4: Point Estimates of Parameterized Offshoring Cost Function 𝑡(𝑖) from Cubic
Spline Interpolation Method for Ten Major Occupational Groups ................................... 99
Table 3.5: Estimates of ̂ 𝑜 , 𝑐̂ 𝑜 by Occupational Groups from Cubic Spline
𝐿
Interpolation Method for Major Ten Occupational Groups ............................................ 100
Table 3.6: NLS Estimates of Cubic Offshoring Cost Function ̂ 𝑜 , 𝑐̂ 𝑜 for the Five
𝐿
Relatively Offshorable Occupational Groups.................................................................. 101

Table 3.7.1: Calculated Number of Jobs Offshored and Offshoring Percentage for Group
1 (Management, Business and Financial Operations Occupations) from NLS Method . 102

Table 3.7.2: Calculated Number of Jobs Offshored and Offshoring Percentage for Group
2 (Professional and Related Occupations) from NLS Method ........................................ 103

Table 3.7.3: Calculated Number of Jobs Offshored and Offshoring Percentage for Group
4 (Sales and Related Occupations) from NLS Method ................................................... 104

Table 3.7.4: Calculated Number of Jobs Offshored and Offshoring Percentage for Group
5 (Office and Administrative Support Occupations) from NLS Method ........................ 105

Table 3.7.5: Calculated Number of Jobs Offshored and Offshoring Percentage for Group
9 (Production Occupations) from NLS Method .............................................................. 106

Table 3.8: Scenario Comparison among the Five Relatively Offshorable Occupational
Groups from NLS Method ............................................................................................... 107

Table 3.A: Monotonic Cubic Spline Interpolation Method Preliminary Point Estimates of
Parameterized Offshoring Cost Function 𝑡(𝑖) ................................................................ 109

xiii

LIST OF FIGURES

Figure 1.1 Timeline for firm export decisions ................................................................... 22

Figure 1.2 Minimum productivity level required for export under alternative depictions of
information cost. ................................................................................................................ 23

Figure 2.1.1: Timeline for firm export decisions when fixed export costs are certain ...... 66

Figure 2.1.2: Timeline for firm export decisions when there is uncertainty in fixed export
costs, such as compliance costs ......................................................................................... 66

Figure 3.1: An Example of Cubic Spline Interpolation ..................................................... 82

Figure 3.2.1: Monotonic Cubic Spline Interpolation Method Offshoring Cost Function by
Occupational Groups in Pre-2000 Period (1983-1999) ................................................... 110

Figure 3.2.2: Monotonic Cubic Spline Interpolation Method Offshoring Cost Function by
Occupational Groups in Post-2000 Period (2000-2011) ................................................. 120

Figure 3.3.1: NLS Method Cubic Offshoring Cost Function for G1 (Management,
Business and Financial Operations Occupations) ........................................................... 130

Figure 3.3.2: NLS Method Cubic Offshoring Cost Function for G2 (Professional and
Related Occupations) ....................................................................................................... 132

Figure 3.3.3: NLS Method Cubic Offshoring Cost Function for G4 (Sales and Related
Occupations) .................................................................................................................... 134

Figure 3.3.4: NLS Method Cubic Offshoring Cost Function for G5 (Office and
Administrative Support Occupations) ............................................................................. 136

xiv

Figure 3.3.5: NLS Method Cubic Offshoring Cost Function for G9 (Production
Occupations) .................................................................................................................... 138
Figure 3.4.1: Change of Offshoring Percentage for G1 (Management, Business and
Financial Operations Occupations) ................................................................................. 140

Figure 3.4.2: Change of Offshoring Percentage for G2 (Professional and Related
Occupations) .................................................................................................................... 141

Figure 3.4.3: Change of Offshoring Percentage for G4 (Sales and Related
Occupations) .................................................................................................................... 142

Figure 3.4.4: Change of Offshoring Percentage for G5 (Office and Administrative
Support Occupations) ...................................................................................................... 143

Figure 3.4.5: Change of Offshoring Percentage for G9 (Production Occupations) ........ 144

Figure 3.5: Changes of Offshoring Percentage for the Five Relatively Offshorable
Occupational Groups ....................................................................................................... 145

xv

INTRODUCTION

Despite significant reductions in tariff barriers and revolutionary advances in
transportation and communication technologies, individual firms still face various trade
costs in order to enter a foreign market. In 2004, ad-valorem tax equivalent trade costs for
industrialized countries were estimated to be 170 percent of producer price (Anderson
and van Wincoop, 2004). Included are impacts from tariffs (less than 5 percent), nontariff
barriers (8 percent), and information costs (6 percent), which implies these three sources
were almost equal impediments to trade on an aggregated basis. As tariffs are reduced
through bilateral and multilateral agreements, concern over the substitution of non-tariff
trade barriers has increased. Among them, increased compliance costs (costs which are
necessary to conform with the regulations governing market access) are an obvious
consequence from the proliferation of non-tariff barriers, which are uncertain prior to
export or collection of information which is costly for an individual firm.
In the first essay, the heterogeneous firm model of Melitz (2003) is extended to
analyze how an individual firm may reduce or eliminate the uncertainty of compliance
costs by paying for the information cost prior to making decisions to export. Firms’
ability to eliminate uncertainty over compliance costs depends on their undertaking the
information cost necessary to learn about regulations. Results suggest that in the presence
of uncertain compliance costs and non-zero information cost, average profits and
productivity differences between exporting and non-exporting firms are reduced.
Two separate scenarios are discussed where the presence of positive information
cost will suppress the likelihood of exporting. In the first scenario, firms capable of
1

exporting in the absence of information cost will be limited to the domestic market due to
prior overestimation of compliance costs and lack of incentive to pay the information
cost. As a consequence, firms with high productivity trapped in the domestic market will
increase the overall productivity and profits of domestic firms, (i.e., non-exporting firms).
In the second scenario, firms with low productivity are overoptimistic to pay the
information cost but not able to export due to prior underestimation of compliance costs.
Exit of some low-productivity firms from the domestic market due to the loss of
information cost will also in turn increase the overall productivity and profits of domestic
firms, (i.e., non-exporting firms).
Partially inspired by the theoretical prediction in the first essay, the second essay
attempts to seek empirical evidence of the existence of information costs and compliance
costs using real data. The second essay uses a hurdle model approach to separate the
effect of information costs and compliance costs and empirically examines the effect of
the presence of positive information costs and compliance costs on (a) the likelihood of
an individual firm’s export decision in the first stage, and (b) the export value in the
second stage after the export decision is made.
To capture the two-stage export decision process of an individual firm, a hurdle
model approach is employed. In particular, two model specifications, lognormal hurdle
model and exponential type II Tobit (ET2T) model (or Heckman method) are adopted
and results are compared between these two models. Two indicators are created to infer
and separate the existence of information costs and compliance costs based on an
individual firm’s export status in previous years. Under some theoretical restrictions, a
bilateral trade flow data at SITC4-digit industry level from 1991-2000 are used to
approximate the export and value decisions of heterogeneous firms.

2

A statistical 𝜒 2 test is strongly in favor of the exponential type II Tobit (ET2T)
model (correlations between the two-stage decisions are assumed) against the lognormal
hurdle model (independence between the two-stage decisions are assumed). Results from
the exponential type II Tobit (ET2T) model indicate the effect of fixed export costs is
two-fold. Information costs are crucial in determining whether or not to export. It
decreases the probability of export by about five to six percentage points in the first stage.
Once the export decision is made, firms paying information costs tend to export in the
second stage to cover such costs. On the other hand, paying compliance costs decreases
the probability of export by about 36 percentage points. Compliance costs are more
prohibitive in the subsequent export value decision.
If the first two essays directly evaluate the existence of trade barriers on firms’
participation in international trade, then the third essay investigates the consequences of
firms’ participation through offshoring in international trade as offshoring has spread
rapidly from the jobs of blue-collar workers in manufacturing sectors to those of whitecollar workers in service sectors. Workers in all sectors in most developed countries,
including the United States became more concerned about the security of their jobs as the
global economy continued to integrate.
The third essay generalizes the Grossman and Rossi-Hansberg (2008) offshoring
model to include numerous tasks/skill levels (tasks correspond to specific occupations in
the empirical framework) and then empirically investigate the effect of offshoirng on
occupational employment for major occupational groups (at 2-digit SOC level) in the
U.S. labor market by (a) estimating the offshoring cost functions for the ten major
occupational groups and identifying relatively offshorable occupational groups based on

3

estimated offshoring costs; (b) calculating the number of jobs offshored and offshoring
percentage for relatively offshorable occupational groups.
This research uses the CPSMORG (Current Population Survey Merged Outgoing
Rotation Groups) data from year 1983 to 2011 aggregated at the occupational level. A
non-parametric monotonic cubic spline interpolation method is introduced to approximate
the offshoring cost functions for the ten major occupational groups because standard
parametric methods cannot directly estimate functions. Estimated offshoring costs
demonstrate that among the ten occupational groups, groups involved with more
impersonal and/or routine-tasks have relatively lower offshoring costs in comparison to
groups involved more personal and/or non-routine manual tasks. Included in the former
group are production occupations and office and administrative support occupations.
Occupations in the latter group include farming, fishing, construction, extraction, repair,
and transportation. This finding is consistent with the initial hypotheses and with Blinder
and Krueger (2009) who examined offshorability in major occupational groups based on
a telephone survey.
Motivated by the practical issue of difficulty in obtaining a time-variant
offshoring/offshorability index faced by majority empirical studies interested in
identifying the effect of offshoring, this research calculates the number of jobs offshored
as well as the offshoring percentage by occupation for the five relatively offshorable
occupations over the sample period under three different scenarios. My calculation
indicates that production occupations are most offshorable among all five offshorable
occupational groups in all three scenarios. In addition offshoring percentage for
production occupations has been increasing in both pre- and post-2000 periods while the
offshoring percentage for professional and related occupations, and management,

4

business, and financial operations occupations has been decreasing over time. The
model’s calculation also provide time-variant offshoring indices for more than 300 major
U.S. detailed occupations in these five relatively offshorable groups that can be employed
in other empirical studies.

5

REFERENCES

6

REFERENCES

Anderson, J. E. and E. van Wincoop (2004). Trade costs. Journal of Economic Literature 42
(3), 691–751.

Blinder, A. S. and A. B. Krueger (2013). Alternative measures of offshorability: A survey
approach. Journal of Labor Economics 31 (2), S97–S128.

Grossman, G. M. and E. Rossi-Hansberg (2008). Trading tasks: A simple theory of
offshoring. American Economic Review 98 (5), 1978–97.

Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate
industry productivity. Econometrica 71 (6), 1695–725.

7

CHAPTER 1: FIRM LEVEL EXPORT DECISIONS: THE ROLE OF
INFORMATION COST1

1.1 Introduction
This paper investigates how individual firms eliminate or reduce uncertainty over
compliance costs by paying for information prior to making the decision to export. I
explicitly incorporate such costs into the heterogeneous firm model of Melitz (2003).
Difficulties in accessing information about regulations or exporting procedures could lead to
over- or under-estimation of compliance costs and thus limit export competitiveness well
beyond the content of regulations themselves. Even though regulations and standards are
sometimes publicly accessible through official web sites, it is not an easy or costless task for
firms to go through the tedious, and often obscure, documentation to extract the specific
information they need. Anderson and van Wincoop (2004) estimated the ad-valorem tax
equivalent trade costs for industrialized countries and showed that tariffs, nontariff barriers,
and information costs were almost equal impediments to trade on an aggregated basis.
Increased compliance costs are an obvious consequence from proliferation of nontariff barriers (Thornsbury et al., 2004). A survey across exporting firms in three diverse
industries (telecommunications equipment, dairy products, automotive components) found
that many firms had difficulty determining compliance costs ex ante with part of the
uncertainty related to conformity assessment (for example, inconsistent product evaluation by
assessment bodies) (OECD, 2000). Firms had to make decisions about modifications to meet
anticipated foreign requirements without full information. However, compliance costs are

1

For citation purposes please refer to the article published earlier: Wei, X. and S. Thornsbury
(2012). Firm Level Export Decisions: the Role of Information Cost. Economics Letters 116,
487-90.
8

usually analyzed as if they were certain prior to an export decision (Markus et al., 2001).
Information costs as well as the uncertainty of compliance costs due to lack of information
are not considered and excluded from this line of research.
Melitz (2003) developed a heterogeneous firm model to examine the effect of
compliance costs to export markets on different types of firms, but information cost and
uncertainty over compliance costs were assumed to be nonexistent. This work extends the
Melitz model by explicitly considering these two elements: 1) collecting information is not
free and will be sunk prior to the export decision; 2) compliance cost for each individual firm
is uncertain prior to information collection. Analogous to one-time compliance costs,
information cost is a fixed expense necessary for exporting.

1.2. A Conceptual Model
In this section, the model examines the effect of non-zero information cost on the
likelihood that a firm will export. This model partitions the concept of export cost

the heterogeneous firm model of Melitz (2003) into fixed information cost

compliance costs

 fex  in

 fic  and

 f cc  in the open economy.2

Information cost

 fic  is incurred by a firm in order to participate in foreign

markets. Investment in information collection must be undertaken in a period before the real
fixed compliance costs

 f cc  is revealed; thereafter information costs are sunk (Figure 1).3

2

I retain the Melitz assumptions, derivations and theoretical framework when the economy is
closed.
3
The timing is not altered whether a firm collects all necessary information in-house or hires
9

Prior to paying the information cost

compliance costs

 fic  , firms only have some initial belief that

f cc falls within a certain range. This ex-ante belief about compliance

costs is denoted as a random variable
H B  .

B which follows a distribution hB  on 0,  with cdf

4

For the purpose of comparison, the Melitz modeling strategy of per-period
representation is adopted for the remaining discussion. If firms are assumed to not discount
the future (i.e., discount factor 

time fixed export cost

f ex

 1 ), then firms are indifferent between paying the one

or paying a proportion of this cost in every period. Similarly,

firms are indifferent between paying

f ic

or paying a per-period information cost

f i  f ic  and indifferent between paying B or paying a per-period anticipated
compliance costs 

 B .5 Provided that the real compliance costs  f cc  is revealed after

paying the information cost but prior to export decision, the actual per-period compliance

a broker to provide such service. See footnote 7 for detailed explanation.
4
To be consistent with the Melitz model, firm heterogeneity comes solely from productivity
differences and all firms share the same distribution of beliefs about compliance costs. The
model can be extended to account for heterogeneous belief distributions based on different firm
types by duplicating the current analysis for each type of firm. For example, to obtain unique
productivity cutoffs, it is reasonable to assume beliefs of high productivity firms are first-order
stochastically dominated by those of low productivity firms, i.e., the higher the productivity,
the lower the firm’s belief about compliance costs. In this case all analytical results remain
valid.
5
The anticipated per-period compliance costs 𝛽 = 𝛿В has the same distribution as B.
10

costs paid by a firm is equal to

f c  f cc  . There is no randomness in realization of the

actual per-period compliance costs

fc .

Assuming a continuum of firms with different productivity levels indexed by

where



is distributed as g

,

  with cdf G  ,6 only firms with productivity    

remain in the industry. Hence by construction, a firm seeking to export must have a
productivity level no less than 



.

With the existence of information cost, the profit from domestic sales in every period

   
remains d
 x   

rd  



f

but profit from export is modified to

rx  

export revenues,

 f i  f c , where r   , r   are domestic and
d
x


 is the constant elasticity of substitution, f

production,7 and

fi

and

fc

is the fixed cost for

are per-period information cost and per-period compliance

costs respectively.

6

required for production, is a linear function of output
cost is
7

f  0 . Labor l, the only factor
l  f  q  . This implies marginal

Following Melitz (2003), all firms share the same fixed cost

1

and higher  indicates a higher productivity level.

See footnote 5 for specification of production function.
11

Among firms staying in the industry, the marginal firm which pays information cost

fi

and is actually exporting must have productivity level

  f

*
rx  x



i

 fc  0


x

satisfying

(zero profit condition).

(1.1)

By definition, per-period export cost should equal the information cost plus the compliance
costs, i.e.,

f x  fi  f c .8

Since information about compliance is not costless, the model needs to first consider
which firm will undertake these actions to export. A firm pays information cost and earns the
expected profit denoted as

I   , which is given by:

8

After a broker collected export and compliance information for the first firm, marginal cost
of providing the information to an additional firm approaches zero. Under a competitive
brokerage service assumption, the marginal cost of information collection would then equal

fi  0 . When information cost fi  0 , compliance cost f c
f x  f c , and results revert to the original Melitz solution. In reality,

the price of information

becomes certain,
both brokers and exporting firms consider information proprietary and a potential source of
competitive advantage. Small firms often use brokerage services and pay information cost
between

fi  0 and f i ( f ic ) .

12

rx  
 rd  

I    
f
 fi   h d


 



 rd  
 rx   
 f  fi h d
 fi

 

rx  
 fi

0

(1.2)

 rd  

The first integral is the expected profit from both domestic market 
 


f


and

 rx  

 f i    when the anticipated compliance costs β is less than

foreign market
 


rx  



 fi

and a firm exports. The second integral is the expected profit from domestic

market net information cost when the anticipated compliance costs β is greater than

rx  
 fi


and a firm will not export.

After some simplification,

I    

rx  

0



 fi

rd  
 rx  

  h d 
 f  fi

.

 

(1.2)’

The marginal firm, indifferent between paying the information cost to export or staying in the
domestic market, is the firm with productivity level

ˆ


satisfying the free entry condition

below, such that expected profit from exporting equals domestic profit ,
13

ˆ
ˆ
I     d  

(free entry condition).
(1.3)

From the above free entry condition,

J   is denoted as the difference between

expected profit from exporting and domestic profit for a firm with productivity level

,

J    I     d   
rx  
 fi

0




 rx  
  h d  f i


 
(1.4)

J  

is increasing in

J   



because

rx    rx  
H   
  


 r  
fi   fi h x 

 

rx   0 .
When 


  x , after some simplification

14


f i   0
by


.

 

J  

x





fc

fc

0

 


 rx  x


 fi   h d 
 




fi h d

.

(1.5)
By zero profit condition (equation (1.1)), Equation (1.5) collapses to

 

 f c   h d 
0
f i 1  H  f c 

J  

x

fc

.

(1.6)
Hence, the sign of

  depends on the difference of the two terms in equation (1.6).


J x

The first term is the firm’s expected profits from exporting, while the second term is the
firm’s expected loss from information cost when compliance costs are prohibitive.

1.3. Analytical Results
I distinguish two separate scenarios where the presence of positive information cost
will suppress the likelihood of exporting.

15

Proposition 1. If firms overestimate the real compliance costs a priori (i.e., a high probability


ˆ
x   .
that prior belief about the compliance costs is large), then
Proof:

 


J x  0



fc
1
0  fc   h d  fi
1  H  fc 


fc

H  fc  
h 
 f c  0    d   fi
1  H  fc  
H fc



fc
H  fc  
h   H  fc 
f c  fi
 fc  0    d  
1  H  fc  
H fc  1  H  fc 
In the last line, the first inequality holds because the integral in the bracket is always

positive. Hence, to make the second inequality hold requires
16

H  f c  being small

enough (i.e., there is a very small probability that a firm’s prior belief about the compliance

costs   is less than the real compliance costs

 fc 

, or a firm overestimates the

ˆ
  is increasing in  , I have  x   .


compliance costs). Given J
Q.E.D.

Firms may have little knowledge about the export process (large uncertainty over
compliance costs ex-ante). In this case, information costs will be greater and fewer firms will
have incentive to pay the information cost. A certain number of firms capable of exporting in
the absence of information cost, therefore, are trapped in the domestic market due to prior

overestimation of compliance costs

f c . In comparison to the export cut-off productivity


 x (Figure 1.2a) as originally depicted in Melitz (2003), Figure 1.2b shows that
level
existence of information cost increases the threshold productivity level for exporting from


 x to  . In the presence of information cost, average profit and productivity levels of
ˆ
non-exporters increase, the number of exporters is reduced, and average profit and
productivity differences between exporters and non-exporters are reduced.

17

Proposition 2. If firms underestimate the real compliance costs a priori (i.e., a high
probability that prior belief about the compliance costs is small), then

 


J x  0


ˆ
 x   .9
and

In this case, some firms behave too optimistically and thus more firms than necessary are
paying the information cost but unable to export once compliance costs are known (see

Figure 1.2c). Since firms make their export decision after observing

fc

that becomes

known through paying the information cost, the number of exporters is not altered in this




 x , or  is close
 is close enough to
case. I discuss two extreme cases: either
enough to

rx  



𝜑∗.

In the first case, firms still export because they are better off by exporting if

 f c .10 The Average profit and productivity level of exporters are thus



decreased because firms with


x
close enough to

earn negative profits by exporting.


 close enough to
In the latter case, marginal firms with

9

𝜑 ∗ might have to exit the

Proof of Proposition 2 is parallel to the previous proof with a reverse of the inequality

 


J x  0 .
10

Firms make less negative profits by exporting than remaining in the domestic market.
18

   
industry due to the information cost resulting in d

rd  



 f  fi  0 .

