ESSAYS ON TRADE AND TRANSPORTATION
by
FELIX L. FRIEDT
A DISSERTATION
Presented to the Department of Economics
and the Graduate School of the University of Oregon
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
June 2017
DISSERTATION APPROVAL PAGE
Student: Felix L. Friedt
Title: Essays on Trade and Transportation
This dissertation has been accepted and approved in partial fulfillment of the requirements
for the Doctor of Philosophy degree in the Department of Economics by:
Bruce Blonigen Co-Chair
Wesley Wilson Co-Chair
Anca Cristea Core Member
Jeremy Piger Core Member
Nagesh Murthy Institutional Representative
and
Scott L. Pratt Dean of the Graduate School
Original approval signatures are on file with the University of Oregon Graduate School.
Degree awarded June 2017
ii
c© 2017 Felix L. Friedt
iii
DISSERTATION ABSTRACT
Felix L. Friedt
Doctor of Philosophy
Department of Economics
June 2017
Title: Essays on Trade and Transportation
This dissertation considers the interconnections between trade and transportation.
Through various theoretical and empirical analyses, I provide novel evidence of the
simultaneity of trade and transportation, of spillover effects across integrated transport
markets, and of the influence of the international transport sector on trade policy
effectiveness and natural disaster induced trade disruptions.
In the first substantive chapter, I develop a model of international trade and
transportation. Accounting for the joint-production present in the international container
shipping industry, I illustrate that freight rates adjust to differences in the international
demands for transport and can result in balanced or imbalanced equilibrium trade in the
presence of asymmetric freight rates. The empirical results exhibit the simultaneity of
international trade and transportation costs and show that the dependence of transport
costs on the trade imbalance can lead to spillover effects across bilateral export and import
markets.
In the second substantive chapter, I investigate the effects of maritime trade
policy on bilateral trade in the presence of trade imbalances. Using the previously
developed model, I show that the trade elasticities with respect to carrier costs vary
systematically across transport markets, bilateral trade imbalances and differentiated
iv
products. Empirically, I estimate the varying effects of an EU environmental policy on
U.S.-EU trade and provide strong evidence in support of the theoretical results.
In the third substantive chapter, I analyze the dynamics and spatial distribution
of the trade effects induced by natural disasters. I develop a spatial gravity model
of international trade and apply the model to monthly US port level trade data.
Empirically, I estimate the dynamic evolution of trade effects caused by Hurricane Katrina
differentiating trade disruptions at the local port level. The estimates point to the static
and dynamic resilience of international trade. While ports closest to Katrina’s epicenter
experience significant short-run reductions that can be of permanent nature, international
trade handled by nearby ports rises in response to this disaster, both in the short- and
in the long-run. Overall, the analysis underlines the significance of local infrastructure
networks to reduce the devastation inflicted by natural disasters.
This dissertation includes previously unpublished co-authored material.
v
CURRICULUM VITAE
NAME OF AUTHOR: Felix L. Friedt
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR
Whitworth University, Spokane, WA
DEGREES AWARDED:
Doctor of Philosophy, Economics, 2017, University of Oregon
Master of Science, Economics, 2013, University of Oregon
Bachelor of Arts, Economics & International Business, 2012, Whitworth University
AREAS OF SPECIAL INTEREST:
International Trade
Industrial Organization
Econometrics
Transportation Economics
GRANTS, AWARDS AND HONORS:
Invited Participant, Western Economic Associate International Graduate Student
Workshop, 2016
Gerlof Homan Graduate Scholarship in International Economics, University of
Oregon, 2015
Charles A. Reed Graduate Fellowship, University of Oregon, 2014
Dale Underwood Award, University of Oregon, 2013
Best First-Year Econometrics Performance Award, 2013
Graduate Teaching Fellowship, University of Oregon, 2012-2017
vi
ACKNOWLEDGEMENTS
I am very thankful for the support, guidance and advice I have received from Bruce
Blonigen, Wesley Wilson, Anca Cristea and Jeremy Piger. I am grateful for many fruitful
discussions with Taylor McKenzie, Michael Thacker and Joseph Wyer. Furthermore, I
thank the participants at the University of Oregon IO and Micro Groups for many helpful
comments and Dena Marshall for numerous edits of previous drafts of Chapters II and III.
Earlier versions of Chapters II and III were presented in the TPUG sessions at the 2015
ASSA, 2015 WEAI and 2016 WEAI meetings. I thank the audience and, in particular,
David Austin, Sharat Ganapati, William Huneke, Dennis Novy and Georg Schaur for their
insightful comments. Any remaining errors are my own.