Exit of low productivity firms increases average profits as well as the productivity level of
non-exporters.
I therefore have the following
Proposition 3. Uncertainty over compliance costs which can be mitigated through positive
information cost decreases average profit and productivity gaps between
exporters and non-exporters although the number of exporters may or may not
change.
With positive information cost and uncertainty over compliance costs, the difference in
productivity and average profit between exporting and non-exporting firms is reduced. This is
a new feature not captured in the Melitz standard heterogeneous firm model.

19

1.4. Conclusions
In the presence of positive information cost, ex-ante compliance costs are uncertain
for individual firms when making the decision whether or not to export. Firms’ ability to
eliminate uncertainty over compliance costs depends on undertaking the information cost
necessary to learn about regulations. Two scenarios derived from my theoretical model
predict that profit and productivity differences between exporters and non-exporters are
reduced in industries subject to more stringent regulations where ex-ante compliance costs
tend to be more uncertain for individual firms without prior costly information collection.

20

APPENDIX

21

Figure 1.1: Timeline for firm export decisions

incur
not
incur information reveal of
cost

export
decision

compliance
costs

Note: All notations in this figure are expressed in per-period representation.

22

export

t

Figure 1.2: Minimum productivity level required for export under alternative depictions
of information cost.

f x : Melitz (2003)

1.2a. Firms facing per-period fixed export cost

*



*
ˆ
 x

1.2b.Firms facing uncertain compliance costs

fc

prior to paying information cost

fi

(overestimate)
firms trapped in domestic
market due to information
cost

0



*

x
*

23

ˆ


exporters



Figure 1.2 (Cont’d)
1.2c.Firms facing uncertain compliance costs

fc

prior to paying information cost

fi

(underestimate)
firms paid
information cost, but
not able to export

*

0

ˆ


exporters



 *x

Note: Along the horizontal productivity line,  is firm productivity.  and  x
define the minimum productivity levels for firms to remain in the industry and to export


ˆ




respectively.
defines the productivity level for firms that pay information costs and
consider exporting. All notations in this figure are expressed in per-period representation.

24

REFERENCES

25

REFERENCES

Anderson, J. E. and E. van Wincoop (2004). Trade costs. Journal of Economic Literature
42 (3), 691–751.
Maskus, K. E., T. Otsuki and J. S. Wilson. “An empirical framework for analyzing
technical regulations and trade,” in Quantifying the Impact of Technical Barriers to
Trade, ed. Keith E. Maskus and Jone S. Wilson (Ann Arbor : University of Michigan
Press, 2001), 29.

Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate
industry productivity. Econometrica 71 (6), 1695–725.

Organization for Economic Development and Cooperation. An Assessment of the Costs
for International Trade in Meeting Regulatory Requirements. Technical Report.
TD/TC/WP(99)8/FINAL, OECD, Paris, 2000.

Thornsbury, S., D. Roberts, and D. Orden (2004). Measurement and political economy of
disputed technical regulations. Journal of agricultural and applied economics 36 (3),
559-74.

26

CHAPTER 2: AN EMPIRICAL EXAMINATION OF INFORMATION COSTS AND
COMPLIANCE COSTS USING A HURDLE MODEL APPROACH

2. 1. Introduction
Upon export to a new foreign market, firms seeking to export face per-unit export
costs such as tariff and transport costs, but also face fixed costs that do not vary with export
volume.1 Two typical fixed export costs faced by a new exporting firm are i) Investment in
research to understand the regulations and standards in a potential foreign market (herein
called information costs); ii) product redesign for a specific market, establishing new
processes or procedures to comply with foreign regulations and standards (herein called
compliance costs).2
Several studies have focused on using firm level data to validate the existence of
fixed export costs and quantify the effect on firm participation in foreign markets in both the
extensive margin (i.e., the number of exporting firms) and intensive margin (i.e., volume of
each exporting firm). For Columbian manufacturing firms, Roberts and Tybout (1997) found
that (a) fixed export costs significantly reduce the probability of a firm exporting and (b) a
firm with prior experience is far more likely to export than a firm that has never exported.
Bernard and Jenson (2001) showed that entry costs to foreign markets are substantial for U.S.
manufacturing firms, and firms are increasingly likely to export in consecutive years. Das et
al. (2007) quantified the fixed entry costs for three Colombian manufacturing industries

1

In some international trade literature, fixed export costs are called fixed entry costs, sunk
costs or sunk entry costs.
2
Some compliance costs can be repeatedly occurring for continuous quality control and
testing certification even if a firm keeps exporting to the same market.
27

(basic chemicals, leather products, and knitted fabrics). Their results indicated that average
fixed entry costs to foreign markets were similar across the three manufacturing subsectors
between 1981 and 1991, but were lower for large producers (e.g., $402,000 for knitting mills)
relative to small producers (e.g., $412,000 for knitting mills). These studies estimated how
the existence of fixed export costs affected the extensive margin of trade flow (i.e., how the
existence of fixed export costs affect the likelihood of an individual firm to export). Helpman
et al. (2008) estimated the effect of fixed export costs on both extensive and intensive
margins (i.e., the number of exporting firms and the trade volume per exporting firm
respectively) in a theoretical model accounting for both decisions (self-selection of firms into
the export markets and firms’ export volume) where fixed export costs were assumed to be
known.
Two major assumptions in the existing studies that examining the effect of fixed
export costs have not been addressed. First, fixed export costs (including the theoretical
heterogeneous firm model of Melitz, 2003) are treated as certain when firms make export
decisions (Figure 2.1.1). A second issue that most current empirical studies in this field have
not addressed is differentiation among export destinations. Compliance costs are often
uncertain due to a lack of information prior to the decision of export (Figure 2.1.2). In the
presence of positive information costs and uncertainty of compliance costs, the number of
exporting firms is smaller than that reported by Melitz (2003) where fixed export costs are
assumed to be certain. When compliance costs are uncertain, some firms capable of
exporting may overestimate compliance costs without collecting information and thus are
trapped within the domestic market.
An individual firm’s sequential decision to export can be described by a standard
hurdle model often used in empirical settings. In the first stage, an individual firm determines
28

whether or not to export to a foreign market providing total fixed export costs are known.
Once the export decision is made, the individual firm determines how much to export in the
second stage. However, when there are positive information costs and compliance costs that
are not revealed until after an individual firm pays the information costs the first stage export
decision is contingent on previous export status, i.e., whether a firm is new to this specific
foreign market in period 𝑡 so that collecting information is required prior to export.
Blanes-Cristóbal (2008) empirically tested and confirmed the differences of fixed
export costs across markets, but they only divided export destinations into three general
market areas: EU, OECD and ROW. Morales et al. (2011) developed a structural model and
estimated the fixed entry cost for chemical and chemical products manufacturing firms using
Chilean data. Their results show country-specific entry costs increases significantly as the
destination country is farther away and less similar from the exporting country. Simply
considering the export decision as a single choice of whether or not to export regardless of
destination obscures important market specificity characteristics that define fixed export
costs. Although a firm’s status as an exporter tends to be persistent, an exporting firm may
frequently enter a new foreign market or exit a current export market. Selection of
destination countries depends not only on similarities between a new export market and the
firm’s home country, but also on similarities between a new market and previous destinations
(Morales et al., 2011). Thus, a firm’s export decision should consider the question of whether
or not to export to a specific foreign market because information costs and compliance costs
are market specific and can vary substantially depending on (a) which foreign market is
selected for expansion and (b) which export market is already served by the firm. It is thus
necessary to differentiate export destinations to clarify fixed export costs as a part of
quantifying these costs on firm export decisions.

29

In this paper, the effect of information costs and compliance costs are decomposed to
empirically examine how their presence may affect (a) the likelihood of a positive individual
firm export decision in the first stage, and (b) the export value in the second stage after the
export decision is made in a hurdle model. Panel bilateral trade flow data at SITC4-digit
industry level from 1991-2000 are used to approximate the export decision and export value
decision of heterogeneous firms. Two indicators are created to infer the existence of
information costs and compliance costs based on export status in previous years, as neither
information costs nor compliance costs are directly observed in the data. In addition, export
destinations are distinguished when examining the effect of information costs and compliance
costs. Market heterogeneity provides the basis to determine fixed export costs and to quantify
their effects.
A hurdle model approach using firm-level data can separately identify how the
existence of information costs and compliance costs may affect not only the extensive margin
(i.e., the number of exporting firms), but also the intensive margin (i.e., the trade volume per
exporting firm) of exporting firms. With industry-level data, the heterogeneity of exporting
firms is represented by defining each SITC4-digit group in the form of “a representative
firm”.3 Empirical results capture the estimated effects of information and compliance costs
on decisions about industry export to differentiated markets.
The results show that the effects of fixed export costs are twofold. First, information
costs are crucial in determining whether or not to export. Such costs decrease the probability
of export by about five to six percentage points in the first stage. Once the export decision is
made, firms tend to compensate information costs by exporting more in the second stage. On

3

See Section 2.2.3 for discussions about how to fit industry level data to the firm level
decision model.
30

the other hand, paying compliance costs decreases the probability of export by about 36
percentage points. Compliance costs are more prohibitive in the subsequent export value
decision.

2.2. Empirical Framework
2.2.1 Firm Level Decision Process in the Presence of Information and Compliance Costs

When information costs are assumed to be zero and compliance costs are certain and
known, the individual firm decision to enter a foreign market can be depicted as a simple
two-step process (Figure 2.1.1). In time period 𝑡, a firm decides “whether or not to export” in
the first stage 𝑡1 (defined as the first sub period of 𝑡) given fixed export costs. In the second
stage 𝑡2 (defined as the second sub period of 𝑡) a firm decides how much to export. When
collecting information is costly and compliance costs are known only after a firm collects the
information, the export decision of an individual firm is still a two-step decision but with
some modification in the first-stage 𝑡1 as assumptions that fixed export costs are certain and
known are relaxed. The decision of whether or not to export at time period 𝑡 is now
contingent on whether a firm must pay information costs and/or compliance costs in the first
stage 𝑡1 of time period 𝑡 (Figure 2.1.2).
2.2.2 Econometric Model
Several approaches in the literature are used to model export decision and bilateral
trade between countries. The most straightforward method to model individual firm export
decision is the probit model (e.g., Roberts and Tybout, 1997; Bernard and Jenson, 2001). A
log-linearized gravity equation is often used to predict the volume of trade for an individual
31

exporting firm or the bilateral trade flow between countries. However, there is increasing
evidence showing that zero trade flows are commonly observed in international trade.
Haveman and Hummels (2004) find that nearly 1/3 of the bilateral trade matrix is empty.
Helpmen et al. (2008) find that about half of the country pairs in their sample do not trade
with each other. The problem is expected to be more serious at disaggregated firm level data
when exporting zero volume is often an optimal choice. Under the log-linearized
specification, taking logarithm effectively drops zero observations from the sample and is
likely to produce biased estimates by getting rid of useful information. A commonly used
empirical approximation is to add a small positive number (such as 0.0001) to all zero trade
flows. This is sensible to see how including or excluding zeros make much of difference
empirically, but has no theoretical basis.
A common econometric approach for dealing with corner solution (when some trade
flows are zeros piling up at corner while others are strictly positive values) is the Tobit model
(Tobin, 1958). Tobit model has been commonly applied in dealing with bilateral zero trade
flows (e.g., Eaton and Tamura (1994)). 4

In most recent empirical literature, in order to be consistent with firm heterogeneity
theory (Melitz, 2003; Helpman et al., 2008), a two-step decision is often modeled in a
Heckman procedure to account for bias that productive firms are self-selected into export
market.

4

Tobit model is not feasible in our case because the normality assumption required in the
tobit model is not valid due to extremely large trade values observed in the data set.
32

In this paper, a lognormal hurdle model following Wooldridge (2010) is used to
model the two-stage decision of firm participation in a foreign market.5 The econometric
model is parameterized as follows.
Let 𝑦 be the export value chosen by an individual firm, which is a compound
function of a binary participation decision variable 𝑠, and the continuous choice of a
nonnegative export value

𝑦∗.

𝑦 = 𝑠 ∙ 𝑦∗

(2.1)

When a firm decides to export to a specific foreign destination (𝑠 = 1), a
nonnegative export value

𝑦 ∗ = 𝑦 is observed. On the other hand, when a firm decides not

to export (𝑠 = 0), then 𝑦 = 0 and

𝑦 ∗ is not observed.

The first-stage binary decision variable 𝑠 is assumed to follow a probit model.
1

if a firm overcomes the hurdle of fixed export costs

𝑠={

}
0

(2.2.1)

otherwise
Let 𝑠 ∗ be the latent variable indicating unobserved firm ability to overcome market-

specific fixed export costs which prevent a firm from participating in a foreign market. An
individual firm will export only if it is able to overcome the hurdle of fixed export costs
determined by a vector of attributes 𝒙1 such as firm characteristics including productivity

5 The econometric model is directly formulated in a lognormal hurdle model because the

trade values jump to large numbers after zeros at corner which fits better into a lognormal
distribution than a truncated normal distribution. For details, see Section 17.6.2 in
Wooldridge (2010).
33

and profit margins, compliance requirements in importing countries, etc. The observed
attributes

𝒙1 is independent of the error term

𝑠 ∗ = 𝒙1 𝜸 + 𝑣

𝑠={

∗

0

𝑐𝑜𝑣 ( 𝒙1 , 𝑣) = 0.

𝑣~𝑁𝑜𝑟𝑚𝑎𝑙 (0, 1)

𝑠∗ > 0

1

𝑣, i.e.,

(2.2.2)

}

(2.2.3)

𝑠 ≤0

𝑃( 𝑠 = 1|𝒙1 ) = 𝐸 ( 𝑠|𝒙1 ) = Φ( 𝒙1 𝚼).
The continuous export value of

(2.2.4)

𝑦 ∗ is then chosen based on a vector of attributes 𝒙2

such as firm characteristics, trade barriers, etc. Further,

𝑦 ∗ is assumed to have a lognormal

distribution (see footnote 6),

𝑦 ∗ = 𝑒𝑥𝑝( 𝒙2 𝜷 + 𝑢) > 0 𝑢~𝑁𝑜𝑟𝑚𝑎𝑙 (0, 𝜎 2 )

𝑦={

𝑦∗ > 0
0

𝑖𝑓

(2.3.1)

𝑠=1
}.

(2.3.2)

𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Or,

𝑦 = 1[ 𝒙1 𝜸 + 𝑣] 𝑒𝑥𝑝( 𝒙2 𝜷 + 𝑢) > 0.

(2.4)

Let 𝜌 be the correlation between 𝑣 and 𝑢. Under the assumption that the binary
export decision is independent of the export value decision conditioning on observed
variables 𝒙1 and

𝒙2 ,

𝜌 = 0. As

𝑦 = 𝑦 ∗ when

density of 𝑦 when 𝜌 = 0 is,
34

𝑦 > 0, the expression of the conditional

𝑓( 𝑦|𝒙2 , 𝑦 > 0) = 𝜙 [( 𝑙𝑜𝑔( 𝑦) − 𝒙2 𝜷)⁄ 𝜎]⁄( 𝜎𝑦) , 𝑦 > 0.
(2.5)
The unconditional density of 𝑦 given

𝒙1, 𝒙2 is straightforward by multiplying

𝑃( 𝑦 > 0|𝒙1 ) = Φ( 𝒙1 𝚼),
𝑓(𝑦|𝒙1, 𝒙2 ) = [1 − Φ( 𝒙1 𝜸)]1[𝑦=0]
{Φ( 𝒙1 𝜸) 𝜙 [( 𝑙𝑜𝑔( 𝑦) − 𝒙2 𝜷)⁄ 𝜎]⁄( 𝜎𝑦)}1[𝑦>0] .
(2.6)
For a random firm i, the associated log-likelihood function to be estimated is

𝑙 𝑖 ( 𝜽) = 1[ 𝑦 𝑖 = 0]log[1 − Φ( 𝒙 𝑖 𝜸)] + 1[ 𝑦 𝑖 > 0]log[Φ( 𝒙 𝑖 𝜸)]
+1[ 𝑦 𝑖 > 0]{log{ 𝜙[(log( 𝑦 𝑖 ) − 𝒙 𝑖 𝜷)⁄ 𝜎]} − log( 𝜎) −
log( 𝑦 𝑖 )}.

(2.7)

The conditional and unconditional expectations of 𝑦 then can be derived,

𝐸 ( 𝑦| 𝒙2 , 𝑦 > 0) = 𝐸 ( 𝑦 ∗ |𝒙2 , 𝑠 = 1) = 𝑒𝑥𝑝( 𝒙2 𝜷 + 𝜎 2 ⁄2).
(2.8)

𝐸 ( 𝑦| 𝒙1 , 𝒙2 ) = Φ( 𝒙1 𝚼) 𝑒𝑥𝑝( 𝒙2 𝜷 + 𝜎 2 ⁄2).

(2.9)

The lognormal hurdle model assumes the binary choice decision (𝑠) of whether or
not to export is independent of the export value decision as long as the attributes which
35

determine these two decisions can be controlled through observable firm characteristics. It is
however possible that some unobserved factors that affect an individual firm’s export
decision (

𝑠) are still affecting the export value decision ( 𝑦 ∗ ). Hence the independence

assumption is relaxed and an exponential type II Tobit (ET2T) model is developed to allow
conditional correlation between

𝑠 and 𝑦 ∗ .6

This implies the correlation 𝜌 between 𝑣 and 𝑢

is not zero and the variance-covariance matrix for 𝑣 and 𝑢 is (

1
𝜌𝜎

𝜌𝜎
). In order to
𝜎2

identify 𝜌, an exclusive restriction is needed so that the covariates determining the export

𝒙2 is a

decision strictly contain those affecting the export value decision. In other words,

strict subset of 𝒙1 . When 𝜌 is non-zero, the conditional density of 𝑦 given 𝒙1 can be
derived,

𝑓( 𝑦|𝒙1 , 𝑦 > 0) = Φ([ 𝒙1 𝜸 + ( 𝜌⁄ 𝜎)( 𝑦 − 𝒙2 𝜷)](1 − 𝜌2 )−1⁄2 )
𝜙 [( 𝑙𝑜𝑔( 𝑦) − 𝒙2 𝜷)⁄ 𝜎]⁄( 𝜎𝑦).
The unconditional density of given

(2.10)

𝒙1, 𝒙2 is

𝑓(𝑦|𝒙1, 𝒙2 ) = [1 − Φ( 𝒙1 𝜸)]1[𝑦=0]
{Φ([ 𝒙1 𝜸 + ( 𝜌⁄ 𝜎)( 𝑦 − 𝒙2 𝜷)](1 −
1[𝑦>0]

𝜌2 )−1⁄2 )𝜙 [( 𝑙𝑜𝑔( 𝑦) − 𝒙2 𝜷)⁄ 𝜎]⁄( 𝜎𝑦)}

6

See details in Section 17.6.2 in Wooldridge (2010).
36

.

(2.11)
and the associated log-likelihood function to be estimated is,

𝑙 𝑖 ( 𝜽) = 1[ 𝑦 𝑖 = 0]log[1 − Φ( 𝒙 𝑖 𝜸)]
+1[ 𝑦 𝑖 > 0]{log[Φ([ 𝒙 𝑖 𝜸 + ( 𝜌⁄ 𝜎)(log( 𝑦 𝑖 ) − 𝒙 𝑖 𝜷)])(1 − 𝜌2 )−1⁄2 ]
+ log{ 𝜙[(log( 𝑦 𝑖 ) − 𝒙 𝑖 𝜷)⁄ 𝜎]} − log( 𝜎) − log( 𝑦 𝑖 )}.
(2.12)7

2.2.3 The Data
The major data used in the hurdle model are panel bilateral trade flow data set,
selected from “World Trade Flows: 1962-2000” compiled by Feenstra and Lipsey.8 The
most recent 10 years (1991-2000) were selected as the analysis period and to be consistent
with data information on regulation variables and trade barrier variables used in the empirical
model.9 Trade flows are recorded at the 4-digit SITC level using information from the
importing countries’ data sources wherever they are available for each importing-exporting
country pair.10 Under each particular 4-digit SITC code, trade values are summed to obtain
the total trade flow corresponding to a specific SITC 4-digit product for each country pair in

7 For detailed derivation of Eq. (2.10), (2.11) and (2.12), please see Chapter 17 in

Wooldridge (2010).
Available for download from http://cid.econ.ucdavis.edu/.