The financial support received through the Gerlof Homan Graduate Scholarship in
International Economics is also gratefully acknowledged.
vii
To my wife, Cassie Rene´e, and my parents, Maria-Anna and Gerhard Ludwig, for their
unwavering support.
viii
TABLE OF CONTENTS
Chapter Page
I. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
II. TRADE, TRANSPORTATION AND TRADE IMBALANCES: AN
EMPIRICAL EXAMINATION OF INTERNATIONAL MARKETS AND
BACKHAULS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Institutional Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Theoretical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Empirical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
III. TRADE POLICY OUTCOMES IN THE PRESENCE OF ASYMMETRIC
TRANSPORT COSTS:THEORY AND EVIDENCE . . . . . . . . . . . . . . . . 67
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Institutional Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Empirical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
ix
Chapter Page
IV. THE RESILIENCE OF INTERNATIONAL TRADE: AN EMPIRICAL
EXAMINATION OF THE DYNAMIC SPATIAL TRADE EFFECTS OF
NATURAL DISASTERS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Institutional Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Theoretical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Empirical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
V. CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
APPENDICES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
A. PROOFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
B. DERIVATION OF TRADE ELASTICITIES IN THE VALUE CASE . . . . . 188
C. ADDITIONAL TABLES AND FIGURES . . . . . . . . . . . . . . . . . . . . . 190
REFERENCES CITED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
x
LIST OF FIGURES
Figure Page
1. Linkages between Trade Balance and Transportation Cost Differentials . . . . . 33
2. Trade Imbalances and Freight Rate Differentials . . . . . . . . . . . . . . . . . . . 49
3. Average End-of-the-Year Monthly Growth Rates of U.S.-EU Trade . . . . . . . . 98
4. U.S. Trade Flows across Treatment & Control Groups by Transport Market . . . 108
5. Marginal Treatment Effects by Transport Market and Trade Imbalance . . . . . . 116
6. Aggregate Trade Imbalance by Income Level and Geographic Region . . . . . . . 124
7. Export and Import Share of Ore and Metal by Geographic Region . . . . . . . . 125
8. Global Disaster Trends (1960-2015) . . . . . . . . . . . . . . . . . . . . . . . . . . 136
9. Geographical Movement and Strength of Hurricane Katrina . . . . . . . . . . . . 137
10. Geo-referenced Port Exports Pre & Post Hurricane Katrina . . . . . . . . . . . . 153
11. Geo-referenced Port Imports Pre & Post Hurricane Katrina . . . . . . . . . . . . 154
12. Trends in Aggregate, Regional and Local U.S. Exports and Imports . . . . . . . . 155
13. Trends in the Number of Traded Products . . . . . . . . . . . . . . . . . . . . . . 157
14. Treatment Effects over Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
15. Dynamic Variation in Treatment Effects - New Orleans, LA & Gulfport, MS . . . 171
16. Dynamic Variation in Treatment Effects - Mobile, AL . . . . . . . . . . . . . . . 173
17. Dynamic Variation in Treatment Effects - Panama City, FL . . . . . . . . . . . . 173
18. Dynamic Treatment Effects - Port Arthur, TX . . . . . . . . . . . . . . . . . . . 204
19. Dynamic Treatment Effects - Houston, TX . . . . . . . . . . . . . . . . . . . . . 204
20. Dynamic Treatment Effects - Galveston, TX . . . . . . . . . . . . . . . . . . . . . 204
21. Dynamic Treatment Effects - Tampa, FL . . . . . . . . . . . . . . . . . . . . . . 205
22. Dynamic Treatment Effects - Freeport, TX . . . . . . . . . . . . . . . . . . . . . 205
23. Dynamic Treatment Effects - Miami, FL . . . . . . . . . . . . . . . . . . . . . . . 205
xi
Figure Page
24. Dynamic Treatment Effects - Corpus Christi, TX . . . . . . . . . . . . . . . . . . 206
25. Dynamic Treatment Effects - Port Everglades, FL . . . . . . . . . . . . . . . . . 206
xii
LIST OF TABLES
Table Page
1. Summary Statistics - Trade and Unit-specific Trade Costs . . . . . . . . . . . . . 46
2. Summary Statistics - Trade and Freight Rate Determinants . . . . . . . . . . . . 47
3. Panel Unit Root Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4. Panel Cointegration Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5. Cointegration Relations - Group-mean Panel Dynamic OLS Estimator . . . . . . 58
6. Cointegration Relations - Group-mean Panel Fully Modified OLS Estimator . . . 60
7. Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8. Avg. Containerized Trade and Freight Rate Imbalances . . . . . . . . . . . . . . 75
9. U.S. Containerized Trade by Foreign Country . . . . . . . . . . . . . . . . . . . . 104
10. ATE - PPML . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
11. ATE - Varying Trade imbalance . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
12. ATE - Disaggregated Product Group Level . . . . . . . . . . . . . . . . . . . . . 118
13. Average Continental Disaster Impact . . . . . . . . . . . . . . . . . . . . . . . . . 134
14. U.S. Port of Entry - Export Summary Pre & Post Hurricane Katrina . . . . . . . 149
15. U.S. Port of Entry - Import Summary Pre & Post Hurricane Katrina . . . . . . . 150
16. Aggregate Trade Disruptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
17. Regional Trade Disruptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
18. Port-Specific Trade Disruptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
19. Influence of Port Characteristics on Treatment Effects . . . . . . . . . . . . . . . 166
20. Port-specific Effects on the Number of Exported Products . . . . . . . . . . . . . 177
21. Port-specific Effects on the Number of Imported Products . . . . . . . . . . . . . 180
22. Product Category Legend Key . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
23. Robustness Analysis - Time Trend Inclusive Specification . . . . . . . . . . . . . 190
xiii
Table Page
24. Robustness Analysis - Varying Levels of Clustered Standard Errors . . . . . . . . 191
25. Robustness Analysis - Varying Fixed Effects Specifications . . . . . . . . . . . . . 191
26. Robustness Analysis - Various Sample Restrictions . . . . . . . . . . . . . . . . . 192
27. Robustness Analysis - Alternative Empirical Model Specifications . . . . . . . . . 193