8
9

Data on regulation variables are only available for 1999. Historical data for trade variables
were not available and data for 2009 is used.
10
Exporting country’s information is used instead only when the importing country’s report
is not available when the data are compiled.
37

each year.11 The 111A and 111X 4-digit categories for all country pairs and all years are
deleted to construct a consistent SITC4-digit panel. 12

Regulation variables measure how costly it might be for a domestic firm to meet all
legal requirements before it can enter an industry or operate a new business (Djankov et al.
(2002)). Regulations imposed by an importing country affect not only domestic firms but also
foreign exporting firms seeking to enter the market. These regulations affect fixed export
costs of entry but not variable costs of repeated export. In the empirical model, regulation
variables include (a) the number of legal procedures, (b) number of days, and (c) relative
official costs (as a percentage of GDP per capita) required for a new exporting firm to legally
operate a business in the importing country. These three variables satisfy the exclusive
restriction and can be used to measure the restrictiveness of entry to an importing country and
to predict the likelihood of export for a foreign exporting firm in the first stage.
On the other hand, trade barriers imposed by an importing country are applied to
foreign exporting firms only, which affect not only fixed export costs, but also variable costs
of foreign exporting firms. Trade facilitation indices include a Logistics Performance Index
(a scale number of 1-5 with 1 indicating the worst performance), a Burden of Customs
Procedures (a scale number of 1-7 with 1 indicating the worst, i.e., the largest burden), a Lead
Time measure (number of days delayed in the importing countries) and a Document measure
(number of documents required by the importing countries in order to export) in importing

11

The data reports three forms of trade value: (1) value (in thousand US dollars), (2) value
with unit of number and (3) value with unit of weight.
12
The 111A and 111X 4-digit categories are artificially created by Feenstra and Lipsey to
capture missing or miscellaneous 4-digit trade flows for certain years so that the aggregation
of all 4-digit SITC codes would equal to the value of a higher 3-digit SITC code. With no
detailed information on countries or products included in 111A or 111X, the categories are
inconsistent between years.
38

countries in year 2009 (Word Development Indicators, 2011). A measure of tariff rate of
importing countries is drawn from Kee et al. (2009). This overall tariff rate measure is a
weighted sum of all tariff lines over all goods estimated Kee et al. (2009) using tariff data
between 2000 and 2004.
Other variables are adopted from Helpman et al. (2008), based on factors commonly
used in gravity models to explain bilateral trade flows.13 These variables include country
characteristics (i.e., GDP per capita for importing and exporting countries), geographic
variables (i.e., distance between the importing-exporting country pairs, whether importing
and exporting countries share a common border, whether the exporting country is land locked
or an island), institutional variables (i.e., whether the country pair shares the same legal
system, shares the same colonial origin, whether both countries are members of WTO) and
cultural variables (i.e., whether the country pair has a similarity in religion composition,
speak the same primary language).
The compiled data set includes more than 13 million observations representing ten
years of bilateral trade flow data between 117 importing countries and 117 exporting
countries. To focus the analysis on top importing markets with large import demands, the
model incorporates trade flows between the 117 exporting countries (Table 2.2) and only the
30 largest importing markets (Table 2.3).
2.2.4 Theoretical Assumptions and Restrictions
Several theoretical assumptions and restrictions are made to use the aggregate
industry-level SITC4-digit data in a firm-level export decision model. First, a representative
firm assumption is applied. Firms are assumed to be homogeneous within each SITC4-digit

13

Data set available at http://scholar.harvard.edu/melitz/publications.
39

group and make the same export decisions when facing country-specific fixed information
costs and compliance costs. The STIC4-digit sector within a country is thus representative of
one firm. As there are 900 SITC4-digit codes in the data set, there are 900 representative
firms for 900 different types of industries.
Second, the representative firm is assumed to only produce product(s) within the
SITC4-digit group it represents. For example, a textile firm may produce and export both
curtains and bedspreads, but these two products belong to the same SITC 4-digit code. On the
other hand, if a machine manufacturing firm produces and exports both milling machines and
other food processing machineries that belong to two different SITC 4-digit codes, these two
product lines are assumed to be operated independently and thus are treated as if they were
two different firms. This ensures the validity of aggregating trade flow within each SITC4digit group.
Third, the follower assumptions are imposed for information costs and compliance
cost to distinguish information costs from compliance costs as they cannot be observed in the
data set. Information updates are costless as long as a firm remains active in the export
destination, but information becomes obsolete if the firm exits the market for two consecutive
years. Therefore, as long as a firm is exporting to a specific destination in at least one of the
last two periods, no information costs are required to reenter the same market in period 𝑡.
Information costs are relatively low cost to update as long as a firm remains actively
participating in a specific foreign market.14 Once information about a particular market is
collected it is known by the firm.

14

This research treats an exporting firm as active exporting firm in a foreign market if a firm
continues to export to the same market in consecutive years or exits the market for only one
year due to temporary shocks but reenters the market in the following year.
40

Once a firm exits a specific foreign market in period 𝑡 − 1, it must re-pay
compliance costs upon reentry into the market in period 𝑡. In reality some compliance costs
are variable and must be incurred in each period as long as export continues. For example,
after establishment of a processing line or acquiring certification for testing, packaging or
labeling as part of the fixed export costs, compliance costs may vary with output as well such
as the labor and materials for testing, labeling or packaging carried out in each period. Or, if
investing a new processing line or obtaining a certain certificate for export requires a large
expenditure, a firm may pay a proportion of compliance costs in each period as principle and
interest in each future period on an initial loan.
The model assumes fixed export costs, including information costs, compliance
costs, set-up of local distribution channels, etc., are market dependent. Previous or current
experiences of exporting to a foreign market j neither reduces the information costs nor the
compliance costs for an individual firm to enter another foreign market 𝑖 in period 𝑡.
Research regarding foreign demand, regulations, and standards in a potential foreign market
must be conducted prior to entry into a new foreign market. Product is subject to
modifications in order to meet the demands of individual markets as well as to comply with
the foreign standards. Empirical results show that experience in one market is not relevant or
does not increase the probability of exporting to another market (Blanes-Cristóbal, 2008).
Following these assumptions, the information and compliance costs faced by an
individual firm to export to a specific foreign market in period 𝑡 is summarized in Table 2.1.

41

Table 2.1: Costs of Export Decision at Period t Contingent on Period 𝑡 − 1 and 𝑡 − 2 Status
Group
……
……
𝑡−2
𝑡−1
𝑡
I
II
III
IV

export

export

no information cost,
no compliance costs

not export

export

no information cost,
no compliance costs

export

not export

no information cost,
compliance costs

not export

not export

information cost,
compliance costs

Based on the above four groups, two binary indicators are used to capture the effect
of compliance costs as well as information costs. As firms in Group I and II pay neither
information costs nor compliance costs while firms in Group IV must pay both information
costs and compliance costs in order to export in period 𝑡, the combined effect of both
information costs and compliance costs can be identified by generating a binary variable
both_Index, which equals one if a firm belongs to Group I or II, and zero if a firm belongs to
Group IV. Meanwhile, firms in Group III pay no information costs but only compliance
costs while firms in Group I and II must pay neither information costs nor compliance costs
in order to export in period 𝑡, the effect of compliance costs can be identified by generating a
binary variable Comp_Index, which equals one if a firm belongs to Group III, and zero if a
firm belongs to Group I or II. The difference between both_Index and Comp_Index captures
the effect of information costs.

42

2.2.5. Specification of Estimation Equations for a Representative Firm
With these above mentioned theoretical restrictions, the aggregate bilateral SITC4digit industry-level trade flow data will fit into a representative firm’s two-stage decision
equations.1
In the first stage a probit regression defines firm decision to enter individual export
market.

Pr(𝐸𝑋𝑃𝑂𝑅𝑇 𝑓𝑖𝑗𝑡 = 1|𝒙1 ) = Φ(𝛾0 + 𝐼𝑛𝑑𝑒𝑥 𝑓𝑖𝑗𝑡 𝛾1 +
𝐺𝑟𝑎𝑣𝑖𝑡𝑦 𝑖𝑗𝑡 𝜸2 +𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑖 𝜸3 + 𝑇𝐵 𝑖 𝜸4 )

(2.13)

Conditional on a firm export results from the first stage, an OLS regression for observations
with positive export values only is run in the second stage to determine level of export.

log𝑇𝑉𝑓𝑖𝑗𝑡 = 𝛽0 + 𝐼𝑛𝑑𝑒𝑥 𝑓𝑖𝑗𝑡 𝛽1 + 𝐺𝑟𝑎𝑣𝑖𝑡𝑦 𝑖𝑗𝑡 𝜷2 + 𝑇𝐵 𝑖 𝜷3 + 𝑢 𝑓𝑖𝑗𝑡 .
(2.14)

The dependent variable EXPORT in the probit regression is a binary variable,
corresponding to one if a representative individual firm 𝑓 in country 𝑗 exports to importing
country 𝑖 in year 𝑡, and equal to zero otherwise. The dependent variable log𝑇𝑉 in the
lognormal regression is the log of trade value of SITC4-digit product 𝑓of representative firm
𝑓 in exporting country 𝑗 to importing country 𝑖 in year 𝑡 if an export decision is made in the
first stage (Equation 2.13). Based on the single-product firm assumption, the subscript f
stands for both a representative exporting firm and a SITC4-digit product it exports

1 The same estimating equations could be used for heterogeneous firms if firm-level data

were available.
43

Index aims to pick up the effects of information costs and compliance costs on the
likelihood of export in the first stage (Equation 2.13) and subsequent export value decision in
the second stage (Equation 2.14). Index represents two variables: Comp_Index (measures the
effect of compliance costs in the model) and Both_Index (measures the combined effect of
both information and compliance costs in the model). By construction, Comp_Index is a
binary variable to capture the effect of compliance costs by comparing firms paying no
information costs but only compliance costs (Group III) with firms paying neither
information cost nor compliance costs (Group I and II). Both_Index is a binary variable to
capture the effect of both information costs and compliance costs by comparing firms paying
both information cost and compliance costs (Group IV) with firms paying neither information
cost nor compliance costs (Group I and II). The difference between the effects of
Comp_Index and Both_Index obtained separately from the lognormal hurdle model captures
the effect of information costs. As Both_Index contains the effect of both information costs
and compliance costs, we expect to observe a larger negative impact of Both_Index than
Comp_Index which contains the effect of compliance costs only.
Regulation is a vector of regulation cost variables indicating the restrictiveness of
starting a new business for a domestic firm in importing country 𝑖. The three variables are
number of legal procedures (procedure), number of legal days (time), and relative official
costs, as a percentage of GDP per capita (cost). Since both foreign and domestic firms face
these same regulations, they affect a foreign firm’s decision on whether to enter the foreign
market but not on how much to export. Therefore, the three variables in the Regulation vector
are used as exclusion restrictions in exponential type II Tobit (ET2T) model to identify 𝜌 and
test whether the two decisions are correlated. Given that these regulatory hurdles are faced by
an exporting firm in order to start a new business in an importing country, an increase in

44

number of legal procedures or number of legal days, or an increase in relative official costs, is
expected to decrease the likelihood of export for an exporting firm.
On the other hand, a foreign firm faces additional hurdles to export such as tariff and
non-tariff barriers imposed by importing countries. Hence, the vector of trade barrier (TB)
variables is included to control for these restrictions. Four trade facilitation indices, logistics,
customs, lead_days and documents are used to measure the importing country’s overall
openness. Improvement in logistics and customs (i.e. a larger logistics and customs index
because logistics and customs are scaled numbers with a large number indicating a better
service) encourage export while delays in time and increases in the number of required
documents discourage exports. The variable OTRI_tariff created by Kee et al. (2009), a
measure of the overall trade restrictiveness of tariff, controls for tariff rate in each importing
country. Increase in tariff rate should decrease both export probability and export value.
Gravity is a vector of country-pair specific variables includes: distance between
importing and exporting countries (log(GDP_EX)), whether importing and exporting
countries share a common border (border), the same legal system (legal_system), the same
colonial origin (colonial_tie), whether an exporting country is land locked (landlock_EX) or
an island (island_EX),2 whether both countries are members of WTO (WTO), have a
similarity in religion composition(religion), speak the same primary language (language).
Additionally, importing country’s GDP per capita (log(GDP_IM)) and exporting country’s
GDP per capita (log(GDP_EX)) are included as control for country size and market demand.

2

In standard gravity model, island or landlock dummy is usually generated to indicate
whether both importing and exporting countries are islands (landlocked). As only the 30-top
importing countries are included in this analysis, island and landlock dummies are created for
exporting countries only to indicate whether an exporting country is island or landlocked .
45

By standard recognition of gravity model, increase in distance between importing
and exporting country, sharing a common border, the same legal system and the colonial
origin of both countries, and both being members of WTO will increase the probability of
trade as well as trade volume between importing and exporting countries. A landlocked
exporting country exports less because of little access to ports while an islanded exporting
country exports more because of abundant access to ports, limit resources and economic
dependency. Cultural similarities in religion and language will also increase trade between
importing and exporting countries. Country size has positive impact on trade between two
countries.

2.3 Results and Discussion

Table 2.6.1 and 2.6.2 summarize the effect of compliance costs in the export decision
equation and export value equation respectively. Similarly, Table 2.6.1 and 2.6.2 summarize
the combined effect of both information costs and compliance costs. At a marginal effect
basis, paying compliance costs in the first stage decreases the probability of export by about
35 to 36 percentage points while paying both information costs and compliance costs
decreases the probability of export by about 41 percentage points depending on model
specifications (Table 2.2). The difference between these two effects approximates the
negative effect of information costs, i.e., paying information costs decreases the probability
of export by about five to six percentage point.
The estimated marginal effect of compliance costs and combined effect are
significantly larger in the export value equation, especially in the exponential type II Tobit
(ET2T) model when the export decision and export value decision are jointly determined
46

(Row (iii) and (iv), Table 2.2). In the lognormal hurdle model (i.e., export decision and
export value decision are assumed to be independent), once the hurdle is overcome, paying
compliance costs reduces export value by about 170 percent while paying both information
and compliance costs reduces export value by about 177 percent which implies paying
information costs reduces export value by about seven percent. On the other hand, in the
exponential type II Tobit (ET2T) model, paying compliance costs leads to a reduction of
more than 210 percent in export value. Meanwhile paying both information and compliance
costs leads to a reduction of more than 180 percent in export value, which is smaller than the
effect of compliance costs. This result suggests that in a process where the export decision
and export value decision are not independent (𝜌 ≠ 0), the effect of information costs and
compliance costs are not a simple additive relationship as assumed and observed in the
lognormal hurdle model. Once information is obtained and the export decision is made,
information costs may have positive impact in determining how much to export. In other
words, comparing with a firm that pays compliance costs only, a firm that pays both
information costs and compliance costs tends to export more in order to compensate the
additional information costs it has to pay for export.
The compliance costs index (Comp_Index) and the combined index of both
information costs and compliance costs (Both_Index) were generated based on theoretical
assumptions regarding export status summarized in Table 2.1. By construction, the prediction
power of these two indicators is limited to predicting the probability of export in the first
stage. However, it is not surprising to observe such a significantly negative effect of
information costs and compliance costs on export value. These marginal effects capture the
difference between exporting firms (firms categorized in Group I and II export in period 𝑡 −
1) and non-exporting with zero export values (firms categorized in Group III and IV do not
export in period 𝑡 − 1 ).
47

Table 2.2: Marginal Effects of Compliance Costs and Combined Information and
Compliance Costs
Marginal Effect

(i)
(ii)

Compliance
Costs Effect
Combined
Effect

(iii)

Compliance
Costs Effect

(iv)

Combined
Effect

Lognormal Hurdle Model
(1)
(2)
Export Probability
-0.355**
-0.351**
(0.001)
(0.001)
**
-0.415
-0.412**
(0.000)
(0.000)
Export Value
**
-1.695
-1.686**
(0.006)
(0.006)
-1.770**
(0.005)

-1.762**
(0.005)

ET2T Model
(3)

(4)

-0.361**
(0.001)
-0.412**
(0.000)

-0.358**
(0.001)
-0.410**
(0.000)

-2.189**
(0.006)3
-1.819**
(0.004)3

-2.177**
(0.006)3
-1.811**
(0.004)3

Notes: 1. Marginal effect is the partial effect averaged across sample.
2. Standard errors are reported in parenthesis. **Significant at 1%.
3. To be comparable with the Lognormal Hurdle Model, zero trade values are
excluded when calculating marginal effects.

As the independence assumption of the export decision and export value decision is
strongly rejected at
at

𝜒 2 =32706.27 for compliance costs effect (Column 3 in Table 2.6.2), and

𝜒 2 =45006.84 for combined effect of both information costs and compliance costs

(Column 3 in Table 2.7.2) under the baseline specifications, discussion of results on other
variables are focused the exponential type II tobit (ET2T) model.
For the three regulation variables, the results indicate a small negative effect of time
and cost as expected but not procedure in the export decision equation. An increase in one
hundred percent of relative official cost (as a percentage of GDP per capita) of starting up a
new business in an importing country reduces the probability of export by less than one
percentage point. An increase of ten days to start a new business in importing country reduces
the probability of export by 0.2 percentage point. On the other hand, an additional procedure
required in an importing country for a new business is found to increase the probability of
48

exporting by 0.01 percentage point. As noted earlier, it was expected that the three regulation
variables would have the same inverse relationship with the likelihood of export as they all
measure the entry costs faced by an exporting firm. One possible reason for this unexpected
positive effect of procedure on export value is that changes in the number of procedures in
importing countries are not fully captured because the three regulation variables are only
available for year 1999 while the analysis period is from 1991 to 2000.
For trade barrier variables, an improvement of logistics in an importing country
encourages foreign firms to export and export more while improvement of customs procedure
in an importing country is unexpectedly found to discourage exports. Time delay in execution
of foreign firm’s exporting procedure decreases export value but increases the probability of
export. An increase in the number of required documents regarding exporting decreases both
the probability of export and export value. An increase in tariff rate (OTRI_tariff) leads to
less export and reduced export value. Comparing with information costs and compliance
costs, the negative effect due to tariff rate is minimal.
Gravity variables have a significant effect in the export decision equation and in the
export value equation. In particular, an increase in the distances between the country pair
reduces exports. Both importing and exporting country size matters as both the probability of
exporting and export value increase as country size increases. In addition, the effect of
country size is asymmetric in such a way that the size of an exporting country has a larger
impact on both export decision and value decision. Meanwhile, exporting and importing
countries sharing a border, a common language or a same legal system will increase exports
between the countries. Exports increase for an island country, but decrease for a landlocked
country, keeping other conditions constant.

49

It is notable that the estimated coefficients on the set of gravity variables are not all
consistent with the standardized results which are obtained from a single gravity equation
frequently utilized to either predict the probability of export or the bilateral traded flow. In
particular, both importing and exporting countries being members of WTO, or having a
colonial relationship, or sharing similarities in religions does not always increase the
individual firm’s export probability and export value. The discrepancy with standardized
results estimated from aggregated country level data could be attributable to the fact that the
model is at firm level but an aggregated industry level data are used to estimate the firm level
export decision.
The inconsistency of predicted effects for some of the regulation variables, trade
barrier variables and gravity variables might be caused by the limited scope of this study.3
An aggregated data are used to approximate the firm level export decision. In addition, only
the top 30 importing countries are considered and the characteristics of exporting countries
are not controlled in this study. 4 In addition, data on regulation, trade barriers as well as
country size has no time variation. Changes in regulatory environment, trade policies and
economic development in the 30 importing countries over the analysis period are not
captured.5

3

The scope limitation does not affect the estimated effects of compliance costs and
information costs. Results are not significantly altered by experiments to drop different
gravity variables, year fixed effect and SITC2-digit fixed effect.
4
Exporting country fixed effect is not feasible due to collinearity caused by excessive
dummies in the regression.
5
Interacting with year dummy may help resolve this problem. Unfortunately, once
interaction terms added, log likelihood function is not concave and does not converge.
50

2.4 Conclusion

When compliance costs are unknown until an individual firm pays information costs
prior to exporting, the export decision is no longer as simple as if it were facing fixed export
costs that are certain. Export decisions depend on whether information costs and/or
compliance costs are required in order to export in the current period. This research
investigates the effect of information costs and compliance costs on a firm’s decisions to
export and how much to export through a hurdle model approach. By fitting a panel SITC 4digit bilateral trade flow data into an empirical representative firm-level export decision
model, this research identified and decomposed the different effects of information costs and
compliance costs in firms’ export decision and subsequent export value decision. The effects
of information costs and compliance costs are two-fold. Information costs are crucial in
determining whether or not to export to a specific foreign market. It decreases the probability
of export by about five to six percentage points in the first stage. Once the export decision is
made, firms paying information costs tend to export more in the second stage to cover such
costs. On the other hand, paying compliance costs decreases the probability of export by
about 36 percentage points. Compliance costs are more prohibitive in the subsequent export
value decision.
Due to data limitations, within-industry heterogeneity of exporting firms was ignored
by imposing some theoretical restrictions in order to fit the aggregate data to the firm-level
regression model. The true effect of information costs and compliance costs on
heterogeneous firm’s export and export value decisions would be larger if ignoring withinindustry heterogeneity of exporting firms causes any down-ward bias in the estimation.
Whenever a firm-level panel is available, restrictions imposed in models in this study are
redundant. The logic as well as the estimating procedures can be duplicated to quantify the
51

effect of information costs as well as compliance costs on the export and export value
decisions of heterogeneous firms.