28. Robustness Analysis - Varying Backhaul Identifications . . . . . . . . . . . . . . . 193
29. Robustness Analysis - BH Identification at Varying Levels of Aggregation . . . . 194
30. Distances between U.S. Ports of Entry . . . . . . . . . . . . . . . . . . . . . . . . 195
31. Aggregate Trade Disruptions - Various Spatial Weights . . . . . . . . . . . . . . . 198
32. Regional Trade Disruptions - Various Spatial Weights . . . . . . . . . . . . . . . 199
33. Port-Specific Trade Disruptions - SAC . . . . . . . . . . . . . . . . . . . . . . . . 199
34. Port Characteristics Legend Key . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
35. Influence of Port Characteristics on Treatment Effects - SAC . . . . . . . . . . . 202
36. Treatment Effect over Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
xiv
CHAPTER I
INTRODUCTION
The primary focus of my research considers how transportation industries and their
unique characteristics influence the patterns and composition of international trade, the
effectiveness of trade policy and the resilience of international trade. Engaging in this
research, I draw from the relevant elements of the extensive literatures on international
economics and industrial organization. In the field of international economics, much of the
literature is dedicated to the study of the determinants of international trade.1 Barriers of
trade have long been recognized as one of these major determinants and have received
much academic and policy related attention (e.g. Hummels, 2001; Anderson and van
Wincoop, 2004). Although transportation industries play an integral role in the facilitation
of international transactions and the related costs exemplify a natural barrier to trade,
most of the related literature focuses on artificial trade barriers, such as tariffs.
Recently, however, there has been a growing interest in transportation as a
determinant of trade (e.g. Hummels and Skiba, 2004; Behrens et al., 2006; Hummels, 2007;
Behrens and Picard, 2011; Deardorff, 2014). These studies demonstrate, both theoretically
and empirically, that transportation costs are a central factor in the realm of trade costs
and can influence the patterns of international trade. While distance has been used as the
traditional proxy for transport costs, various studies, including for example Limao and
Venables (2001) and Combes and Lafourcade (2005), have shown that it fails to capture
significant variation concerning these costs. Surprisingly few studies have responded to
these findings and provided a more careful treatment of the transport sector within a
1See, for example, studies by Deardorff (1998), Head et al. (2010) or Head and Mayer (2013).
1
model of international trade (e.g. Behrens and Picard, 2011; Takahashi, 2011; Ishikawa
and Tarui, 2015).
In the first essay of my dissertation, co-authored with Wesley Wilson and entitled
”Trade, Transportation and Trade Imbalances: An Empirical Examination of International
Markets and Backhauls”, we derive and estimate a model of international trade and
transportation. Specifically, we develop a framework of international trade following
Hummels et al. (2009) and incorporate a model of the international transport sector
that accounts for the potential backhaul problem facing carriers facilitating international
bilateral trade.2 The theoretical findings show the simultaneity of trade and transport
costs and co-integration of fronthaul and backhaul transport markets. Empirically, we
use data from three different sets of markets that comprise the bulk of world trade and
show that the simultaneity of trade and transport costs manifests itself in the existence
co-integrating relations that govern the long-run equilibrium in the international container
shipping industries. Simulations based on the estimated structural pricing relations and
demand equations illustrate the potential spillover effects of international trade policy and
demonstrate the significance of fronthaul and backhaul market co-integration.
The simultaneity of trade and transport costs in the presence of the backhaul
problem highlights the transport sector as a potential policy avenue to stimulate
international trade. In my second dissertation essay, entitled ”Trade Policy Outcomes in
the Presence of Asymmetric Transport Costs: Theory and Evidence”, I build on these
findings. In particular, I theoretically evaluate the potential impact of international
carrier-cost-reducing policy on trade and test whether the presence of the backhaul
2The backhaul problem arises when carriers offer a transport service facilitating trade from location A
to location B and, thereby, inadvertently create a transportation service available to facilitate trade from
location B to location A, known as the backhaul. It becomes a problem for carriers when the demands for
transport in markets AB and BA are imbalanced, such that transport capacity is not be fully utilized on
the backhaul.
2
problem in the international transport sector influences the effectiveness of this trade
policy. Theoretical derivations suggest that carrier-cost-reducing policies have larger effects
on trade facilitated in backhaul rather than fronthaul transport markets and that the
difference in these effects varies with the trade imbalance and across product groups.
Empirically, I test these theoretical predictions by estimating the effects of an
exogenous shock to transport costs given by an EU environmental policy on U.S.
containerized trade. I employ a large dataset obtained from the U.S. Census Bureau
that contains trade flows at the U.S.-port-foreign-country level and augment it with
employment statistics from various sources. Overall, the analysis produces novel
theoretical and empirical findings that provide strong evidence that the presence of the
backhaul problem causes systematic variation in policy effectiveness. The significance
of these results follows from the sizable current U.S. trade deficit which prescribes that
U.S. exports(imports) are primarily facilitated in backhaul(fronthaul) transport markets
in recent decades. Given this imbalanced trade pattern, the evidenced variation in
policy effectiveness has strong implications for policies that apply to the U.S. transport
sector, such as the StrongPorts initiative by the Maritime Administration or the Trade
Facilitation Agreement by the World Trade Organization (WTO).