52

APPENDIX

53

Table 2.3: List of 117 Exporting Countries and Areas
Albania
Algeria
Angola
Argentina
Australia
Austria
Bangladesh
Belgium–Lux.
Benin
Bhutan
Bolivia
Brazil
Bulgaria
Burkina Faso
Burundi
Cambodia
Cameroon
Canada
Central Africa Republic
Chad
Chile
China
Colombia
Congo
Costa Rica
Cote D’Ivoire
Czech Republic (Fm. Czechoslovakia)
Denmark
Dominican Rep.
Ecuador
Egypt
El Salvador
Ethiopia
Fiji
Finland
Fm. USSR
Fm. Yugoslavia
France
Germany

Ghana
Greece
Guatemala
Guinea
Haiti
Honduras
Hong Kong, China
Hungary
India
Indonesia
Iran
Ireland
Israel
Italy
Jamaica
Japan
Jordan
Kenya
Kiribati
Korea Rep. (South)
Kuwait
Laos
Lebanon
Madagascar
Malawi
Malaysia
Maldives
Mali
Mauritania
Mexico
Mongolia
Morocco
Mozambique
Nepal
Netherlands
New Zealand
Nicaragua
Niger
Nigeria

54

Norway
Oman
Pakistan
Panama
Papua N.Guinea
Paraguay
Peru
Philippines
Poland
Portugal
Romania
Rwanda
Saudi Arabia
Senegal
Sierra Leone
Singapore
Solomon Islds.
South Africa
Spain
Sri Lanka
Sweden
Switzerland
Syrn Arab Rp
Taiwan, China
Tanzania
Thailand
Togo
Tunisia
Turkey
Uganda
United Kingdom
United Arab Em.
United States
Uruguay
Venezuela
Vietnam
Yemen
Zambia
Zimbabwe

Table 2.4: List of Top 30 Importing Countries*
Australia (island)

Mexico

Austria

Norway

Belgium–Lux

Portugal

Brazil

Russia

Canada

Saudi Arabia

China

Singapore (island)

Czech Republic (landlocked)

Spain

France

Sweden

Germany

Switzerland (landlocked)

India

Thailand

Indonesia (island)

Turkey

Italy

United Kingdom (island)

Japan (island)

United Arab Em.

Korea Rep. (South)

United States

Malaysia

Vietnam

*Notes: Countries are listed in alphabetical order. The rank is based on World Trade Report
2012 by WTO
(http://www.wto.org/english/res_e/booksp_e/anrep_e/world_trade_report12_e.pdf)

55

Table 2.5: Descriptive Statistics of Variables
Variable Name

Definition

Complinace_Index

Binary, 1 if a firm doesn’t pay information costs but
does pay compliance costs in order to export to a
specific destination, zero otherwise
Binary, 1 if a firm pays both information costs and
compliance costs in order to export to a specific
destination, zero otherwise
Binary, 1 if a firm exports to a specific destination, 0
otherwise
Log of yearly trade value between a importing and
exporting county pair at SITC4-digit level
No. of procedures required for a domestic start-up firm
to legally operate a business in the importing country
No. of official days required for a domestic start-up
firm to legally operate a business in the importing
country
Official costs required for a domestic start-up firm to
legally operate a business in the importing country (%
of GDP per capita)
Logistics Performance Index (a scale of 1-5 with 1
indicating the worst performance
Burden of Customs Procedures (a scale number of 1-7
with 1 indicating the largest burden),
No. of days between initiation and execution of a
exporting process in importing country
No. of documents required by the importing countries
in order to export

Both_Index

EXPORT
Log(TV)
Regulat
ion

Procedure
Time

Cost (%)

Trade
Barrier

Logistics
Customs
Lead_Days
Documents

56

No. of
Observations
2,012,587

0.121

Std.
Dev.
0.326

4,493,027

0.606

0.489

0

1

5,919,840

0.373

0.484

0

1

2,208,910

14.071

1.821

6.908

24.26
1

5,919,840

8.039

3.355

2

17

5,919,840

38.903

37.580

2

152

5,919,840

16.517

25.031

0.6

130.7

5,285,150

3.636

0.376

2.61

4.11

5,285,150

4.442

0.997

0.6

5.8

5,285,150

3.414

1.484

1

7.1

5,285,150

5.108

1.4870

2

13

Mean

Min

Max

0

1

Table 2. 5 (Cont’d)

Gravity

OTRI_tariff
(%)
Log(GDP_IM)
Log(GDP_EX)
Log(distance)
Border
Island_EX
Landlock_EX
Language
Legal system
Religion

Colonial tie
WTO

Overall trade restrictiveness index (tariff data only)
Log of GDP per capita in importing country (in 1999
US dollars)
Log of GDP per capita in exporting countries (in 1999
US dollars)
Log of the distance (in km) between importing and
exporting country’s capital
Binary, 1 if importing and exporting country shares a
common border, 0 otherwise
Binary, 1 if exporting country is islands, 0 otherwise
Binary, 1 if exporting country has no direct access to
sea, 0 otherwise
Binary, 1 if both importing and exporting country use
the same language as official language
Binary, 1 if both importing and exporting country share
the same legal origin
(% Protestants in importing country* % Protestants in
exporting country )+(% Catholics in importing country
* %Catholics in exporting country) + (% Muslims in
importing country* %Muslims in exporting country )
Binary, 1 if importing country ever colonized exporting
country or vice versa, 0 otherwise.
Binary, 1 if both exporting and importing country
belong to WTO, 0 otherwise

57

5,285,150

5.615

5.560

1.7

26.1

5,919,840

9.401

1.227

6.174

5,908,740

9.060

1.425

4.500

5,919,840

3.908

1.016

0.882

5.661

5,919,840

0.069

0.253

0

1

5,919,840

0.106

0.308

0

1

5,919,840

0.114

0.318

0

1

5,919,840

0.214

0.410

0

1

5,919,840

0.276

0.447

0

1

5,919,840

0.174

0.254

0

0.987

5,919,840

0.044

0.206

0

1

5,919,840

0.739

0.439

0

1

10.67
7
10.67
7

Table 2.6.1: Compliance Costs Effect in Export Decision Equation
Explanatory
Variable

Export Decision Equation
Dependent Variable: Export
ET2T Model
(Heckman Method)

Lognormal Hurdle Model

Compliance Index
No. of procedures
Time
Cost (%)
Logistics
Customs
Lead Time
Documents
OTRI_tariff
Log(GDP_IM)
Log(GDP_EX)
Log(distance)
Border
Island_EX
Landlock_EX
Language
Legal System
Religion
Colonial tie
WTO

(1)
-1.609**
0.004**
-0.001**
0.001**
0.204**
-0.075**
-0.027**
-0.042**
-0.006**
0.031**
0. 114**
-0. 019**
0. 329**
0.141**
-0.150**
0. 018**
0.036**
-0.024**
-0.041**
-0.027**

(0.004)
(0.001)
(0.000)
(0.000)
(0.010)
(0.002)
(0.001)
(0.001)
(0.000)
(0. 003)
(0. 001)
(0.002)
(0.007)
(0.005)
(0.005)
(0.004)
(0.004)
(0.007)
(0.008)
(0.004)

(2)
-1.601**
0.002*
-0.001**
0.001**
0.079**
-0.042**
-0.003**
-0.007**
0.020**
-0.044**
0.047**
-0.031**
0.328**
0.136**
-0.139**
0.034**
0.027**
0.035**
-0.016*
0.017**

(0.003)
(0.001)
(0.000)
(0.000)
(0.012)
(0.003)
(0.001)
(0.002)
(0.001)
(0.005)
(0.004)
(0.002)
(0.007)
(0.005)
(0.005)
(0.004)
(0.004)
(0.007)
(0.008)
(0.005)

58

(3)
-1.618**
0.008**
-0.001**
-0.000**
0.331**
-0.087**
0.001
-0.071**
-0.006**
0.006*
0.115**
-0.020**
0.505**
0.156 **
-0.223**
0.044**
0.040**
-0.106**
-0.042**
-0.034**

(0.004)
(0.000)
(0.000)
(0.000)
(0.010)
(0.002)
(0.001)
(0.001)
(0.000)
(0.003)
(0.001)
(0.002)
(0.008)
(0.005)
(0.005)
(0.004)
(0.004)
(0.007)
(0.009)
(0.004)

(4)
-1.611**
0.009**
-0.001**
-0.000
0.194 **
-0.070**
0.003+
-0.063**
0.003**
0.007
0.105**
-0.023**
0.505**
0.150**
-0.227**
0.063**
0.032**
-0.058**
-0.040**
-0.042**

(0.004)
(0.000)
(0.000)
(0.000)
(0.012)
(0.002)
(0.002)
(0.002)
(0.001)
(0.005)
(0.005)
(0.002)
(0.008)
(0.005)
(0.005)
(0.004)
(0.004)
(0.007)
(0.009)
(0.005)

Table 2.6.1(Cont’d)
Year fixed effects
SITC2-digit
fixed
effects
Importing country
income group fixed
effectsa,b
Exporting country
income group fixed
effectsa,b
Observations
Log likelihood

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

-

Yes

-

Yes

-

Yes

-

Yes

1,783,220
-713,346

1,783,220
-708,650

1,783,221
-3,416,187

1,783,221
-3,410,884

Notes: Robust standard errors (clustering by country pair SITC 4-digit level) are reported in parenthesis.
+Significant at 10%.
*Significant at 5%.
**Significant at 1%.
a.
Importing and exporting country fixed effects are suppressed as the log likelihood function does not converge.
b. According to World Bank estimates of 1999 GNP per capita. Low income group: $755 or less; Lower middle income group: $7562,995; Upper middle income group: $2,996-9,265; High income group: $9,266 or more.

59

Table 2.6.2: Compliance Costs Effect in Export Value Equation
Explanatory
Variable

Export Value Equation
Dependent Variable: log(TV)
ET2T Model
(Heckman Method)

Lognormal Hurdle Model

Index
No. of procedures
Time
Cost (%)
Logistics
Customs
Lead Time
Documents
OTRI_tariff
Log(GDP_IM)
Log(GDP_EX)
Log(distance)
Border
Island_EX
Landlock_EX
Language
Legal System
Religion
Colonial tie
WTO

(1)
-1.695**
0.384**
-0.162**
-0.065**
-0.025**
-0.005**
0.043**
0.121**
-0.023**
0.992**
0.209*
-0.405**
0.095**
0.069**
-0.266**
-0.079**
-0.218**

(0.006)
(0.022)
(0.004)
(0.003)
(0.003)
(0.001)
(0.007)
(0.003)
(0.004)
(0.016)
(0.011)
(0.011)
(0.009)
(0.009)
(0.015)
(0.019)
(0.001)

(2)
-1.686**
0.078**
-0.150**
-0.079**
-0.022**
-0.002
0.096**
0.248**
-0.024**
0.996**
0.222**
-0.442**
0.119**
0.065**
-0.232**
-0.091**
-0.288**

(0.006)
(0.027)
(0.005)
(0.004)
(0.003)
(0.002)
(0.012)
(0.011)
(0.004)
(0.017)
(0.011)
(0.011)
(0.009)
(0.009)
(0.015)
(0.019)
(0.011)

60

(3)c
-3.374**
0.409**
-0.200**
-0.094**
-0.040*
-0.010**
0.080**
0.203**
-0.033**
1.130**
0.290**
-0.481**
0.103**
0.088**
-0.268**
-0.110**
-0.244**

(0.008)
(0.022)
(0.004)
(0.003)
(0.003)
(0.001)
(0.007)
(0.003)
(0.004)
(0.017)
(0.012)
(0.011)
(0.010)
(0.009)
(0.016)
(0.019)
(0.010)

(4)d
-3.350**
0.070**
-0.153**
-0.089**
-0.015**
0.012**
0.073**
0.262**
-0.041**
1.133**
0.304**
-0.504**
0.136**
0.077**
-0.213**
-0.095**
-0.265**

(0.009)
(0.028)
(0.005)
(0.004)
(0.003)
(0.002)
(0.012)
(0.011)
(0.004)
(0.017)
(0.011)
(0.011)
(0.010)
(0.009)
(0.016)
(0.019)
(0.011)

Table 2.6.2 (cont’d)
Year fixed effects
SITC2-digit
fixed
effects
Importing country
income group fixed
effectsa, b
Exporting country
income group fixed
effectsa, b
Observations
Log likelihood

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

-

-

Yes

Yes

-

-

Yes

Yes

1,419,403
-28,823,138

1,743,634
-28,816,311

1,783,221
-3,416,187

1,783,221
-3,410,884

Notes: Robust standard errors (clustering by country pair SITC4-digit level) are reported in parenthesis.
+Significant at 10%.
*Significant at 5%.
**Significant at 1%.
a.
Importing and exporting country fixed effects are suppressed as the log likelihood function does not converge.
b. According to World Bank estimates of 1999 GNP per capita. Low income group: $755 or less; Lower middle income group: $7562,995; Upper middle income group: $2,996-9,265; High income group: $9,266 or more.
c.
Independent assumption of two stages(𝜌 = 0) is rejected with 𝜒 2 =30961.78.
d.

Independent assumption of two stages(𝜌 = 0) is rejected with 𝜒 2 =31144.09.

61

Table 2.7.1: Combined Information Costs and Compliance Costs Effect in Export Decision Equation
Explanatory
Variables

Export Decision Equation
Dependent Variable: Export
ET2T Model
(Heckman Method)

Lognormal Hurdle Model

Both Index
No. of procedures
Time
Cost (%)
Logistics
Customs
Lead Time
Documents
OTRI_tariff
Log(GDP_IM)
Log(GDP_EX)
Log(distance)
Border
Island_EX
Landlock_EX
Language
Legal System
Religion
Colonial tie
WTO

(1)
-2.502**
-0.001**
-0.001**
0.001**
-0.012**
-0.109**
-0.015**
-0.069**
0.004**
-0.005**
0.054**
0.003**
0.068**
0.096*
-0.071**
0.105**
0.063**
0.006
-0.130**
0.043**

(0.002)
(0.000)
(0.000)
(0.000)
(0.005)
(0.001)
(0.001)
(0.001)
(0.000)
(0.002)
(0.001)
(0.001)
(0.004)
(0.003)
(0.003)
(0.002)
(0.002)
(0.004)
(0.004)
(0.002)

(2)
-2.494**
-0.002**
-0.002**
0.001**
-0.064**
-0.083**
0.003**
-0.052**
0.022**
-0.041**
-0.037**
-0.001**
0.060**
0.091**
-0.061**
0.108**
0.060**
0.042**
-0.113**
0.077**

(0.002)
(0.000)
(0.000)
(0.000)
(0.006)
(0.002)
(0.001)
(0.001)
(0.000)
(0.003)
(0.003)
(0.001)
(0.004)
(0.003)
(0.003)
(0.002)
(0.002)
(0.004)
(0.004)
(0.003)

62

(3)
-2.480**
0.005**
-0.001**
-0.000**
0.091**
-0.081**
0.001
-0.082*
0.002**
-0.003
0.066**
-0.000
0.242**
0.119**
-0.142**
0.110**
0.065**
-0.062**
-0.118**
0.019**

(0.002)
(0.005)
(0.000)
(0.000)
(0.006)
(0.001)
(0.001)
(0.001)
(0.000)
(0.002)
(0.001)
(0.001)
(0.005)
(0.004)
(0.003)
(0.003)
(0.003)
(0.005)
(0.006)
(0.003)

(4)
-2.473**
0.006**
-0.001**
-0.000**
-0.033**
-0.090**
0.005**
-0.077**
0.009**
0.019**
0.021**
-0.003*
0.242**
0.114**
-0.142**
0.121**
0.061**
-0.032**
-0.116**
0.014**

(0.003)
(0.000)
(0.000)
(0.000)
(0.007)
(0.002)
(0.001)
(0.001)
(0.000)
(0.004)
(0.003)
(0.001)
(0.005)
(0.004)
(0.003)
(0.003)
(0.003)
(0.005)
(0.006)
(0.003)

Table 2.7.1 (cont’d)
Year fixed effects
SITC2-digit
fixed
effects
Importing country
income group fixed
effectsa, b
Exporting country
income group fixed
effectsa, b
Observations
Log likelihood

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

-

Yes

-

Yes

-

Yes

3,978,791
-1,220,891

3,978,791
-1,216,647

Yes
4,007,318
-4,083,178

4,007,318
-4,078,049

Notes: Robust standard errors (clustering by country pair SITC4-digit level) are reported in parenthesis.
+Significant at 10%.
*Significant at 5%.
**Significant at 1%.
a.
Importing and exporting country fixed effects are suppressed as the log likelihood function does not converge.
b. According to World Bank estimates of 1999 GNP per capita. Low income group: $755 or less; Lower middle income group: $7562,995; Upper middle income group: $2,996-9,265; High income group: $9,266 or more.

63

Table 2.7.2: Combined Information Costs and Compliance Costs Effect in Export Value Equation
Export Value Equation

Explanatory Variable

Dependent Variable: log(TV)

ET2T Model
(Heckman Method)

Lognormal Hurdle Model
(1)
Both Index
No. of procedures
Time
Cost (%)
Logistics
Customs
Lead Time
Documents
OTRI_tariff
Log(GDP_IM)
Log(GDP_EX)
Log(distance)
Border
Island_EX
Landlock_EX
Language
Legal System
Religion
Colonial tie
WTO

-1.770**
0.346**
-0.154**
-0.065**
-0.020**
-0.005**
0.039**
0.110**
-0.021**
0.971**
0.203*
-0.374**
0.087**
0.065**
-0.246**
-0.082**
-0.201**

(0.005)
(0.020)
(0.004)
(0.003)
(0.003)
(0.001)
(0.006)
(0.003)
(0.004)
(0.016)
(0.011)
(0.010)
(0.009)
(0.008)
(0.013)
(0.008)
(0.009)

(2)
-1.763**
0.050**
-0.143**
-0.080**
-0.017**
-0.001
0.084**
0.227**
-0.020**
0.975**
0.216**
-0.408**
0.112**
0.062**
-0.211**
-0.094**
-0.266**

(0.004)
(0.025)
(0.004)
(0.003)
(0.003)
(0.001)
(0.011)
(0.010)
(0.004)
(0.016)
(0.011)
(0.010)
(0.009)
(0.008)
(0.014)
(0.017)
(0.009)

64

(3)c
-4.973**
0.172**
-0.217**
-0.094**
-0.057**
-0.000
0.066**
0.145**
-0.010**
0.877**
0.255**
-0.370**
0.178**
0.124**
-0.202**
-0.222**
-0.155**

(0.009)
(0.017)
(0.004)
(0.003)
(0.002)
(0.001)
(0.005)
(0.002)
(0.004)
(0.015)
(0.010)
(0.009)
(0.008)
(0.007)
(0.012)
(0.015)
(0.008)

(4)d
-4.950**
-0.159**
-0.172**
-0.089**
-0.037**
0.020**
0.072**
0.145**
-0.018**
0.882**
0.266**
-0.385**
0.206**
0.117**
-0.164**
-0.209**
-0.173**

(0.009)
(0.020)
(0.004)
(0.003)
(0.003)
(0.001)
(0.010)
(0.009)
(0.004)
(0.015)
(0.010)
(0.009)
(0.008)
(0.007)
(0.013)
(0.015)
(0.008)

Table 2.7.2 (cont’d)
Year fixed effects
SITC2-digit
fixed
effects
Importing
country
income group fixed
effects
Exporting
country
income group fixed
effect
Observations
Log likelihood

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

-

Yes

-

Yes

-

Yes

-

Yes

1,533,769
-29,518,233

1,533,769
-29,511,899

4,007,318
-4,083,178

4,007,318
-4,078,049

Notes: Robust standard errors (clustering by country pair SITC4-digit level) are reported in parenthesis.
+Significant at 10%.
*Significant at 5%.
**Significant at 1%.
a.
Importing and exporting country fixed effects are suppressed as the log likelihood function does not converge.
b.
According to World Bank estimates of 1999 GNP per capita. Low income group: $755 or less; Lower middle income group: $7562,995; Upper middle income group: $2,996-9,265; High income group: $9,266 or more.
c.
Independent assumption of two stages(𝜌 = 0) is rejected with 𝜒 2 =45006.84.
d.

Independent assumption of two stages(𝜌 = 0) is rejected with

65

𝜒 2 =45368.32.

Figure 2.1.1: Timeline for firm export decisions when fixed export costs are certain

facing fixed
export costs

incurring
fixed export
costs

export
decision

Firm interested in exporting:

not export

export
value
decision

t

Figure 2.1.2: Timeline for firm export decisions when there is uncertainty in fixed export
costs, such as compliance costs

Firm interested in exporting:

incurring
information
cost

incurring
compliance
costs

not export

revealing
compliance
cost

export
decision

66

export
value
decision

t

REFERENCES

67

REFERENCES

Bernard, A. B. and J. B. Jensen (2004). Why some firms export. The Review of Economics
and Statistics 86 (2), 561–69.

Blanes-Cristóbal, J. V., M. Dovis, J. Milgram-Baleix, and A. I. Moro-Egido (2008). Do sunk
exporting costs differ among markets? evidence from spanish manufacturing firms.
Economics Letters 101 (2), 110 – 12.

Cragg, J. G. (1971). Some statistical models for limited dependent variables with application
to the demand for durable goods. Econometrica 39 (5), 829–44.

Das, S., M. J. Roberts, and J. R. Tybout (2007). Market entry costs, producer heterogeneity,
and export dynamics. Econometrica 75 (3), 837–73.

Eaton, J. and A. Tamura (1994). Bilateralism and regionalism in Japanese and US trade and
direct foreign investment patterns. Journal of the Japanese and international economies 8
(1), 478–510.

Feenstra, R. C., R. E. Lipsey, H. Deng, A. C. Ma, and H. Mo (2005). World trade flows:
1962-2000. Working Paper 11040, National Bureau of Economic Research.

Haveman, J., & Hummels, D. (2004). Alternative Hypotheses and the Volume of Trade: The
Gravity Equation and the Extent of Specialization. Canadian Journal of Economics 37 (1),
199-218.

Helpman, E., M. Melitz, and Y. Rubinstein (2008). Estimating trade flows: Trading partners
and trading volumes. The Quarterly Journal of Economics 123 (2), 441–87.

Kee, L. H., A. Nicita, and M. Olarreaga (2009). Estimating trade restrictiveness indices. The
Economic Journal 119 (534), 172–99.
Maskus, K. E., T. Otsuki and J. S. Wilson. “An empirical framework for analyzing
technical regulations and trade,” in Quantifying the Impact of Technical Barriers to Trade,
ed. Keith E. Maskus and Jone S. Wilson (Ann Arbor : University of Michigan Press,
2001), 29.

68

Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate
industry productivity. Econometrica 71 (6), 1695–725.

Morales, E., G. Sheu, and A. Zahler (2011). Gravity and extended gravity: estimating
a structural model of export entry. MPRA Paper 30311, University Library of Munich,
Germany.

Roberts, M. J. and J. R. Tybout (1997). The decision to export in Colombia: An
empirical model of entry with sunk costs. American Economic Review 87 (4), 545–64.

Thornsbury, S., D. Roberts, and D. Orden. Measurement and political economy of disputed
technical regulations. Journal of agricultural and applied economics 36 (3).

Tobin, J. (1958). Estimation of relationships for limited dependent variables. Econometrica
26 (1), 24–36.

Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data, Volume
1. The MIT Press.

69

CHAPTER 3: A STRUCTURAL ESTIMATION OF THE EMPLOYMENT
EFFECTS OF OFFSHORING IN THE U.S. LABOR MARKET

3.1 Introduction
The debate over offshoring has intensified in the United States as offshoring has
spread from the jobs of blue-collar workers in the manufacturing sector to those of whitecollar workers in service sectors. The service sector comprises about 80 percent (U.S.
Department of Commerce) of the U.S. employment and most white-collar workers are
employed in the service sector. U.S. workers in all sectors have become more concerned
about the security of their jobs due to increased offshoring activities as the global economy
has continued to integrate. Given changes in technology (the internet), a well-educated
radiologist and a low-skilled auto assembly line worker could both be susceptible to
offshoring. These concerns are well reflected in results from Princeton University’s
telephone survey conducted in summer 2008.1 Survey results indicate occupational
offshorability reported by individual survey respondents are much higher than those
predicted by economists.
The increase in offshoring along with a persistently high unemployment rate in
recent years, has heightened policymaker concerns and has been the subject of increased
economic research on the short- and long-run labor market implications of offshoring and in
particular, the potential for U.S. job loss. The actual impact of offshoring is multidimensional and difficult to quantify. Existing empirical estimates (Bardhan and Kroll,
2003; Blinder, 2007; Blinder, 2009) provide a wide range of estimates for offshorable jobs in
the U.S. labor market, varying from 11 to 47 percent. With relatively little theoretical

1

For details, see Blinder and Krueger (2009).
70

guidance, the wide range in early empirical estimates provided limited information to
policymakers facing tensions from a high national unemployment rate exceeding nine
percent.
Under such circumstances, an economic theory of offshoring has been exposited by
Grossman and Rossi-Hansberg (2008). In their parsimonious framework, job tasks are
defined as either low-skilled or high-skilled. Using comparative static analysis, they analyze
the synergic action of productivity effect, relative-price effect and labor supply effect of
offshoring on these two groups due to a change of offshoring costs. Their results show that
offshoring might lead to wage gains for both low-skilled and high-skilled workers and create
a win-win situation for all types of workers, but not necessarily reward one player by
harming the others as stated in the traditional Stolper-Samuelson results. Motivated by these
results, several papers empirically tested the effect of offshoring in the United States
(Harrison and McMillan, 2010; Ebenstein et al., 2013; Crinò, 2010b) and in European
countries (Goos et al., 2010; Crinò, 2010a; Criscuolo and Garicano, 2010).
Harrison and McMilan (2010) estimated a reduction of four million jobs in U.S.
manufacturing employment due to offshoring over the period of 1982 to 1999. Ebenstein et
al. (2013) found that the impact of offshoring on U.S. worker’s wages has been
underestimated by previous studies because offshoring has driven workers from high-wage
manufacturing jobs to low-wage service jobs. In addition, workers performing routine tasks
are most affected by offshoring and experience larger wage decline. On the other hand,
studying the effects of service offshoring on white-collar employment in more than 100 U.S.
occupations, Crinò (2010 b) concluded that (a) service offshoring increases employment in
more skilled occupations relative to less skilled occupations; (b) at a given skill level, service
offshoring penalizes offshorable occupations while benefiting less-offshorable occupations.
However, evidence from European countries is mixed. Goos et al. (2010) found that
71

offshoring was associated with reduced employment in offshorable occupations across 16
European countries as opposed to Crinò is (2010 a) finding that service offshoring has no
effect on employment in Italian firms. Using occupational licensing as a shifter of offshoring
costs, Criscuolo and Garicano (2010) found that an increase in service offshoring increased
both wages and employment in less-offshorable service occupations (i.e., licensed
occupations) in the UK.
Grossman and Rossi-Hansberg’s theoretical framework includes wage implications and
may partially relieve policymaker concerns over increased wage inequality due to offshoring
in the U.S. labor market, but it does not address the core question of to what extent
offshoring will affect labor demand (i.e., number of jobs). Goos et al. (2010) did find
offshoring to be an explanatory factor affecting the conditional demand for labor in different
occupations in their theoretical model and estimation, but other existing studies simply
extend their empirical investigation to the effects of offshoring on wage or employment and
provide some empirical evidences.
In this paper, Grossman and Rossi-Hansberg’s (2008) offshoring model is generalized
to include numerous tasks/skill levels (tasks correspond to specific occupations in this
empirical framework) and investigate the effect of offshoring on occupational employment
for ten major occupational groups (at 2-digit SOC level) in the U.S. labor market (see Table
3.2.1 and 3.2.2 for details of occupational groups). Using the CPSMORG (Current
Popolution Survey Merged Outgoing Rotation Groups) data from year 1983 to 2011,
analysis is conducted in two phases. First, the monotonic cubic spline interpolation method
is used to estimate the offshoring cost functions for all ten occupational groups. The
monotonic cubic spline interpolation method requires no specific functional form other than
the assumption that offshoring costs are non-decreasing in the percentage of tasks being
offshored. This nice property makes monotonic cubic spline interpolation method a perfect
72

fit for this study because offshorability for one occupational group could largely differ from
another. Next, a parametric method-nonlinear least squares (NLS)-is utilized for the five
relatively offshorable occupational groups. Based on the monotonic cubic spline
interpolation results, a cubic functional form is attached to the five relatively offshorable
occupational groups to approximate their offshoring cost functions. Then, the number of
jobs offshored and the offshoring percentage over the sample period for the five offshorable
occupational groups are calculated.
Aside from a limited number of studies with primary information on offshoring
activities (see for example, Crinò, 2010), researchers have used two alternative approaches
to measure offshoring over time. The first approach is to approximate or infer offshoring
activities using relevant information. For example, Ebenstein et al. (2013) use foreign
affiliate employment of U.S. multinational firms as a measure capturing U.S. firms’
offshoring activities. Criscuolo and Garicano (2010) use occupational licensing to infer the
offshorability of an occupation in their study of offshoring of UK service sectors.
Approximation of offshoring activities circumvents the issue of time-invariance of
offshoring/offshorabilty index, but reliability of the approximation is unknown.
The second approach is to generate a time-invariant offshoring index based on firm
offshoring activities. For example, Goos et al. (2010) construct an occupational
offshorability index based on offshoring activities of 415 European firms between 2002 and
2008. Applying a time-invariant index assumes that the offshoring activities are either not
influenced by the reduction of offshoring costs or that costs are constant over time. A time
variant offshoring index is thus especially important when investigating the effect of
offshoring over a relative long-time span. For example, the occupation of a radiologist
would be considered as non-offshorable without the advancement in recent
telecommunication technology which makes transformation of large image data a relatively
73

costless task. An important contribution of this paper is to provide time-variant estimates of
offshoring for more than 400 major U.S. occupations over the period of 1983 to 2011.

3.2 A Simple Structural Model of Offshoring
Inspired by empirical findings about the impact of characteristics of tasks on wage
inequality and employment structure (e.g., Autor et al., 2003), Grossman and RossiHansberg (2008) proposed a theoretical model of task offshoring to explain the impact of
offshoring on the wage rates of different types of workers. In the Grossman and RossiHansberg model, tasks are limited to only two types: low-skill and high- skill. Under a
standard Heckscher-Ohlin set-up, Grossman and Rossi-Hansberg (2008) show how
changing offshoring costs will affect the wage rates of low-skilled and high-skilled workers
in the home country through static comparative analysis.
This research generalizes the analysis to include numerous tasks and link the concept
of tasks to detailed occupations that are actually offshored. While the focal point of
Grossman and Rossi-Hansberg (2008) is to decompose effects of offshoring on factor prices
i.e., wage rates, this research focuses on exploring the effect of offshoring on employment
levels in different occupations. To be consistent and comparable with Grossman and RossiHansberg (2008), this research uses the term “task” instead of “occupation” in the structural
model specification, but freely changes between these two in the remaining of this paper
depending on the context. 2

3.2.1 Model Specification
The production process requires many types of tasks and each type of task is denoted

2

Each task corresponds to an occupation in our empirical framework.
74

by 𝑜. Producing one unit of a specific good involves a continuum of each type of task.
Without loss of generality, the measure of each type of task can be normalized to one.
Firms in the home country produce many goods. The number of goods produced in the
home country is assumed to be larger than the number of types of tasks.3 All tasks are
involved in order to produce one unit of specific good,4 i.e.,

𝑎 𝑜𝑗 is the total amount of

domestic factor o that would be needed to produce a unit of good j in the absence of any
offshoring. Firms can undertake an 𝑜-type task either at home or abroad depending on the
offshoring costs and the relative wage of task 𝑜 between home and foreign country. An 𝑜type task is indexed by 𝑖 ∈ [0, 1] and ordered in a manner such that the offshoring cost of
task 𝑜, denoted by 𝑡(𝑖), is non-decreasing in 𝑡.
3.2.2 Model Derivation
As some tasks are more difficult to offshore than others, offshoring costs are assumed
to be varying across different tasks and changing over time. Denote offshoring costs shifter
as

𝛽 𝑜,𝑠 with subscript

𝑜 indicating task type and 𝑠 indicating time period. Let

𝑤 ∗ be respectively the home and foreign wage of task
𝑜,𝑠
home and foreign country of each task 𝑜, denoted by

𝑤 𝑜,𝑠 and

𝑜 . Then the relative wage between

𝜔 𝑜,𝑠 , satisfies 𝜔 𝑜,𝑠 =

𝑤 𝑜,𝑠
𝑤∗
𝑜,𝑠

for all

periods s.

3

This assumption is to guarantee a unique solution to the factor price of each type of task
given the price and production technology of each good.
4
If the cost-minimizing demand for factor o is zero, the o-type task will be missing in the
production process.
75

Following Grossman and Rossi-Hansberg (2008)’s formulation, 𝐼 𝑜,𝑠 , the equilibrium
marginal task 𝑜 performed at home (or the cutoff point of task 𝑜 at equilibrium) in period s
in each industry is determined by the following condition such that wage savings exactly
balance the offshoring cost of task 𝑜:

𝑤 𝑜,𝑠 = 𝑤 ∗ 𝛽 𝑜,𝑠 𝑡(𝐼 𝑜,𝑠 ) .
𝑜,𝑠
Then by my relative wage assumption

𝑡(𝐼 𝑜,𝑠 ) =

𝜔 𝑜,𝑠
𝛽 𝑜,𝑠

(3.1)

𝜔 𝑜,𝑠 =

𝑤 𝑜,𝑠
𝑤∗
𝑜,𝑠

,

= ρo,s ,

(3.2)

where

ρo,s denotes the equilibrium offshoring costs, which depends on the ratio of relative

wage

𝜔 𝑜,𝑠 and the offshoring cost shifter 𝛽 𝑜,𝑠 at each period

𝑠. Given that 𝑡(∙) is an

increasing function in 𝐼 𝑜,𝑠 , a higher proportion of task 𝑜 will be moved offshore as

𝐼 𝑜,𝑠

increases. As 𝐼 𝑜,𝑠 is the cutoff point of the marginal task 𝑜 performed at home country,

ρo,s precisely captures the offshoring decisions made by home firms.
Denote
offshoring and

𝐿 𝑜 the initial total employment of occupation
𝐿 𝑜,𝑠 the employment of occupation

observed in data, then

𝑜 at home country without

𝑜 in period 𝑠 with offshoring, which is

𝐿 𝑜,𝑠 , can be calculated as following:

𝐿 𝑜,𝑠 = (1 − 𝐼 𝑜,𝑠 ) ∙ 𝐿 𝑜 ,

(3.3)

76

where 1

− 𝐼 𝑜,𝑠 indicates the fraction of

𝑜-type tasks that are performed at home.

Under the perfect competitive assumption, the price of any good 𝑗 is equal to the unit
cost of production (if a positive quantity of the good is produced):

𝑝 𝑗 = ∑ 𝑜 𝑤 𝑜,𝑠 Ω( Io,s )𝑎 𝑜𝑗 (∙),

(𝑗 > 𝑜)

where, the arguments in the function for the factor intensity

(3.4) 5

𝑎 𝑜𝑗 are the relative costs of the

various sets of tasks when they are located optimally with offshoring,
and Ω(

Io,s ) = 1 − Io,s +

In other words, Ω(

𝐼
∫ 𝑜,𝑠 𝑡(𝑖)𝑑𝑖
0

𝑡(𝐼 𝑜,𝑠 )

.

(3.5)

Io,s ) consists of two parts, 1 − Io,s (the proportion of tasks that

remains in home country) and

𝐼
∫ 𝑜,𝑠 𝑡(𝑖)𝑑𝑖
0

𝑡(𝐼 𝑜,𝑠 )

(the proportion of tasks conducted in foreign

country expressed in equivalent home-country factor employment).
As

𝐼 𝑜,𝑠 =

𝐿 𝑜 −𝐿 𝑜,𝑠
𝐿𝑜

=1−

𝐿 𝑜,𝑠
𝐿𝑜

is a function of

𝐿 𝑜 , Ω( Io,s ) is a function of 𝐿 𝑜 .

Since the number of the goods is larger than the number of factors (𝑗 > 𝑜), factor
prices(𝑤 𝑜,𝑠 Ω(

Io,s )) can be uniquely determined and solved from the systems of

equations (3.4). That is,

𝑤 𝑜,𝑠 Ω( Io,s ) = 𝑐 𝑜 ,

(3.6)

5 Equivalent to Equation (3) in Section I, Grossman and Rossi-Hansberg (2008). For

detailed derivation, please refer to Grossman and Rossi-Hansberg (2008).
77

where

𝑐 𝑜 depends on the prices 𝑝 𝑗 and all production technologies of all goods produced in

home country. Identity (3.6) is the key equation in identifying the equilibrium cutoff point
of offshoring percentage (Io,s ) of task 𝑜, offshoring cost 𝑡(𝑖) as well as constant

𝑐 𝑜.

Section 3.2.3 explains estimation of Equation (3.6).
3.2.3 Model Interpretation
Although the Grossman and Rossi-Hansberg (2008) model is static, it can be
interpreted with some dynamics within each period. Given the wage differential between the
home and foreign country, the equilibrium cutoff point of offshoring 𝐼 𝑜,𝑠 is determined by
Equation (3.1) at the beginning of period 𝑠, which automatically determines the domestic
labor demand for task 𝑜 (in Equation (3.3)). By the zero-profit condition under perfect
competition, the new wage

𝑤 𝑜,𝑠 for task

𝑜 at the end of period 𝑠 in the home country is

obtained by solving Equation (3.4) (or equivalently Equation (3.6)). If the new wage

𝑤 𝑜,𝑠 is

higher (or lower) than the starting wage in period 𝑠, the firm in the home country increases
(or decreases) offshoring until it reaches its new equilibrium cutoff point at the end of period
𝑠 that we observe in the data. The same process repeats in all periods.
By this interpretation, it is explicitly assumed the wage and employment observed in
our data set are equilibrium wage and employment at the end of each period, which are both
driven by offshoring. Then by estimating Equation (3.6), we can identify the offshroing cost
function 𝑡(𝑖)6 and the initial employment without offshoring for each task 𝑜.

6

However, the offshoring cost function 𝑡(𝑖) can only be identified up to a constant scale
because multiplying a scalar to 𝑡(𝑖), Equation (3.6) still holds.
78

3.3 Estimation Framework and Method
3.3.1 The Empirical Framework
To estimate Equation (3.6), take logarithm and reorder, which leads to,

𝑙𝑛𝑤 𝑜,𝑠 = −𝑙𝑛Ω(Io,s ) + 𝑙𝑛𝑐 𝑜 = 𝑙𝑛𝑐 𝑜 − 𝑙𝑛Ω(Io,s ).
As Ω(

(3.7)

Io,s ) is a function of the observed variable 𝐿 𝑜,𝑠 , unobserved parameters

𝐿 𝑜 and the offshoring cost function

𝑡(∙), standard linear estimation methods which can only

estimate unknown parameters but not unknown functions are not applicable.
Further denote

𝑦 𝑜,𝑠 = 𝑙𝑛𝑤 𝑜,𝑠 , 𝑥 𝑜,𝑠 = 𝐿 𝑜,𝑠 . Then the conditional mean of

𝑦 𝑜,𝑠 can be specified as
𝐸(𝑦 𝑜,𝑠 |𝑥 𝑜,𝑠 ) = 𝑚(𝑥 𝑜,𝑠 , 𝜽 𝟎 ) = 𝑙𝑛𝑐 𝑜 − 𝑙𝑛Ω ( Io,s (𝐿 𝑜,𝑠 , 𝐿 𝑜, 𝑡(∙)))
(3.8)
Where

𝜽 𝟎 = ( 𝐿 𝑜 , 𝑐 𝑜 , 𝑡(∙)) consists of two parameters and one function to be

identified. Since

𝜽 𝟎 contains the offshoring cost function that cannot be directly estimated,

I need to parameterize 𝑡(∙) in order to proceed to estimate 𝑡(∙) together with the other two
parameters.
No specific structure except the monotonicity of 𝑡(𝑖) (i.e., 𝑡(𝑖) is non-decreasing in
𝑖) is assumed in the Grossman and Rossi-Hansberg (2008) framework. Hence, using a
parametric estimation method and attaching any specific functional form to the offshoring
cost function 𝑡(𝑖) for all ten occupational groups in empirical estimation will likely result in
misspecification problems. Instead the non-parametric cubic spline method, in particular, the

79

monotonic cubic spline interpolation method is adopted to approximate the offshoring cost
function 𝑡(𝑖).
Once parameterization of 𝑡(∙) is resolved, estimation of equation (3.8) becomes a
standard non-linear estimation problem. The NLS estimators
𝑁
𝑆
𝜽 = 𝑚𝑖𝑛 𝑁 −1 𝑆 −1 ∑ 𝑜=1 ∑ 𝑠=1{𝑦 𝑜,𝑠 − 𝑚(𝒙 𝑜,𝑠 , 𝜽)}

2

𝜃∈Θ

(3.9)

minimize the sum of least squared residuals of the sample average and should solve the
sample minimization problem if the true parameters

𝜽 𝟎 = 𝑎𝑟𝑔𝑚𝑖𝑛 𝜃∈Θ 𝐸 {[ 𝑦 −

𝑚( 𝑥, 𝜽)]2 } solve the population minimization problem.
Ideally we would estimate Equation (3.8) occupation by occupation to identify the
initial employment without offshoring

𝐿 𝑜 at home country, the constant parameter 𝑐 𝑜 and

the set of parameters for each occupation 𝑜 in the parameterized offshoring cost function
𝑡(𝑖). Due to data restrictions, 7 the individual occupations are grouped into ten broad
occupational groups for pre- and post-2000 periods respectively and these groups are used as
the basis to estimate Equation (3.8). 8

3.3.2 Application of Monotonic Cubic Spline Interpolation Method
A two-step monotonic cubic spline interpolation procedure is used to estimate

̂=
𝜽

(𝐿 𝒐 , 𝑐 𝒐 , 𝑡( 𝑖)) for the ten occupational groups based on the algorithm of monotonic cubic

7

See data description for details.