In the final essay of my dissertation, entitled ”The Resilience of International Trade:
An Empirical Examination of the Dynamic Spatial Trade Effects of Natural Disasters”,
I investigate how the transportation network of U.S. seaports influences the response
of U.S. trade to natural disasters. The central hypotheses of interest are whether or not
international trade exhibits a static and/or dynamic resilience to natural disaster induced
trade disruptions and what role infrastructure and transportation networks play in the
determination of these resiliences. An extensive literature review highlights the gap in
the current state of the literature and reveals that these important questions have not
3
yet been addressed. Theoretically, I build on the traditional general equilibrium gravity
equation and extend it to allow for the spatial and dynamic responses of disaster induced
trade effects on international trade. Empirically, I test the resilience of trade estimating
the spatial distribution and the dynamic response of U.S. containerized trade to Hurricane
Katrina controlling for spatial correlations in the data. Various analyses of the U.S. port-
level trade data point to very localized offsetting short-run as well as long-run disaster
induced trade disruptions that underlie the aggregate resilience of international trade and
evidence the importance of local transportation and infrastructure networks in shaping
these responses.
My research contributes to both the international economics and industrial
organization’s literatures, and should be of considerable interest to researchers and
policy-makers within these areas. While world trade has been growing at a rate faster
than world GDP over the past decades (Blonigen and Wilson, 2013), more recent data
reveal a relatively uncharacteristic downturn. In light of this recent development and
the current political uncertainty concerning traditional commercial policies, my work
points to transportation industries and infrastructure as an alternative policy instrument
to stimulate trade. Moreover, my theoretical and empirical findings provide supporting
evidence that the international transportation industries, in fact, influence trade policy
effectiveness and are a cornerstone of the resilience of international trade.
4
CHAPTER II
TRADE, TRANSPORTATION AND TRADE IMBALANCES: AN EMPIRICAL
EXAMINATION OF INTERNATIONAL MARKETS AND BACKHAULS
The present essay emanated from a term paper project I completed during Professor
Wesley Wilson’s Industrial Organization seminar. While Professor Wilson suggested the
topic of the paper and provided initial points of reference in the literature, I compiled an
extensive literature review on the issue of trade and transportation, refined the specific
research questions addressed in this essay and developed the theoretical model underlying
our analysis. Under my advisor’s guidance, I refined the theory and significantly
augmented the initially provided data. Having compiled the dataset, I derived the
empirical model. Based on this model, I then conducted the empirical analysis researching
the appropriate estimation techniques and finally implementing the advanced time-
series econometrics discussed in the paper. While Professor Wesley Wilson contributed
to various parts of this project, I believe it is fair to say that I took the lead on the
development of this research and each of its aspects. From the initial analysis to the final
write-up, I worked through each of the intricate details of this study.
Introduction
International trade has been growing for decades and has been rising faster
than world gross domestic product (Blonigen and Wilson, 2013). This growth has put
tremendous pressure on international transport markets which have responded with
considerable innovations; most notably the introduction of containers in the late 1950’s.
The resulting reduction of the costs of transportation between countries has, along with
5
rising incomes, fueled the growth of international trade and the interest in developing
models that link trade and transportation together.
In the present study, I derive a model of transportation demand based on trade
determinants and incorporate the supply-side, similar to Behrens and Picard (2011),
with direct attention to the fact that most supply is provided between pairs of countries
under conditions of joint production. I apply the model to three different sets of markets
encompassing trade by container between Asia and the U.S., Asia and Europe, and
the U.S. and Europe. The empirical tests employed in this study exhibit strong long-
run relationships between trade and transportation that reflect the static equilibrium
relations derived from the theory. The results point to the importance of freight rates in
the determination of international trade and their respective dependence on the trade
imbalance.
I estimate that in the long-run a 1% permanent increase in freight rates leads to a
0.06% permanent decline of trade, while freight rates show varying long-run responses
to changes in the trade imbalance between two trading countries. The specific point
estimates accentuate the importance of transportation in the determination of trade and
highlight port and maritime transit policy measures as a key instrument to stimulate
the growth of international trade. The primary contributions of this study are threefold.
First, the presentation of a model that integrates trade and transportation with an
explicit representation of the joint production present in the liner shipping industry and
the ability to explain the large freight rate differentials observed between the U.S. and
China, for example. Second, the estimation of the structural relationships that capture
the long-run equilibrium conditions suggested by the theory. Third, simulations based on
these estimates that highlight that the dependence of transportation costs on the trade
6
imbalance can lead to spillover effects from policy across bilateral export and import
markets.
Despite its importance to the determination of trade, historically transportation
has played only a secondary role in the trade literature. In recognition of this gap in
the literature, Behrens et al. (2006), Luo (2011) and Kleinert and Spies (2011) have
integrated a transport sector into various models of trade. I extend this line of research
by developing a model that combines the literatures on both trade and transportation.
The trade framework follows Hummels et al. (2009), while equilibrium transportation costs
are derived from the integration of a transport sector that follows the literature on the
backhaul problem1 by Wilson (1987), Wilson (1994) and Wilson and Beilock (1994). More
specifically, I expand the model discussed by Hummels et al. (2009) by incorporating an
international transport sector that accounts for the possibility of the backhaul problem
faced by international carriers operating on bidirectional markets. I use this structural
model to gain insights into the simultaneity of trade and transport costs as well as the
effects of backhaul problem and show that there are potential spillover effects deriving
from the cointegration of bilateral transport costs. This result compliments the findings
by Deardorff (2014) illustrating that not only the presence of unit-specific trade costs, but
also their cointegration can distort traditional trade theory results.
Empirically, I show that there are cointegration relations that establish the
simultaneity between trade and transport costs suggested by the theory and govern the
long-run equilibrium supply and demand conditions of the international transport markets
facilitating international trade. The estimated cointegration equations provide long-run
1This ’problem’ is an artifact of the market structure that freight carriers face. Allocating capacity
and serving the transport market facilitating goods from country i to country j on a fixed round trip,
inadvertently creates capacity to serve the transportation market facilitating trade from country j to
country i. If this capacity available in the secondary backhaul transport market is not met with the
demand for transport by shippers, it creates the, so called, backhaul problem for carriers that incur the
joint cost of allocating capacity and serving the market pair ij.
7
structural relationships that support the general findings of static cross-sectional analyses
in the trade literature. That is, trade between two trading regions is driven by aggregate
income measures and reduced by trade costs. In particular, I illustrate that the estimated
long-run equilibrium trade relation supports the general finding of unit elasticity of trade
with respect to exporter and importer aggregate income, with point estimates ranging
from 0.927 to 1.189 which are shown not to be statistically different from one at any
conventional significance level.
In addition to that, the estimates demonstrate that, in the long-run, a 1% permanent
increase in freight rates leads to, on average, a 0.058% permanent decline in trade.
This relatively inelastic long-run response of the volume of containerized trade to an
increase in unit-specific trade costs appears reasonable given the fact that container
freight rates represent only a small fraction of the value of the goods shipped2 but also
masks the considerable changes in trade in absolute terms. That is, the results suggest
that a $12.5 permanent increase in freight rates reduces the long-run volume of trade by
670,000 containers and the value of trade by $13 billion to $2 trillion depending on the
containerized cargo.
Furthermore, the empirical results lend themselves for the evaluation of potential
transport cost reducing policy measures and their long-run impact on trade. Simulations
based on the coefficient estimates suggest, for example, that a persistent 10% increase in
market shipping capacity is anticipated to permanently lower unit-specific trade costs in
the form of container freight rates by 7.284% in fronthaul transport markets and 4.290%
in backhaul transport markets. Based on the simulations, these long-run equilibrium
2According to the Rodrigue et al. (2013) the average value of a forty foot container ranges between
$20,000 and $3,600,000 depending on the type of cargo. In contrast, the data show that average freight
rates for a twenty foot container range from $835 to $1717 depending on the market.
8
reductions of freight rates coincide with a 0.321% permanent increase in fronthaul trade
and a 0.189% permanent increase in backhaul trade.
The remainder of this paper is organized as follows: In section 2, I describe the
institutional background of the container shipping industry. This background section is
complemented by the literature review concerning the areas of trade, transportation and
the integration of the two subjects and is presented in section 3. In section 4, I present a
theoretical model of transportation markets. The empirical model is developed in section
5. Section 6 summarizes the data employed. These data exhibit dramatic differences
across markets in terms of the level of trade imbalances and and freight rate differentials.
Section 7 presents the empirical results which are generated from panel time series
techniques. Indeed, the examination points to most of the variables having unit roots and
the existence of long-run cointegration relations between trade and transportation. Section
8 provides a summary as well as conclusion and points to areas of further inquiry.
Institutional Background
Prior to modeling the international transportation markets of containerized
shipments, a thorough analysis of the industry and its unique characteristics is imperative.
In this section, I provide the information necessary to gain a basic understanding of the
container shipping industry, its historical development and changing regulation.
In 2006, container shipping had its 50 year anniversary. During the previous 50
years, the industry that was sparked by Malcolm McLean’s historic innovation of a vessel
carrying 58 trailer-truck bodies from Newark, New Jersey to Houston, Texas, has rapidly
grown and revolutionized international trade. As the demand for international transport of
goods has risen dramatically, world shipping capacity3 as well as the capacity of individual
3By 2010, the global container shipping fleet was comprised of 4,677 vessels with a maximum capacity
of 12.8 million twenty-foot equivalent units (TEUs) (UNCTAD, Secretariat, 2014)
9
vessels has grown to match this demand. In 1956, 58 trailer-truck bodies were transported
on board of the Ideal X. By 2006, it was not uncommon for vessels to carry over 9,000
twenty-foot equivalent units (TEUs) (Transportation Research Board, 2006). More
recently, however, Port Finance International (2013) reports, that this capacity has been
more than doubled by the largest container vessel, the triple E-class, operated by Maersk,
with a capacity of 18,270 TEUs per ship.