8

To distinguish, subscript 𝒐 (bold italic) is used to represents an occupational group in the
remaining of this paper.
80

spline interpolation developed by Wolberg and Alfy (1999, 2002). Figure 3.1 illustrates an
example of monotonic cubic spline: the interpolating cubic spline passing through its control
points is smooth and monotonic. While it is not yet often used in the field of economics,
monotonic cubic spline interpolation is a well-developed method and widely used in
numerical and statistical data analysis to solve engineering problems. The most compelling
reason for the use of cubic polynomials is the property of twice differentiable continuity,
which guarantees continuous first and second derivatives across all intervals. The goal of
cubic spline interpolation is to determine the smoothest possible curve that passes through
designated control points while simultaneously preserving the property of piecewise
monotonicity within each interval.

81

Figure 3.1: An Example of Cubic Spline Interpolation

Source: Wikipedia (http://en.wikipedia.org/wiki/Monotone_cubic_interpolation).
(For interpretation of the references to color in this and all other figures, the reader is
referred to the electronic version of this dissertation.)

The algorithm of Wolberg and Alfy (2002) is adopted in the first step. The
algorithm itself consists of two parts designated WAA Step-1 (abbreviation of Wolberg and
Alfy Algorithm) and WAA Step-2 to distinguish from the overall two-step interpolation
procedure and avoid confusion. The WAA Step-1 attempts to find a twice continuously
differentiable cubic spline which minimizes the modified second derivative discontinuity in
the spline.9 If a twice continuously differentiable cubic spline exists, the WAA Step-2 is
then employed to obtain the optimal twice continuously differentiable cubic spline by

−
+
Definition of second derivative discontinuity: ∑ 𝑖 [ 𝑓 ′′ ( 𝑥 𝑖 ) − 𝑓 ′′ ( 𝑥 𝑖 )]2 . Definition of
modified second derivative discontinuity: summation of second derivative difference is non−
+
−
negative, i.e., ∑ 𝑖 [ 𝑓 ′′ ( 𝑥 𝑖 ) − 𝑓 ′′ ( 𝑥 𝑖 ) + 𝐾 ] ≥ 0, where 𝐾 satisfies 𝑓 ′′ ( 𝑥 𝑖 ) −
𝑓 ′′ ( 𝑥 + ) + 𝐾 ≥ 0 for any arbitrary 𝑖. The reason to use modified second derivative
𝑖
discontinuity is to turn the objective function into a linear function so that linear
programming can be applied. See Wolberg and Alfy (2002) for details.
9

82

computing the integral of the spline curvature. If not, the best first differentiable cubic
spline is obtained in the WAA Step-1 and the WAA Step-2 is canceled.
To estimate 𝜽, the offshoring percentage interval 𝑖 ∈ [0, 1] is partitioned into ten
equal sub-intervals, representing the percentage increment of 𝑖 being offshored. The WAA
Step-1 and WAA Step-2 are applied to locally approximate the offshoring cost function 𝑡(𝑖)
and obtain the monotonic cubic spline interpolation for each occupational group. The
interpolated offshoring cost function is then used to calculate Ω (

in Equation (3.9). Then the optimal estimators of
linear least square errors by iterations.

Io,s (𝐿 𝑜,𝑠 , 𝐿 𝑜, 𝑡(∙)))

̂ is obtained by minimizing the non𝜽

̂ is a vector containing 13 estimators. They are
𝜽

estimator of the initial employment of occupation 𝑜 at home country without offshoring
estimator of the constant parameter

𝑐̂ 𝒐

̂ 𝒐,
𝐿

and the set of estimators for parameterized

offshoring cost function 𝑡(𝑖) , which corresponds to 11 control points that portioned 𝑖 ∈
[0, 1] into ten equal sub-intervals.
3.3.3 Estimating Offshoring Cost Functions for the Ten Major Occupational Groups
Implementation of the monotonic cubic spline approximation to estimate offshoring
cost functions for the ten major occupational groups requires updating the initial values of
the cost function 𝑡(𝑖) at each control point of 𝑖 . 10 Hence initial starting values for 𝑡(∙)
must be obtained. Blinder and Krueger (2009) provide values for offshorability in major
occupational groups11 as the starting point to differentiate relatively offshorable occupations

10

The 11 control points of 𝑖 are 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8., 0.9, 1.

11

See Table 2, Column 5, titled Externally-Coded Percent Offshorable in Blinder and
83

from relatively non-offshorable occupations.12 Based on their externally-coded estimates,
the ten occupational groups are divided into two broad categories: Offshorable Groups and
Non-offshorable Groups (Table 3.1).
Table 3.1: Offshorablility in Major Occupational Groups
Rank of Offshorablity
1
2
3
4
5

6
6
6
7
8

Occupational
Group
(Externally-coded
Percentage)
Offshorable Groups

Offshorable

G9: Production occupations13 (80.7%)
G5: Office and administrative support occupations (41.2%)
G2: Professional and related occupations (20.5%)
G4: Sales and related occupations (17.8%)
G1: Management, business, and financial operations
occupations (16.4%)
Non-offshorable Groups
G6: Farming, fishing, and forestry occupations (0.0%)
G7: Construction and extraction occupations (0.0%)
G10: Transportation and material moving occupations (0.0%)
G3: Service occupations (0.7%)
G8: Installation, maintenance, and repair occupations (1.3%)

Notes: Prepared by authors based on the externally-coded offshorable percentage (Column 2,
Table 2) in Blinder and Krueger (2009).

Adjustment of employment size for each occupation within an occupational group is
necessary before the monotonic cubic spline approximation is applied to estimate the
offshoring cost functions for the ten major occupational groups. There are large betweenoccupation variations in employment within a same occupational group (Table 3.4).

Krueger (2009).
12
There are sharp disagreements between self-classified and externally coded offshorability
for some occupational groups. This research uses the externally-coded offshorability by
professionals as the criterion to divide offshorable and non-offshorable groups.
13
For the purpose of simplicity and comparability with Blinder and Krueger (2009)’s
results, only post-2000 occupation titles are used to indicate occupational groups in the main
text unless otherwise specified.
84

However, by estimating Equation (3.8) at the basis of occupational groups, the to-beidentified parameter

𝐿 𝒐 (i.e., the initial total employment without offshoring) is implicitly

assumed to be same for all occupations within a group. This is a relatively strong
assumption for the ten occupational groups with large between-occupation variations in
employment within each occupational group. In order to identify
meaningful

𝐿 𝒐 and obtain a

̂ 𝒐 for each occupational group, this study adjusts employment size to make
𝐿

employment size for each occupation relatively homogenous within an occupational
group.14

As

̂ 𝒐 (i.e., estimated total employment without offshoring) is heavily dependent on
𝐿

the within-group variations of adjusted occupational employment

̃ 𝑜,𝑠 between different
𝐿

occupations, offshoring percentage is restricted to not exceed 10% of the maximum

̃ 𝑜,𝑠
𝐿

(i.e., the maximum adjusted employment of all occupations across all years in the sample
period) in each occupational group. In other words, the estimated

̂𝒐
𝐿

is restricted to

̂𝒐≤
𝐿

̃
1.1 ∗ 𝑚𝑎𝑥(𝐿 𝑜,𝑠 ) . This restriction is also used as one stopping criterion for iterations
when applying monotonic cubic spline interpolation to approximate the offshoring cost
functions for the ten occupational groups.

14

Detailed adjustment method of employment size is discussed in Section 3.4 after
introducing the data set.
85

3.3.4 Estimating Number of Jobs Offshored and Offshoring Percentage for the Five
Relatively Offshorable Occupational Groups
After estimation of the offshoring cost functions for the ten major occupational
groups, this analysis is focused on the five relatively offshorable occupational groups in
Table 3.1. Nonlinear least squares (NLS) with a specific cubic functional form

𝑡( 𝑖 ) =

𝑎𝑖 3 + 𝑏𝑖 2 + 𝑑𝑖 + 𝑒 is employed to re-estimate the offshoring cost functions for the
five relatively offshorable occupational group. The number of jobs offshored and the
offshoring percentage by detailed occupation in pre- and post-2000 sample period are
calculated after estimation of the cubic offshoring cost function. There are a few reasons to
focus on the relatively offshorable occupational groups. First, offshoring cost is relatively
tractable because fluctuations of employment at offshorable occupations over time observed
in data reflect the change of offshoring costs. Second, factors (e.g., technology, institutional
restructuring) that could potentially affect the occupational employment except offshoring
are not controlled in this study. In other words, changes of employment over time are
assumed to be purely attributable to offshoring in this framework. While this is a strong
assumption, it is more realistic for the relatively offshorable occupations which are primary
focus of this study.
To calculate the number of jobs offshored and the offshoring percentage for the five
offshorable occupational groups over the pre- and post-2000 sample period, this study uses

̂ 𝒐 , which is estimated from the adjusted employment size ̃ 𝑜,𝑠
𝐿
𝐿
the two-step cubic spline interpolation method, to recover

of each occupation from

̃ 𝒐 , the unadjusted initial
𝐿

employment without offshoring for each occupational group by reversing the adjustment
method.

86

Different scenarios are applied when re-estimating the cubic offshoring cost
functions using NLS, calculating the number of jobs offshored and offshoring percentage for
the five offshorable occupational groups. Based on Blinder and Krueger’s (2009) estimates
for offshorable occupational groups (re-organized in Table 3.1), the 20% scenario is chosen
as a benchmark case for all five groups because the externally coded offshorability are
relatively close to 20 percent (Group 1, Production occupations, 16.4%; Group 2,
Professional and related occupations, 20.5%; Group 4, Sales and related occupations,
17.8%). In the 20% scenario, the offshoring percentage is assumed to not exceed 20 percent
of the maximum

̃ 𝑜,𝑠 , i.e., the estimated ̂ 𝒐 ≤ 1.2 ∗ 𝑚𝑎𝑥(𝐿 𝑜,𝑠 ). The maximum
̃
𝐿
𝐿

offshoring percentage is then gradually relaxed to 40 percent (externally coded offshorable
percentage is 41.2 percent for Group 5, Office and Administrative Support Occupations) and
80 percent (externally coded offshorable percentage is 80.7% for Group 9, Production
Occupations) for all five offshorable groups.

3.4 Data Description and Adjustment
The CPSMORG (Current Population Survey Merged Outgoing Rotation Groups)
data from years 1983 to 2011 are used to implement the two-step monotonic cubic spline
interpolation procedure. The data are discontinuous due to a complete switch in the
occupational and industrial classification system in CPS (Current Population Survey) in
2003.15 This substantial change in the composition of detailed occupations between the
1980 and 2002 occupation codes makes linking data by occupation codes impossible.
Hence, the sample is divided into two periods: pre-2000 (1983-1999) and post-2000 period
(2000-2011) to conduct analysis at occupational level.

15

Years 2000-2002 are dual-coded in both 1980 and 2002 census classifications systems.
87

Observations for individuals with age less than 18 and or more than 65 are dropped
from the sample to maintain focus on the labor force. Hourly wage series for each individual
is created following Schmitt 2003 and inflated by 2000 CPI index to obtain the real hourly
wage. Wage and employment are aggregated to occupation level based on 1980 census
codes for the pre-2000 period and based on 2002 census codes for the post-2000 period.
CPS earning weights are used to obtain occupational hourly wage while CPS final weights
are used to obtain occupational employment during aggregation. To maintain balanced
panels for both the pre- and post-2000 periods, occupations not present in all years of each
analysis period were omitted from the data set. After aggregation, there are 486 occupations
in the pre-2000 period and 460 occupations in the post-2000 period (Table 3.2.1 and 3.2.2).
As mentioned earlier, by estimating offshoring cost functions by occupational groups

𝐿 𝒐 is implicitly assumed to be same for all occupations within an occupational group. But
the large between-occupation variations in employment within an occupational group is not
in favor of this assumption. Several adjustments are made to reduce between-occupation
variations and homogenize the employment size within each occupational group.
For both pre- and post-2000 sample period, mean employment for each occupation is
calculated and a median employment for all occupations within an occupational group is
obtained. Relative employment size for each occupation is mean employment of each
occupation by this occupational group median employment.16 Finally, the adjusted
employment for each occupation in each year

̃ 𝑜,𝑠 is observed employment 𝐿 𝑜,𝑠 divided by
𝐿

the relative employment size of each occupation. The adjusted employment for each

16

If there are even-numbered groups within an occupational group, we use the larger of the
two medians as the denominator.
88

occupation

̃ 𝑜,𝑠 is used in the monotonic cubic spline interpolation to approximate the
𝐿

offshoring cost function.
The estimated
adjusted employment

̂ 𝒐 largely depends on the maximum or minimum value of the
𝐿
̃ 𝑜,𝑠 within each occupational group. The estimated ̂ 𝒐 is likely to be
𝐿
𝐿

misleadingly inflated if there are extreme values of

̃ 𝑜,𝑠
𝐿

within an occupational group.

Hence, occupations with observations falling in the upper and lower five percentile of the
adjusted employment are dropped to further homogenize the employment size of
occupations within each occupational group. Table 3.4 summarizes the employment size
variations for each occupational group before and after adjustment for pre- and post-2000
periods respectively. After adjustment, the mean and median employment size within each
occupational group are quite close. The between-occupation employment variations within
an occupational group are largely reduced.

3.5 Results and Discussion
3.5.1 Offshoring Costs for the Ten Major Occupational Groups
The 11 point estimates of the parameterized offshoring cost functions from the
monotonic cubic spline interpolation method for the ten major occupational groups are
summarized in Table 3.5. A corresponding interpolated offshoring cost function 𝑡(𝑖) for
each occupational group are plotted in Figure 3.2.1 and Figure 3.2.2 for the pre- and post2000 periods respectively. The estimated

𝑐̂ 𝑜

and

̂ 𝑜 are reported in Table 3.6.
𝐿

One issue to be emphasized in front is that any direct comparison between pre- and
post-2000 periods is not feasible although results for the pre- and post-2000 periods are
89

sometimes displayed in parallel. As mentioned earlier, compositions of occupations within
each occupational group for pre- and post-2000 periods are completely different.
Consequently, the estimated

̂𝑜
𝐿

̂ 𝑜 in pre-2000 period is not comparable with the estimated
𝐿

in post-2000 period due to this occupational composition difference. Nonetheless,

results from these two sample periods are consistent and some general patterns can be
observed. Estimated offshoring cost functions indicate an effect of economies of scale in
offshoring. The offshoring cost increases in the first ten percent of offshoring and then
decreases as more jobs offshored.17 Among the ten occupational groups, Group 1
(Management, business, and financial operations occupations), Group 2 (Professional and
related occupations), Group 4 (Sales and related occupations), Group 5 (Office and
administrative support occupations) and Group 9 (Production occupations) have relatively
lower costs at any given level of offshoring percentage 𝑖 in both the pre- and post-2000
periods. In particular, production occupations in Group 9, which are commonly regarded to
contain the most impersonal and/or routine tasks and easiest to offshore, have the lowest
offshoring costs when the offshoring percentage is below 40 percent (Table 3.4). The
estimated offshoring cost for production occupations has a sharp increase when offshoring
moves from the first 40 percent to 50 percent in the pre-2000 period, and from the first 30
percent to 40 percent in the post-2000 period. The remaining five occupational groups,
Group 3 (Service occupations), Group 6 (Farming, fishing, and forestry occupations),18

17

The partition of 10 subintervals is arbitrary. But increasing the numbers of subintervals
does not alter the result because the nice monotonic property of monotonic cubic spline
interpolation within each interval.
18
Group 6 has low offshoring cost (small point estimates) in the first 30 percent of
offshoring due to few observations between interval 0.0 and 0.3.
90

Group 7 (Construction and extraction occupations), Group 8 (Installation, maintenance, and
repair occupations) and Group 10 (Transportation and material moving occupations), have
relatively higher offshoring costs.
The rank of offshorability in this study based on estimated offshoring costs for both
pre- and post-2000 periods is different from the externally coded offshorability of Blinder
and Krueger (2009) based on individual telephone survey in 2008, but most results are
consistent with them. Blinder and Krueger (2009) found Group 6 (Farming, fishing, and
forestry occupations), Group 7 (Construction and extraction occupations) and Group 10
(Transportation and material moving occupations) to be the least offshorable. This study
identified farming, fishing, and forestry occupations (Group 6), construction and extraction
occupations (Group 7), and service occupations (Group 3) with the highest offshoring costs
while transportation and material moving occupations in Group 10 with the second highest
offshoring cost.
3.5.2 NLS Results for the Five Relatively Offshorable Occupational Groups
The estimated coefficients of the cubic offshoring cost function together with

𝑐̂ 𝑜 and

̂ 𝑜 by NLS method under three different scenarios are reported in Table 3.5.
𝐿
Corresponding offshoring cost functions 𝑡(𝑖) of the five relatively offshorable groups are
displayed respectively in Figure 3.3.1 through Figure 3.3.5. Unlike the monotonic cubic
spline interpolation method, it is difficult to directly compare the estimated cubic offshoring
cost functions among different occupational groups within the same scenario, or the same
occupational group among three different scenarios given the fact that the single cubic
functional form attached to all five relatively occupational groups cannot be uniquely
identified because there is only one moment condition (i.e., Eq. 3.6) available in the
structural model.
91

The number of jobs offshored and the offshoring percentage are calculated based on
the estimated

̂ 𝑜 for the five relatively offshorable groups in pre- and post-2000 periods are
𝐿

summarized in Table 3.8.1 through Table 3.8.5. First, the initial total employment for each
occupation 𝑜 in each year 𝑠 within an offshorable occupational group is recovered by
multiplying the relative employment size of each occupation to its corresponding

̂𝒐
𝐿

estimated for each occupational group. The number of jobs offshored at each occupation 𝑜
in each year 𝑠 is then the difference between the recovered initial total employments of
occupation 𝑜 and

𝐿 𝑜,𝑠 observed in data. The offshoring percentage is then obtained using

the number of jobs offshored divided by the initial total employment without offshoring.
Both the number of jobs offshored and offshoring percentage increase as the
maximum offshoring capacity increases from 20% scenario to 80% scenario. Production
occupations in Group 9 have been consistently increasing over time in both pre- and post2000 periods under all three scenarios. Under the 20% scenario that maximum 20 percent of
production occupations are offshorable, the offshoring percentage for production
occupations increases from 36.5 to 46.3 percent in the pre-2000 period and increases from
36.1 to 48.5 percent in the post-2000 period. Under the 40% scenario, offshoring percentage
for production occupations increases from 45.6 to 54.0 percent in the pre-2000 period and
increases from 45.2 to 55.9 percent in the post-2000 period. Under the 80% scenario,
offshoring percentage for production occupations increases from 57.4 to 64.0 percent in the
pre-2000 period and from 52.9 to 62.1 percent in the post-2000 period, which are less than
the estimated 80.7 percent by Blinder and Krueger (2009).
Changes in the offshoring percentage for the five relatively offshorable occupational
groups over the two sample periods are depicted in Figure 3.4.1 through Figure 3.4.5
92

additionally. Offshoring percentage for sales and related occupations in Group 4 and office
and administrative support occupations in Group 5 have been gradually increasing over time
in the post-2000 period. On the other hand, for management, business and financial
operations occupations (Group 1) and professional and related occupations (Group 2),
offshoring percentage actually has decreased over time.
In addition, using externally coded offshorability estimated for the five offshorable
groups from Blinder and Krueger (2009) (reorganized in Table 3) as a criterion, results of
the 20% scenario for Group 1, Group 2 and Group 4, the 40% scenario for Group 5, and the
80% scenario for Group 9 to are selected to make comparisons among groups. This
comparison shows that occupations in sales and related occupations in Group 2 are the least
offshorable among the five offshorable occupational groups followed by the management,
business and financial operations occupations and professional and related occupations.

93

3.6 Conclusion
This research generalizes the Grossman and Rossi-Hansberg (2008) offshoring
model to include numerous tasks/skill levels. This generalization allows a possible and
direct linkage between the theoretical task offshoring model and occupational data that can
be aggregated from the CPSMORG (Current Population Survey Merged Outgoing Rotation
Groups) data from year 1983 to 2011. Empirical investigation of the effect of offshoring on
occupational employment for ten major occupational groups (at 2-digit SOC level) in the
U.S. labor market is conducted by estimating their offshoring cost functions using a nonparametric monotonic cubic spline interpolation method. Based on the estimated offshoring
costs, five relatively offshorable occupational groups are identified including production
occupations, office and administrative support occupations, sales and related occupations,
professional and related occupations, and management, business, and financial operations
occupations.
Motivated by the practical issue of difficulty in obtaining a time-variant
offshoring/offshorability index faced by majority empirical studies interested in identifying
the effect of offshoring, this study calculates the number of jobs offshored and the
offshoring percentage under the NLS method for the five relatively offshorable occupational
groups under different scenarios. Calculated offshoring percentage provides time-variant
offshoring indices for more than 300 major detailed occupations in these five relatively
offshorable groups that can be employed in other empirical studies.
The results of this research indicate that offshoring percentage for each
occupational group may vary under different scenarios, but the evolution pattern is
consistent. Production occupations are the most offshorable while sales and related
occupations are the least offshorable among all five offshorable occupational groups under
all three scenarios. The offshoring percentage for production occupations has been
94

increasing in both pre- and post-2000 periods while the offshoring percentages for
professional and related occupations, and management, business, and financial operations
occupations have been decreasing over time.

95

APPENDIX

96

Table 3.2.1: Major Occupational Groups in Pre-2000 Period (1983-1999)
Group

1980 Census Codes

1

003-037

2

043-199
203-235
403-469
243-285
303-389
473-499
553-599
613-617
503-549
633-699
703-799

3
4
5
6
7
8
9
10

803-889

Occupation Title
Managerial and professional Specialty
occupations
Professional specialty occupations
Technical occupations
Service occupations
Sales occupations
Administrative support occupations
Farming, forestry, and fishing occupations
Construction trades
Extractive occupations
Mechanics and Repairers
Precision Production Occupations
Operators, fabricators, and laborers
Transportation
and
Material
Moving
Occupations

Total

Number of
Occupations
24
126
42
23
55
19
35
27
99
39
486

*Notes: Occupational group information is obtained from
(http://usa.ipums.org/usa/volii/98occup.shtml), but reorganized and reordered
by author to be comparable with occupational groups in post-2000 period.

Table 3.2.2: Major Occupational Group in Post-2000 Period (2000-2011)
Group 2002 Census Codes
1

0010-0950

2
3
4

1000-3540
3600-4650
4700-4960

5

5000-5930

6
7

6000-6130
6200-6940

8

7000-7620

9

7700-8960

10

9000-9750

Occupation Title
Management, business, and financial
operations occupations
Professional and related occupations
Service occupations
Sales and related occupations
Office and administrative support
occupations
Farming, fishing, and forestry occupations
Construction and extraction occupations
Installation, maintenance, and repair
occupations
Production occupations
Transportation and material moving
occupations

Total

Number of
Occupations
42
107
57
17
50
8
36
34
75
34
460

*Notes: Occupational groups are equivalent to those grouped at 2-digit SOC level.

97

Table 3.3: Occupational Employment Size1 Variation
Pre-2000 Period: 1983-1999

Gro
up
1
2
3
4
5
6
7
8
9
10
Gro
up
1
2
3
4
5
6
7
8
9
10

̅̅̅̅̅
̃
After Adjustment2 (𝐿 𝑜,𝑠 )
Median
Mean
Max
Min
Median
Mean
234,885 509,254
336,438 168,371 257,054 255,055
60,309 155,601
85,892
35,756
60,110
60,309
218,522 433,122
295,584 147,323 224,454 222,077
239,788 624,399
296,549 197,353 247,542 245,897
186,986 368,996
260,335 117,224 188,326 186,986
39,952 154,177
61,434
22,577
40,062
39,952
42,083 135,112
55,790
27,795
42,113
42,083
101,811 161,553
131,050
76,843 101,989 101,811
42,481 121,924
59,714
25,258
42,741
42,689
87,280 289,827
54,206 120,427
87,752
7,280
Post-2000 Period: 2000-2011

Before Adjustment
Max
5,179,799
1,829,530
2,463,471
3,470,040
4,064,116
1,173,238
1,103,129
746,818
1,409,946
2,780,569

Min
10,103
2,439
11,309
16,358
4,676
1,894
3,192
3,043
2,775
3,098

̅̅̅̅̅
(𝐿 𝑜,𝑠 )

̅̅̅̅̅
Before Adjustment (𝐿 𝑜,𝑠 )
Max
2,402,506
2,819,706
2,274,862
3,548,378
3,507,671
868,469
1,440,582
766,161
1,185,664
3,089,586

Notes:

Min
8,412
2,485
4,214
34,789
5,690
2,253
3,107
2,940
3,646
3,943

Median
180,133
80,408
115,561
325,546
150,514
20,680
45,802
48,820
40,289
52,905

Mean
359,564
202,254
377,052
851,402
388,964
154,201
182,013
126,207
113,951
257,220

̅̅̅̅̅
̃
After Adjustment (𝐿 𝑜,𝑠 )
Max
Min
Median
Mean
231,030 138,907 181,858 181,400
106,722
55,535
81,007
81,271
145,055
81,288 116,166 115,561
386,948 250,422 323,554 325,546
214,546 112,681 158,846 159,529
29,347
11,050
20,308
20,680
66,277
27,194
46,395
46,183
65,622
31,711
49,031
49,308
60,173
21,929
39,664
40,289
76,506
34,108
52,943
53,229

1. For each occupation, employment is averaged across years within sample period.
2. Occupations with employment falling in the upper and lower five percentile are
dropped after adjustment.

98

Table 3.4: Point Estimates of Parameterized Offshoring Cost Function
Spline Interpolation Method for Ten Major Occupational Groups
Gro
up
1
2
3
4
5
6
7
8
9
10

𝑡(𝑖) from Cubic

Value of 𝑡(𝑖) at Control Points 𝑖
Pre-2000: 1983-1999
0
5.33
1.12
14.8
0.01
10.2
7.44
4.19
0.40
8.24
11.7

0.1
3.38
2.65
6.05
6.60
1.18
12.0
9.27
0.20
0.14
9.35

0.2
11.5
10.3
6.04
12.6
5.36
19.6
17.4
16.4
1.07
19.1

0.3
25.8
15.4
22.2
20.8
24.5
40.0
25.2
16.3
7.34
42.8

0.4
38.4
32.5
49.3
40.5
41.2
51.1
75.4
73.5
18.9
74.3

0.5
49.6
71.9
173.7
56.1
45.8
101.8
52.3
110.6
53.8
182.2

0.6
70.3
111.5
252.7
87.8
55.2
119.0
92.7
128.8
103.5
279.1

0.7
77.7
137.7
372.5
120.5
83.1
170.5
136.0
162.2
132.0
424.0

0.8
95.6
130.7
297.8
185.8
107.3
249.8
193.1
265.8
151.5
551.8

0.9
137.3
124.4
452.8
167.5
136.5
369.4
307.4
356.5
182.6
766.9

1.0
202.1
108.8
259.4
144.5
145.7
505.1
448.7
551.8
166.0
1006.3

91.6
227.4
543.5
104.5
101.7
247.8
264.8
144.7
85.6
121.8

121.1
175.7
747.3
120.1
133.4
222.3
363.0
182.2
117.3
195.2

188.1
460.9
1076.9
184.9
168.8
262.3
543.6
269.8
134.2
241.1

Value of 𝑡(𝑖) at Control Points 𝑖
Post-2000 Period: 2000-2011
1
2
3
4
5
6
7
8
9
10

2.71
5.98
20.3
0.00
3.18
20.4
2.21
5.67
0.54
17.9

6.35
2.02
18.3
0.38
1.04
23.5
27.8
7.27
0.09
11.1

12.5
4.88
57.8
13.0
0.88
25.4
26.6
2.13
0.90
9.83

20.4
9.74
73.5
29.3
38.3
35.8
40.8
18.2
41.1
53.7

30.5
31.9
142.5
29.6
35.5
53.7
23.2
34.6
20.7
80.0

42.6
113.0
233.0
53.2
47.0
77.8
82.2
40.9
69.8
99.9

58.43
147.2
180.9
61.3
64.4
109.2
116.5
52.5
103.4
132.7

77.38
197.9
404.9
86.4
85.4
168.0
203.8
92.0
95.5
170.6

Notes: 1. No other control variables are included in the model.
2. The upper bound is set that offshoring cannot exceed the 10% of the maximum
employment of all occupations across all years within each group.
3. The maximum iterations is 500 times.
4. Initial value is adopted from the first-round cubic spline interpolation results
without dropping any observations. See Table 3.A for details of initial value.

99

Table 3.5: Estimates of ̂ 𝑜 , 𝑐̂ 𝑜 by Occupational Groups from Cubic Spline Interpolation
𝐿
Method for Major Ten Occupational Groups
Pre-2000 Period (1983-1999)
Group
1
2
3
4
5
6
7
8
9
10

̂𝑜
𝐿

̂𝑜
𝐿

𝑐̂ 𝑜

370,034
89,912
323,423
314,449
284,791
67,578
61,369
137,680
62,628
367,550

Post Post-2000 Period (2000-2011)

5.86
6.69
2.72
3.74
3.27
1.85
4.21
5.05
3.40
5.95

𝑐̂ 𝑜

245,310
110,980
152,963
425,643
225,284
29,860
69,592
69,206
66,190
84,049

Notes: 1. No other control variables are included in the model.
2. The upper bound is set that offshoring cannot exceed the 10% of the maximum
employment of all occupations across all years within each group.
3. The maximum iterations is 500 times.
4. Initial value is adopted from the first-round cubic spline interpolation results
without dropping any observations. See Appendix Table 3.A for details of initial
value.

100

7.00
6.51
3.82
5.05
3.77
3.43
4.98
5.50
3.13
4.20

Table 3.6: NLS Estimates of Cubic Offshoring Cost Function

̂ 𝒐 ,𝑐̂ 𝒐 for the Five Relatively Offshorable Occupational Groups
𝐿

Pre-2000 (1983-1999)
Group

Coefficients of Cubic

̂
𝑎

̂
𝑏

̂
𝑑

𝑡(𝑖)
𝑒̂

Post-2000 (2000-2011)

̂𝑜
𝐿

1
2
4
5
9

-3875 -4341
-4062 -4267
-2085 -2335
-16.91 -17.49
-5686 -6249

-1678
-2823
-896
-7.04
-2337

348.2
549
195
1.54
624

403,726
103,070
355,858
312,402
71,622

1
2
4
5
9

5136 5783
2021 1801
-0.001 -0.001
-460
-543
3392 3706

2149
966
-0.0003
-403
1421

-922.8
-394
0.0002
145
-685

471,014
120,249
414,872
364,469
83,600

1813
-2649
0.79
284
2205

-1037
1658
-0.59
-110
-1330

543,488
141,377
533,774
449,143
106,827

1
2
4
5
9
Notes:

4639
-6880
2.09
133
5430
1.
2.
3.
4.

5217
-7702
2.45
334
6137

Coefficients of Cubic

𝑐̂ 𝑜

̂
𝑎

20% Scenario
5.03 16397
4.94
-8.77
2.88 -1733
2.85
2272
2.72
17.92
40% Scenario
4.02
1578
3.72 -1549
2.25
2511
2.18 -1922
2.16
4024
80% Scenario
3.61 -4753
3.73
378
2.06 -2316
2.11
3589
2.10
2338

̂
𝑏

̂
𝑑

𝑡(𝑖)
𝑒̂

̂𝑜
𝐿

𝑐̂ 𝑜

19786
-9.30
-1916
2582
20.86

5933
-3.55
-754
972
7.23

-1380
0.81
160
-205
-1.63

277,236
128,066
464,337
257,455
72,208

5.63
5.24
4.30
3.09
3.00

5273
-1910
2795
-2181
5176

218
-723
1042
-809
1581

-407
338
-462
344
-779

323,442
149,410
541,727
300,364
84,242

4.46
3.99
3.24
2.50
2.42

-5446
917
-2570
4001
2662

-1848
63
-819
1382
917

1115
-117
547
-801.8
-540.2

390,700
181,275
650,472
366,463
98,066

4.01
3.84
3.04
2.24
2.29

No other control variables are included in the model.
Occupations with employment falling in the upper and lower five percentile are dropped after adjustment.
The maximum iterations is500 times.
Initial value is adopted from the first-round cubic spline interpolation results without dropping any observations. See Table 3.A for
details of initial value.

101

Table 3.7.1: Calculated Number of Jobs Offshored and Offshoring Percentage for Group 1
(Management, Business and Financial Operations Occupations) from NLS Method
Year

Pre- 1983
2000 1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
Post- 2000
2000 2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011

20% Scenario
40% Scenario
80% Scenario
No. of Jobs Offshore No. of Jobs Offshore No. of Jobs Offshore
Offshored
Percentag Offshored
Percenta Offshored
Percentag
e
ge
e
7,771,465
47.6%
10,400,000
55.1% 13,200,000
61.1%
7,085,993
45.8%
9,713,385
53.5% 12,500,000
59.7%
6,707,947
46.1%
9,335,339
53.8% 12,200,000
60.0%
6,292,123
42.1%
8,919,514
50.3% 11,700,000
57.0%
5,989,792
40.0%
8,617,183
48.5% 11,400,000
55.4%
5,500,926
37.7%
8,128,317
46.6% 11,000,000
53.7%
4,891,834
35.6%
7,519,225
44.8% 10,300,000
52.2%
4,902,728
34.8%
7,530,119
44.1% 10,400,000
51.6%
4,671,854
34.7%
7,299,245
44.0% 10,100,000
51.5%
6,299,921
36.2%
8,927,313
45.4% 11,800,000
52.6%
6,233,880
32.9%
8,861,271
42.5% 11,700,000
50.2%
6,354,836
35.1%
8,982,227
44.3% 11,800,000
51.8%
5,822,961
33.1%
8,450,352
42.6% 11,300,000
50.3%
5,651,819
33.3%
8,279,210
42.9% 11,100,000
50.5%
5,185,593
32.9%
7,812,984
42.5% 10,600,000
50.2%
4,869,771
30.6%
7,497,162
40.5% 10,300,000
48.4%
4,454,344
27.5%
7,081,736
37.9%
9,911,625
46.2%
7,696,804
37.5%
11,100,000
46.4% 16,000,000
55.6%
7,312,754
35.2%
10,700,000
44.4% 15,600,000
54.0%
7,152,252
36.1%
10,500,000
45.2% 15,500,000
54.7%
7,336,184
37.0%
10,700,000
46.0% 15,700,000
55.3%
7,286,110
35.7%
10,700,000
44.9% 15,600,000
54.4%
7,108,696
34.0%
10,500,000
43.4% 15,400,000
53.2%
6,919,571
34.1%
10,300,000
43.5% 15,300,000
53.2%
6,652,990
33.0%
10,000,000
42.6% 15,000,000
52.5%
6,526,403
34.0%
9,920,160
43.4% 14,900,000
53.1%
6,630,334
32.5%
10,000,000
42.2% 15,000,000
52.1%
6,953,488
32.1%
10,300,000
41.8% 15,300,000
51.9%
6,892,056
33.6%
10,300,000
43.1% 15,200,000
52.9%

Notes: 1. The number of job offshored is the sum of job offshored across all occupations
within an occupational group.
2. The offshoring percentage is the average offshoring percentage across all
occupations within an occupational group.
3. A cubic offshoring cost function is assumed.

102

Table 3.7.2: Calculated Number of Jobs Offshored and Offshoring Percentage for Group 2
(Professional and Related Occupations) from NLS Method
Year

Pre- 1983
2000 1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
Post- 2000
2000 2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011

20% Scenario
40% Scenario
80% Scenario
No. of Jobs Offshore No. of Jobs Offshore No. of Jobs Offshore
Offshored
Percentag Offshored
Percentag Offshored
Percentag
e
e
e
13,300,000
46.8% 17,800,000
54.4% 23,400,000
61.2%
13,200,000
46.6% 17,700,000
54.3% 23,300,000
61.1%
12,800,000
46.1% 17,300,000
53.8% 22,900,000
60.7%
12,500,000
44.8% 17,100,000
52.7% 22,600,000
59.8%
12,400,000
45.5% 16,900,000
53.3% 22,500,000
60.3%
12,000,000
43.1% 16,500,000
51.3% 22,100,000
58.5%
11,700,000
44.0% 16,300,000
52.0% 21,800,000
59.2%
11,200,000
40.9% 15,700,000
49.4% 21,300,000
56.9%
11,100,000
39.5% 15,700,000
48.2% 21,200,000
55.9%
11,000,000
41.0% 15,500,000
49.4% 21,100,000
57.0%
10,600,000
39.1% 15,100,000
47.8% 20,700,000
55.6%
10,900,000
40.1% 15,500,000
48.6% 21,000,000
56.3%
10,600,000
40.1% 15,100,000
48.7% 20,700,000
56.4%
10,100,000
38.9% 14,700,000
47.6%
20200,000
55.4%
9,780,010
37.5% 14,300,000
46.4% 19,900,000
54.4%
9,640,700
36.4% 14,200,000
45.5% 19,700,000
53.6%
8,816,018
34.8% 13,300,000
44.1% 18,900,000
52.5%
12,300,000
39.4% 17,300,000
48.1% 24,800,000
57.2%
11,800,000
38.3% 16,800,000
47.1% 24,200,000
56.4%
11,600,000
38.5% 16,600,000
47.3% 24,000,000
56.5%
11,500,000
37.8% 16,500,000
46.7% 24,000,000
56.1%
11,400,000
37.4% 16,400,000
46.3% 23,800,000
55.7%
11,200,000
37.2% 16,200,000
46.1% 23,600,000
55.6%
11,000,000
36.1% 16,000,000
45.2% 23,500,000
54.8%
10,400,000
35.5% 15,400,000
44.7% 22,900,000
54.4%
10,100,000
35.0% 15,100,000
44.3% 22,500,000
54.1%
9,978,409
34.9% 15,000,000
44.2% 22,400,000
54.0%
10,000,000
34.2% 15,000,000
43.6% 22,500,000
53.5%
9,962,497
34.2% 15,000,000
43.6% 22,400,000
53.5%

Notes: 1. The number of job offshored is the sum of job offshored across all occupations
within an occupational group.
2. The offshoring percentage is the average offshoring percentage across all
occupations within an occupational group.
3. A cubic offshoring cost function is assumed.

103

Table 3.7.3: Calculated Number of Jobs Offshored and Offshoring Percentage for Group 4
(Sales and Related Occupations) from NLS Method
Year

Pre- 1983
2000 1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
Post- 2000
2000 2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011

20% Scenario
40% Scenario
80% Scenario
No. of Jobs Offshore No. of Jobs Offshore No. of Jobs Offshore
Offshored
Percentag Offshored
Percentag Offshored
Percentag
e
e
e
4,489,843
35.1%
6,585,252
44.3% 10,800,000
56.7%
4,182,837
32.5%
6,278,245
42.1% 10,500,000
55.0%
4,178,781
32.4%
6,274,190
42.0% 10,500,000
54.9%
4,014,622
30.7%
6,110,031
40.6% 10,300,000
53.8%
3,805,164
29.2%
5,900,573
39.3% 10,100,000
52.8%
3,846,256
30.6%
5,941,665
40.5% 10,200,000
53.7%
3,670,658
30.7%
5,766,067
40.6%
9,987,918
53.8%
3,370,897
29.0%
5,466,306
39.1%
9,688,157
52.6%
3,407,134
30.8%
5,502,543
40.7%
9,724,394
53.9%
3,370,818
27.4%
5,466,226
37.8%
9,688,077
51.6%
3,440,809
27.9%
5,536,218
38.2%
9,758,068
51.9%
4,456,491
34.7%
6,551,899
44.0% 10,800,000
56.5%
4,284,704
32.7%
6,380,113
42.3% 10,600,000
55.1%
4,027,732
31.4%
6,123,140
41.1% 10,300,000
54.3%
3,933,802
31.0%
6,029,211
40.8% 10,300,000
54.0%
3,888,172
30.2%
5,983,581
40.2% 10,200,000
53.5%
4,005,604
28.9%
6,101,013
39.0% 10,300,000
52.6%
5,993,041
32.1%
9,110,449
41.8%
5,993,041
32.1%
5,701,304
28.8%
8,818,713
39.0%
5,701,304
28.8%
5,635,168
28.9%
8,752,576
39.1%
5,635,168
28.9%
5,805,051
29.9%
8,922,459
39.9%
5,805,051
29.9%
5,677,808
29.7%
8,795,216
39.7%
5,677,808
29.7%
5,556,994
30.0%
8,674,402
40.0%
5,556,994
30.0%
5,363,608
27.9%
8,481,016
38.2%
5,363,608
27.9%
5,251,183
28.0%
8,368,592
38.3%
5,251,183
28.0%
5480,845
29.9%
8,598,254
39.9%
5,480,845
29.9%
5,400,839
29.6%
8,518,247
39.7%
5,400,839
29.6%
5,567,462
31.5%
8,684,871
41.3%
5,567,462
31.5%
5,655,523
32.4%
8,772,932
42.0%
5,655,523
32.4%

Notes: 1. The number of job offshored is the sum of job offshored across all occupations
within an occupational group.
2. The offshoring percentage is the average offshoring percentage across all
occupations within an occupational group.
3. A cubic offshoring cost function is assumed.