In light of such rapid and recent developments, the question of how a seemingly
inconspicuous innovation like a metal container has been able to revolutionize
international trade becomes interesting and intriguing. The answer to this question lies
within the costs associated with international trade. Internationally traded goods must be
transported between the two trading regions. In many cases this implies that the goods
have to be shipped long distances between the U.S., Europe, and Asia. The freight rates
and related charges associated with these shipments are an integral part of international
trade costs and have been strongly influenced by the introduction and adoption of the
container.4
Due to the importance of this industry, in 1961, the Federal Maritime Commission
(FMC) was charged with the regulatory oversight of the U.S. liner shipping industry.
The goal in establishing the FMC has been to ensure that no unfair business practices by
the liner groups or foreign governments harm American consumers of imported goods or
American exporters. With the beginning of containerized shipments in 1956, the FMC was
instated in the wake of this revolutionizing innovation and had to reevaluate the existing
regulation set forth in the Shipping Act of 1916. This reevaluation led to the Shipping Act
of 1984. The dilemma the FMC had to address was the disconnect between promoting
4Jacks et al. (2008), for example, find that trade costs play a varying role in the determination of trade
growth over the past two centuries. In particular, authors find that the reduction in trade costs post world
war II has contributed about 33% of total trade growth.
10
competitive freight rates, while allowing conferences to set market rates high enough to
ensure sufficient supply of carrier capacity to support growing U.S. trade flows.
The Shipping Act of 1984 limited the power of ocean common carrier conferences
over each member. Conferences could no longer prohibit its carrier members from
privately negotiating a service contract with a shipper. This policy reform was completed
in 1998 with the passing of the Ocean Shipping Reform Act (OSRA). Although the limited
antitrust immunity of carrier conferences remained, the industry was further deregulated
under OSRA. Conference members were not only allowed to privately negotiate contracts
determining the pricing terms, but OSRA also prohibited the conferences from requiring
disclosure of such contracts.5
This deregulation has led to the restructuring of the global liner6 shipping industry.7
Historically, container freight rates on various trade routes have been set by the associated
conferences of carriers serving the respective trade route. But in the aftermath of OSRA,
these conferences have lost some of their control over the separate markets of this industry.
In 1999, for example, the Asia North America Eastbound Rate Agreement (ANERA)
and the Japan U.S. Eastbound Freight Conference (JUEFC), which previously controlled
some of the freight rates charged on Trans-Pacific trade routes, were suspended. Further
deregulation of the liner shipping industry has not only been initiated by the FMC, but
has been a global issue, instead. In 2006, the EU repealed the block exemption from EU
5”(...) With respect to agreements, OSRA has maintained antitrust immunity for concerted carrier
actions, but has limited the permissible activities to which such immunity attaches. Agreements no
longer may limit or prohibit service contracting by their members. Moreover, agreements are precluded
from requiring members to disclose their service contract negotiations or the details of any contracts into
which they have entered. An agreement may publish general guidelines applicable to members’ individual
contracting practices, but these guidelines must be voluntary and non-enforceable by the agreement and
filed confidentially with the Commission. (...)” Federal Maritime Commission et al. (2000, p.3)
6According to the World Shipping Council liner shipping encompasses all modes of high capacity
transport services. Moreover, liner vessels are primarily containerships. Thus, the deregulation of liner
shipping directly applies to the containerized freight studied in this paper.
7Fusillo (2013) finds that post-OSRA market shares have been much less stable.
11
competition law that it had provided for the liner shipping conferences in EU trades.
This new law became effective in 2008 and led to the termination of the Transatlantic
Conference Agreement (TACA).
Today’s container shipping industry is comprised of many different firms
incorporated all over the globe, holding capacity shares ranging from less than 0.02% to
14.7%8 of the global capacity measured in TEUs. After the demise of the P3 alliance9,
there are currently nineteen major competitors each holding more than 1% of global
capacity shares. Most notable is the largest competitor in this industry, Maersk, holding
14.7%. Combined, these nineteen firms account for over 82% of the global container
shipping capacity, while the remaining market share is scattered across many smaller
carriers. Considering these market shares raises the question of how the Shipping Act of
1984 and OSRA have shaped the market structure in this industry and how much market
power is currently exercised by container carriers? Furthermore, the question of how the
current market structure influences the dynamics that determine international container
freight rates becomes of central interest.
The interplay between trade and transportation that gives rise to the demand and
supply of the international transport sector and determines the asymmetric unit-specific
trade costs given by container freight rates, is the central issue analyzed throughout the
remainder of this study. The following section gives an overview of the existing literature
on international trade and its associated trade costs as well as the existing literature
analyzing transportation markets. Following these separate literature reviews, I present
a summary of the existing literature that develops and tests the theoretical and empirical
8Source: Alphaliner - TOP 100 Operated fleets as per April 11th, 2016
9The P3 alliance was a planned cooperation between Maersk, Mediterranean Shipping Co., and CMA
CGM, the three largest competitors world wide, on the major east-west trade routes. However, after a
year of negotiations, China denied regulatory approval of this alliance and thereby, led to its demise.
12
interconnections between international trade and transportation markets with a particular
focus on the liner shipping industry.