104

Table 3.7.4: Calculated Number of Jobs Offshored and Offshoring Percentage for Group 5
(Office and Administrative Support Occupations) from NLS Method
Year

Pre- 1983
2000 1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
Post- 2000
2000 2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011

20% Scenario
40% Scenario
80% Scenario
No. of Jobs Offshore No. of Jobs Offshore No. of Jobs Offshore
Offshored
Percentag Offshored
Percentag Offshored
Percentag
e
e
e
10,200,000
41.9% 14,600,000
50.2% 21,800,000
59.6%
10,400,000
43.7% 14,800,000
51.7% 21,900,000
60.8%
10,200,000
43.1% 14,600,000
51.2% 21,700,000
60.4%
10,100,000
41.8% 14,500,000
50.1% 21,600,000
59.5%
9,897,058
40.4% 14,300,000
48.9% 21,400,000
58.5%
9,852,361
41.0% 14,200,000
49.5% 21,400,000
59.0%
9,901,249
39.6% 14,300,000
48.2% 21,400,000
58.0%
9,387,039
36.5% 13,800,000
45.6% 20,900,000
55.8%
9,582,273
36.2% 14,000,000
45.3% 21,100,000
55.6%
9,565,003
36.1% 13,900,000
45.2% 21,100,000
55.5%
9,722,870
35.8% 14,100,000
44.9% 21,200,000
55.3%
11,800,000
42.2% 16,200,000
50.5% 23,400,000
59.8%
11,800,000
42.5% 16,200,000
50.7% 23,300,000
60.0%
11,700,000
41.6% 16,100,000
49.9% 23,200,000
59.4%
11,800,000
40.4% 16,200,000
48.9% 23,300,000
58.5%
11,800,000
39.5% 16,200,000
48.2% 23,300,000
57.9%
11,800,000
40.3% 16,200,000
48.8% 23,300,000
58.5%
10,600,000
36.7% 15,400,000
45.7% 22,900,000
55.5%
10,600,000
38.6% 15,400,000
47.4% 23,000,000
56.9%
10,900,000
39.9% 15,800,000
48.5% 23,300,000
57.8%
11,000,000
36.6% 15,800,000
45.7% 23,300,000
55.5%
11,100,000
36.1% 16,000,000
45.2% 23,500,000
55.1%
11,100,000
35.3% 15,900,000
44.5% 23,400,000
54.5%
11,000,000
36.6% 15,900,000
45.6% 23,400,000
55.4%
11,200,000
38.8% 16,100,000
47.6% 23,600,000
57.0%
11,000,000
38.3% 15,900,000
47.1% 23,400,000
56.7%
11,500,000
39.9% 16,400,000
48.5% 23,900,000
57.8%
11,600,000
39.3% 16,500,000
48.0% 24,000,000
57.3%
12,000,000
40.3% 16,900,000
48.9% 24,400,000
58.1%

Notes: 1.The number of job offshored is the sum of job offshored across all occupations
within an occupational group.
2. The offshoring percentage is the average offshoring percentage across all
occupations within an occupational group.
3. A cubic offshoring cost function is assumed.

105

Table 3.7.5: Calculated Number of Jobs Offshored and Offshoring Percentage for Group 9
(Production Occupations) from NLS Method
Year

Pre- 1983
2000 1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
Post- 2000
2000 2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011

20% Scenario
40% Scenario
80% Scenario
No. of Jobs Offshore No. of Jobs Offshore No. of Jobs Offshore
Offshored
Percentag Offshored
Percentag Offshored
Percentag
e
e
e
6,709,289
36.5%
9,768,267
45.6% 15,700,000
57.4%
6,856,422
38.4%
9,915,399
47.2% 15,800,000
58.7%
6,862,317
37.9%
9,921,295
46.8% 15,900,000
58.4%
6,864,584
37.9%
9,923,561
46.8% 15,900,000
58.4%
6,944,952
38.7% 10,000,000
47.5% 15,900,000
58.9%
7,105,089
38.4% 10,200,000
47.2% 16,100,000
58.7%
6,948,416
37.9% 10,000,000
46.8% 15,900,000
58.4%
6,878,102
37.1%
9,937,079
46.1% 15,900,000
57.8%
7,062,489
38.8% 10,100,000
47.6% 16,100,000
59.0%
7,208,100
39.1% 10,300,000
47.8% 16,200,000
59.2%
7,240,489
40.0% 10,300,000
48.6% 16,200,000
59.8%
8,090,008
43.1% 11,100,000
51.2% 17,100,000
61.8%
8,044,484
43.7% 11,100,000
51.8% 17,000,000
62.3%
7,987,601
44.3% 11,000,000
52.3% 17,000,000
62.7%
8,040,568
44.1% 11,100,000
52.1% 17,000,000
62.5%
8,274,450
44.4% 11,300,000
52.3% 17,300,000
62.7%
8,496,861
46.3% 11,600,000
54.0% 17,500,000
64.0%
4,711,486
36.1%
6,992,078
45.2%
9,611,898
52.9%
4,922,465
36.6%
7,203,058
45.6%
9,822,878
53.3%
5,399,961
40.9%
7,680,554
49.4% 10,300,000
56.5%
5,928,591
42.0%
8,209,184
50.3% 10,800,000
57.3%
6,092,295
44.3%
8,372,887
52.2% 11,000,000
59.0%
6,095,895
44.4%
8,376,488
52.4% 11,000,000
59.1%
6,241,160
45.6%
8,521,752
53.4% 11,100,000
60.0%
6,225,142
45.7%
8,505,735
53.5% 11,100,000
60.0%
6,290,275
47.3%
8,570,868
54.8% 11,200,000
61.2%
6,905,819
49.8%
9,186,411
57.0% 11,800,000
63.0%
6,962,225
49.1%
9,242,818
56.4% 11,900,000
62.5%
6,812,466
48.5%
9,093,059
55.9% 11,700,000
62.1%

Notes: 1. The number of job offshored is the sum of job offshored across all occupations
within an occupational group.
2. The offshoring percentage is the average offshoring percentage across all
occupations within an occupational group.
3. A cubic offshoring cost function is assumed.

106

Table 3.8: Scenario Comparison among the Five Relatively Offshorable Occupational Groups from NLS Method

Year

Pre2000

1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999

Group 1 (20%)
Offshored
Jobs
%
7,771,465 47.6%
7,085,993 45.8%
6,707,947 46.1%
6,292,123 42.1%
5,989,792 40.0%
5,500,926 37.7%
4,891,834 35.6%
4,902,728 34.8%
4,671,854 34.7%
6,299,921 36.2%
6,233,880 32.9%
6,354,836 35.1%
5,822,961 33.1%
5,651,819 33.3%
5,185,593 32.9%
4,869,771 30.6%
4,454,344 27.5%

Group 2 (20%)
Offshored
Jobs
%
13,300,000
46.8%
13,200,000
46.6%
12,800,000
46.1%
12,500,000
44.8%
12,400,000
45.5%
12,000,000
43.1%
11,700,000
44.0%
11,200,000
40.9%
11,100,000
39.5%
11,000,000
41.0%
10,600,000
39.1%
10,900,000
40.1%
10,600,000
40.1%
10,100,000
38.9%
9,780,010
37.5%
9,640,700
36.4%
8,816,018
34.8%

Group 4 (20%)
Offshored
Jobs
%
4,489,843 35.1%
4,182,837 32.5%
4,178,781 32.4%
4,014,622 30.7%
3,805,164 29.2%
3,846,256 30.6%
3,670,658 30.7%
3,370,897 29.0%
3,407,134 30.8%
3,370,818 27.4%
3,440,809 27.9%
4,456,491 34.7%
4,284,704 32.7%
4,027,732 31.4%
3,933,802 31.0%
3,888,172 30.2%
4,005,604 28.9%

107

Group 5 (40%)
Offshored
Jobs
%
14,600,000
50.2%
14,800,000
51.7%
14,600,000
51.2%
14,500,000
50.1%
14,300,000
48.9%
14,200,000
49.5%
14,300,000
48.2%
13,800,000
45.6%
14,000,000
45.3%
13,900,000
45.2%
14,100,000
44.9%
16,200,000
50.5%
16,200,000
50.7%
16,100,000
49.9%
16,200,000
48.9%
16,200,000
48.2%
16,200,000
48.8%

Group 9 (80%)
Offshored
Jobs
%
15,700,000 57.4%
15,800,000 58.7%
15,900,000 58.4%
15,900,000 58.4%
15,900,000 58.9%
16,100,000 58.7%
15,900,000 58.4%
15,900,000 57.8%
16,100,000 59.0%
16,200,000 59.2%
16,200,000 59.8%
17,100,000 61.8%
17,000,000 62.3%
17,000,000 62.7%
17,000,000 62.5%
17,300,000 62.7%
17,500,000 64.0%

Table 3.8 (cont’d)

Post2000

2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011

7,696,804
7,312,754
7,152,252
7,336,184
7,286,110
7,108,696
6,919,571
6,652,990
6,526,403
6,630,334
6,953,488
6,892,056

37.5%
35.2%
36.1%
37.0%
35.7%
34.0%
34.1%
33.0%
34.0%
32.5%
32.1%
33.6%

12,300,000
11,800,000
11,600,000
11,500,000
11,400,000
11,200,000
11,000,000
10,400,000
10,100,000
9,978,409
10,000,000
9,962,497

39.4%
38.3%
38.5%
37.8%
37.4%
37.2%
36.1%
35.5%
35.0%
34.9%
34.2%
34.2%

5,993,041
5,701,304
5,635,168
5,805,051
5,677,808
5,556,994
5,363,608
5,251,183
5480,845
5,400,839
5,567,462
5,655,523

32.1%
28.8%
28.9%
29.9%
29.7%
30.0%
27.9%
28.0%
29.9%
29.6%
31.5%
32.4%

15,400,000
15,400,000
15,800,000
15,800,000
16,000,000
15,900,000
15,900,000
16,100,000
15,900,000
16,400,000
16,500,000
16,900,000

45.7%
47.4%
48.5%
45.7%
45.2%
44.5%
45.6%
47.6%
47.1%
48.5%
48.0%
48.9%

9,611,898
9,822,878
10,300,000
10,800,000
11,000,000
11,000,000
11,100,000
11,100,000
11,200,000
11,800,000
11,900,000
11,700,000

52.9%
53.3%
56.5%
57.3%
59.0%
59.1%
60.0%
60.0%
61.2%
63.0%
62.5%
62.1%

Notes: 1. The number of job offshored is the sum of job offshored across all occupations within an occupational group.
2. The offshoring percentage is the average offshoring percentage across all occupations within an occupational group.
3. A cubic offshoring cost function is assumed.
4. In 20%, 40% and 80% scenario, offshoring is set not to exceed the 20%, 40% and 80% of the maximum adjusted employment of all
occupations across all years within each occupational group respectively.

108

Table 3.A: Monotonic Cubic Spline Interpolation Method Preliminary Point Estimates of
Parameterized Offshoring Cost Function 𝑡(𝑖)
Value of 𝑡(𝑖) at Control Points 𝑖

Gr
ou
p
1
2
3
4
5
6
7
8
9
10

0
3.02
1.11
8.40
0.04
13.9
8.82
5.37
8.09
8.29
8.25

0.1
8.98
2.65
6.90
7.00
5.53
13.4
9.24
0.12
0.14
10.1

0.2
14.1
8.75
12.8
12.3
11.5
20.2
17.2
15.3
0.94
15.8

0.3
26.2
15.4
18.7
20.6
20.2
32.6
25.2
14.7
7.27
35.6

Pre-2000: 1983-1999
0.4
0.5
0.6
36.3
48.6
61.7
32.5
72.0
121.4
47.3
174.3 262.4
38.8
57.8
87.8
30.6
43.2
59.5
46.2
83.6
119.4
52.4
75.5
92.5
75.6
110.6 131.8
18.9
54.0
104.4
71.7
208.7 284.9

0.7
80.1
138.1
372.7
120.9
79.1
182.9
136.1
172.0
136.7
421.4

0.8
98.83
131.7
298.5
185.8
103.7
272.1
191.1
266.1
151.5
561.2

0.9
126.5
123.4
452.6
167.1
131.8
401.7
307.4
355.9
173.3
798.0

1.0
185.1
109.1
259.2
144.0
146.1
537.2
438.8
552.8
165.1
1011.5

Value of 𝑡(𝑖) at Control Points 𝑖
Post-2000 Period: 2000-2011
1
3.09 6.85 12.6
20.5 30.8
43.05 58.9
78.18
102.2 121.8 186.0
2
8.49 2.09 4.45
7.80 29.7
110.2 157.6
197.4
225.4 173.1 459.4
3
17.4 26.9 55.4
69.5 141.6 232.7 180.7
405.3
533.4 744.8 1076.7
4
0.87 6.92 9.56
19.8 24.4
51.8
59.7
86.4
102.0 129.5 184.9
5
3.27 1.01 0.88
37.2 33.8
48.0
64.3
85.5
101.7 133.3 165.2
6
22.4 23.2 25.3
35.7 53.5
77.0
111.3
165.6
249.8 215.3 259.1
7
2.21 27.9 26.6
40.8 23.2
82.3
116.8
203.8
265.3 369.0 533.2
8
5.69 7.30 2.14
18.2 34.6
41.3
62.5
92.1
134.7 182.3 269.8
9
0.70 0.09 1.32
40.6 19.5
71.3
105.2
95.2
85.91 117.7 124.6
10 16.1 11.2 11.4
53.7 79.8
99.4
132.5
171.3
124.0 195.0 241.0
Notes: 1. No other control variables are included in the model.
2. The upper bound is set that offshoring cannot exceed the 10% of the maximum
employment of all occupations across all years within each group.
3. No observations are dropped.
4. The maximum iterations is500 times.
5. For Group 1, 2, 4, 5, 9the functional form for iteration to start with is
(4𝑥 + 1.5)3 + 𝜀, where 𝜀 is a random shock with normal distribution
𝑁(0, 0.01).
For Group 3, 6, 7, 8 and 10, the functional form for iteration to start with is 10 ∗
𝑒𝑥𝑝(4𝑥) + 𝜀, where 𝜀 is a random shock with normal distribution 𝑁(0, 0.01).

109

Figure 3.2.1: Monotonic Cubic Spline Interpolation Method Offshoring Cost Function by
Occupational Groups in Pre-2000 Period (1983-1999)

110

Figure 3.2.1 (cont’d)

111

Figure 3.2.1 (cont’d)

112

Figure 3.2.1 (cont’d)

113

Figure 3.2.1 (cont’d)

114

Figure 3.2.1 (cont’d)

115

Figure 3.2.1 (cont’d)

116

Figure 3.2.1 (cont’d)

117

Figure 3.2.1 (cont’d)

118

Figure 3.2.1 (cont’d)

119

Figure 3.2.2: Monotonic Cubic Spline Interpolation Method Offshoring Cost Function by
Occupational Groups in Post-2000 Period (2000-2011)

120

Figure 3.2.2 (cont’d)

121

Figure 3.2.2 (cont’d)

122

Figure 3.2.2 (cont’d)

123

Figure 3.2.2 (cont’d)

124

Figure 3.2.2 (cont’d)

125

Figure 3.2.2 (cont’d)

126

Figure 3.2.2 (cont’d)

127

Figure 3.2.2 (cont’d)

128

Figure 3.2.2 (cont’d)

129

Figure 3.3.1: NLS Method Cubic Offshoring Cost Function for G1 (Management, Business and Financial Operations Occupations)

130

Figure 3.3.1 (cont’d)

Notes: 95 percent confidence band is calculated with 50 times bootstrapping.
131

Figure 3.3.2: NLS Method Cubic Offshoring Cost Function for G2 (Professional and Related Occupations)

132

Figure 3.3.2 (cont’d)

Notes: 95 percent confidence band is calculated with 50 times bootstrapping.

133

Figure 3.3.3: NLS Method Cubic Offshoring Cost Function for G4 (Sales and Related Occupations)

134

Figure 3.3.3 (cont’d)

Notes: 95 percent confidence band is calculated with 50 times bootstrapping.
135

Figure 3.3.4: NLS Method Cubic Offshoring Cost Function for G5 (Office and Administrative Support Occupations)

136

Figure 3.3.4 (cont’d)

Notes: 95 percent confidence band is calculated with 50 times bootstrapping.
137

Figure 3.3.5: NLS Method Cubic Offshoring Cost Function for G9 (Production Occupations)

138

Figure 3.3.5 (cont’d)

Notes: 95 percent confidence band is calculated with 50 times bootstrapping.
139

Figure 3.4.1: Change of Offshoring Percentage for G1 (Management, Business and
Financial Operations Occupations)

65.0%

Pre-2000 (G1: Management, Business and
Financial Operations Occupations )

55.0%
45.0%

20% Scenario
40% Scenario

35.0%
25.0%
1980

80% Scenario

1985

1990

1995

2000

Post-2000 (G1: Management, Business and
Financial Operations Occupations )
60.0%
55.0%
50.0%
20% Scenario
45.0%

40% Scenario
80% Scenario

40.0%
35.0%
30.0%
1998

2003

2008

140

2013

Figure 3.4.2: Change of Offshoring Percentage for G2 (Professional and Related
Occupations)

Pre-2000 (G2: Pressional and Related Occupations)
65.0%
60.0%
55.0%

20% Scenario

50.0%

40% Scenario
80% Scenario

45.0%
40.0%
35.0%
30.0%
1980

60.0%

1985

1990

1995

2000

Post-2000 (G2: Professional and Related
Occupations)

55.0%
50.0%
45.0%

20% Scenario

40% Scenario

40.0%

80% Scenario
35.0%
30.0%
1998

2000

2002

2004

2006

141

2008

2010

2012

Figure 3.4.3: Change of Offshoring Percentage for G4 (Sales and Related Occupations)

60.0%

Pre-2000 (G4: Sales and Related
Occupations)

55.0%
50.0%

45.0%
40.0%

20% Scenario

35.0%

40% Scenario

30.0%

80% Scenario

25.0%
20.0%
1980

1985

1990

1995

2000

Post-2000 (G4: Sales and Related Occupations)
55.0%
50.0%

20% Scenario

45.0%

40% Scenario

40.0%

80% Scenario

35.0%
30.0%
25.0%
1998

2003

2008

142

2013

Figure 3.4.4: Change of Offshoring Percentage for G5 (Office and Administrative Support
Occupations)

Pre-2000 (G5: Office and Administrative
Suppot Occupations)
65.0%
60.0%
55.0%
50.0%

20% Scenario

45.0%

40% Scenario

40.0%

80% Scenario

35.0%
30.0%
1980

60.0%

1985

1990

1995

2000

Post-2000 (G5: Office and Administrative
Occupations)

55.0%
50.0%
20% Scenario

45.0%

40% Scenario
80% Scenario

40.0%
35.0%
1998

2000

2002

2004

2006

143

2008

2010

2012

Figure 3.4.5: Change of Offshoring Percentage for G9 (Production Occupations)

Pre-2000 (G9: Production Occupations)
70.0%
65.0%
60.0%
55.0%
20% Scenario
50.0%

40% Scenario

45.0%

80% Scenario

40.0%
35.0%
1980

1985

1990

1995

2000

Post-2000 (G9: Production Occupations)
65.0%
60.0%
55.0%
20% Scenario

50.0%

40% Scenario

45.0%

80% Scenario

40.0%
35.0%
1998

2003

2008

144

2013

Figure 3.5: Changes of Offshoring Percentage for the Five Relatively Offshorable
Occupational Groups

Pre-2000 Changes of Offshoring Percentage
70.0%
65.0%
60.0%
55.0%
50.0%
45.0%
40.0%
35.0%
30.0%
25.0%
1980

Management, Business
and Finance (20%)
Professional (20%)
Sales (20%)
Office and
Administrative (40%)
Production (80%)
1985

1990

1995

2000

Post-2000 Changes of Offshoring Percentage
65.0%
60.0%
55.0%
50.0%
45.0%
40.0%
35.0%
30.0%
25.0%
1995

Management, Business
and Finance (20%)
Professional (20%)
Sales (20%)
Office and
Administrative (40%)
Production (80%)
2000

2005

2010

145

2015

REFERENCES

146

REFERENCES

Autor, D. H., F. Levy, and R. J. Murnane (2003). The skill content of recent technological
change: An empirical exploration. Quarterly Journal of Economics 118 (4), 1279–1333.
Bardhan, A. D. and C. Kroll (2003). The new wave of outsourcing. Technical report, UC
Berkeley: Fisher Center for Real Estate and Urban Economics.
Blinder, A. S. (2006). Offshoring: The next industrial revolution. Foreign Affairs 85, 1113298.
Blinder, A. S. (2009). How many U.S. jobs might be offshorable. World Economics 10 (2),
41–78.
Blinder, A. S. and A. B. Krueger (2013). Alternative measures of offshorability: A survey
approach. Journal of Labor Economics 31 (2), S97–S128.
Crinó, R. (2010). Employment effects of service offshoring: Evidence from matched firms.
Economics Letters 107 (2), 253–256.
Criscuolo, C. and L. Garicano (2010). Offshoring and wage inequality: Using occupational
licensing as a shifter of offshoring costs. American Economic Review 100 (2), 439–43.
Ebenstein, A., A. Harrison, M. McMillan, and S. Phillips (2009). Why are American
workers getting poorer? estimating the impact of trade and offshoring using the CPS.
Working Paper 15107, National Bureau of Economic Research.
Friedman, T. L. (2007). The World Is Flat: A Brief History of the Twenty-First Century.
D&M Publishers Incorporated.
Goos, M., A. Manning, and A. Salomons (2008). Recent changes in the European
employment structure: the roles of technology, globalization and institutions. Technical
report.
Grossman, G. M. and E. Rossi-Hansberg (2008). Trading tasks: A simple theory of
offshoring. American Economic Review 98 (5), 1978–97.
Harrison, A. and M. McMillan (2011). Offshoring jobs? multinationals and U.S.
manufacturing employment. Review of Economics and Statistics 93 (3), 857–875.
Wolberg, G. and I. Alfy (1999). Monotonic cubic spline interpolation. Proceedings of
Computer Graphics International, 188–195.
Wolberg, G. and I. Alfy (2002). An energy-minimization framework for monotonic cubic spline
interpolation. Journal of Computational and Applied Mathematics 143 (2), 145 – 188.

147