Literature Review
In order to accurately model the international transport sector and integrate
it into a model of international trade, a clear understanding of the theoretical and
empirical developments in the fields of international trade as well as transportation
economics is necessary. This literature review aims to provide such an understanding, first
highlighting the major developments in international trade and subsequently focusing on
the advancements in the field of transportation economics. For the research purposes of
this study, the overlap between the two fields of international trade and transportation
economics is of special interest. To this end, the third section of this literature review is
dedicated towards a concise summary of the historical and recent progress in combining
the two areas of research. In conclusion of this section, I highlight the remaining gaps in
this strand of the literature and describe how this study contributes towards the closure of
these gaps.
International Trade
There is a plethora of different models of international trade. Some of the underlying
foundations of these models include absolute cost advantages (Smith, 1776), Ricardian
comparative cost advantages (Dornbusch et al., 1977; Eaton and Kortum, 2002; Bernard
et al., 2003), varying factor endowments (Heckscher and Ohlin, 1991), differences in
incomes and trade costs (Samuelson, 1952), economies of scale and a taste for variety
(Krugman, 1979), and varying productive efficiencies (Melitz, 2003), among others.
Empirically, the workhorse for modeling trade has been the gravity model introduced
13
by Tinbergen (1962), which takes trade between countries as a function of GDP and
trade cost measures, such as distance between the two countries. The resulting gravity
equation, which can be derived from a multitude of trade models, has been one of the key
instruments used to empirically analyze the determinants of trade flows. There are many
applications for which the gravity equation has been utilized.10 One of the key results
highlighted by both the theoretical and empirical studies is the dependence of trade on
international trade costs.
Regardless of the differences between Krugman’s New Trade Theory11, Anderson
and van Wincoop’s gravity equation12, or the New New Trade Theory based on the
Melitz model of trade13 (Melitz, 2003), all of these models incorporate symmetric trade
costs. Through the theoretical derivation of each of these models, trade costs have been
established to be a major determinant of international trade. Obstfeld and Rogoff (2001)
go as far as claiming that all remaining macroeconomic puzzles identified in their research
critically hinge on trade costs.
Early empirical studies simply controlled for trade costs via distance between two
trading regions. However, as the trade literature has advanced, focus has shifted towards
a more careful integration of trade costs and its individual components. Anderson and
van Wincoop (2004) survey the literature and establish that trade costs are equivalent
10See, for example, surveys by Bergstrand and Egger (2011) or Anderson (2011) and applications by
Bergstrand (1985), Thursby and Thursby (1987), McCallum (1995), Anderson and van Wincoop (2003)
and Carrere (2006).
11Krugman is credited with the introduction of the New Trade Theory, due to his research on the
patterns of trade and role that economies of scale play in the determination of these patterns (Krugman,
1980)
12Anderson and van Wincoop (2003) provide a theoretical derivation of the gravity equation that
stresses the importance of multilateral resistance terms.
13In his 2003 article, Melitz incorporates heterogeneous firms into a trade model and thereby, provides
the theoretical basis for a new trend in the international economics literature that focuses on the role that
individual firms with varying productive efficiencies play in the determination of trade, labeled as the New
New Trade Theory.
14
to a 170% ad valorem tax of a representative rich country. Furthermore, the authors
point out that trade costs can be divided into three main categories, which include border
related costs, local distribution costs and, most importantly for this study, transportation
costs.14 In particular, according to the authors, 21% of all trade costs are attributed to
transportation costs.15
In addition to these empirical findings, it is important to point out that the majority
of international transactions are facilitated by seaborne transportation services. According
to Rodrigue et al. (2013), as of 2008, roughly 90%16 of the volume, measured in shipped
tonnage, and about 73% of the value of international trade was handled by seaborne
transportation. As container shipping is one of the major contributors to seaborne
transportation17, these figures highlight the importance of the international container
shipping industry in explaining trade costs and motivates my research in this particular
industry.
14Hummels (2001) points out that explicit trade costs such as tariffs and freight rates are more
significant contributors to trade costs than implicit determinants such as common language and colonial
linkages.
15Anderson and van Wincoop (2004) point out that this is a rough estimate, and, indeed, there is some
debate over the true effects of transport costs and trade liberalization on trade flows and growth. While
Baier and Bergstrand (2001) find that the reduction of transport costs contribute only 8% of total trade
growth, compared to 33% due to trade liberalization, a more recent study by Bernhofen et al. (2016) finds
that the reduction in transport costs due to containerization has had a much more significant impact
on trade growth than trade liberalization efforts. In particular, the authors find that the cumulative
average treatment effect (ATE) of containerization on ’North-North’ trade 15 years after treatment is
1240% compared to free trade agreements and GATT which have a cumulative ATE of 68% and 194%,
respectively. Furthermore, Hummels (2007) points out that transportation costs on U.S. imports far
outweighed the costs imposed by tariffs in 2004. However, Hummels (2007) argues that the cost reduction
of air rather than seaborne transportation has been a critical factor driving the second era of globalization.
16The data used by Rodrigue et al. (2013) was obtained by IHS. Global Insight, Inc., World Trade
Services and does not include intra-EU trade.
17According to the U.S. Department of Transportation Maritime Administration (2013) containerized
trade, measured in metric tons, accounted for 18.1% of total U.S. waterborne trade in 2011.
15
Transportation Economics
Transportation economics have a rich background within the Industrial Organization
literature. The field spans several different industries besides the international liner
shipping industry, such as railroads and trucking, for example. This subsection will
highlight advances in a variety of industries, while pointing to the potential overlap and
applicability, as well as the limitations of the theoretical and empirical results obtained
across these industries.
One of the earlier works, which directly ties into the liner shipping industry,
is the study by Nicholson (1958), who analyzed the minimum rate regulation in the
California trucking industry. The author finds that one of the main shortcomings of the
regulation was the fact that the minimum rate assessment did not incorporate the joint
cost component that exists when serving two markets linked by a round trip. This joint
cost component is one of the key elements to modeling the liner shipping industry. A
carrier operates on round trips and has to take the inseparable joint cost associated with
providing capacity to a transport market pair into account.
Since then, the issue of joint costs has been connected to the backhaul problem
in various transportation industries and has received a fair amount of attention in
the previous literature. Basemann and Daugherty (1977), for example, claim that an
insignificant differential in costs between empty and full backhauls allows the backhaul
to be viewed as a complementary good of the fronthaul. The simple general equilibrium
model employed by the authors shows the existence of a competitive equilibrium with
empty backhauls, while regulation is shown to be inconsistent with empty backhauls.
However, this result is only obtained at backhaul freight rates equal to zero which in turn
is only feasible if the additional marginal cost between full and empty backhauls tends
to zero. The freight rate differentials observed in the container shipping industry reflect
16
a similar pricing behavior, although backhaul rates are on average statistically different
from zero.18 This observation is intuitive when considering the dimensions of seaborne
transportation. Although there may be small differences between the cost of moving an
empty and a fully loaded container per mile traveled, these differences add up quickly
when considering vessels that carry 10,000 or more containers and travel over 6,000 miles
on the backhaul leg.
Demirel et al. (2010) introduce search time as a market friction to overcome this
issue. The authors use a matching model to demonstrate that backhaul rates of inland
barge shipments depend on the search time spent by carriers. Since search times are
costly, this market friction provides a theoretical explanation for non-zero backhaul rates
that are suggested by Basemann and Daugherty (1977). With the inclusion of search
costs, this study hints at the importance of accurately each of a given carrier’s cost
components.19
As previously noted, the issue of empty backhauls from a major importing region
is not unique to the container shipping industry. The trucking industry, for example,
faces the same issue of empty backhauls on a much smaller scale and generally at a
domestic level. Again, the main problem is described by binding capacity constraints on
the fronthaul and uncertain and insufficient demand for transportation on the backhaul.
Wilson (1987) analyzes firm behavior under this uncertainty. In particular, he studies
trucking firms’ decisions to serve markets under the ICC regulation and varying backhaul
probabilities. Jordan (1987) presents two alternative mathematical models to analyze
the trucking industry operating on networks with multiple nodes. He finds that the
18Average freight rates by market are provided in Table 1.
19One example of a careful treatment of transportation costs is the study by Baumol and Vinod (1970).
The authors introduce a thoroughly developed cross-industry cost structure, in order to study the modal
choice of shippers and derive the demand for transport from an inventory theoretic model.
17
heuristic model he sets up minimizes the empty truck miles under the restrictions that
only one backhaul load can be carried and that the backhauls between terminal pairs
must be balanced. Rietveld and Roson (2002) develop a monopolistic model addressing
the backhaul problem and direction dependent pricing. Their main finding is that price
restricting policies that limit the discriminatory pricing of the monopolist can lead to
an overall welfare loss, where both the supplier and average consumer lose. All of these
papers relate to the container shipping industry in certain aspects. Seaborne vessels travel
in round trips between two trading regions and face uncertainty about the backhaul
demand. Due to this uncertainty, it is common practice that prices depend on the
direction of trade flow and can exhibit large differentials within a given market pair.
However, as the research by Felton (1981) shows, the trucking industry is subject
to a set of regulations that not only heavily influences trucking freight rates and the
commodities transported, but is also very different from those regulations20 the container
shipping industry must comply with. Even after the trucking deregulation through the
late 1970’s and the Motor Carrier Act of 1980, Wilson and Beilock (1994) find that the
remaining entry regulation has a significant effect on the access of regulated markets
which continues to cause empty backhauls. In addition, Wilson (1994) documents that the
trucking regulation leads to underutilization and displacements of capacity across varying
markets. The dynamics created by the remaining trucking regulation, although similar
to those found in the container shipping industry, come about for very different reasons.
Therefore, a cross-industries comparison, drawing conclusions over the container shipping
industry, might be quite misleading.
20See the previous discussion of the institutional background of the container shipping industry and
Federal Maritime Commission et al. (2000) for further regulatory detail.
18
Liner Shipping Industry
Due to these distinct differences that set the liner shipping industry apart from
other transportation industries and because of its importance concerning the facilitation of
international transactions, it has been the subject of a variety of economic research. One
strand of this research, for example, focuses on the impact of containerization on global
economic developments. One of the seminal studies in this literature is the book, titled
”The Box”, by Levinson (2010). In his book on the evolution of containerized shipments,
the author points out that one of the most contributing factors to the development of the
world economy has been the sharp reduction in freight rates caused by the cost savings
incurred due to containerization. While this observation alone raises many research
questions of great importance, the attention liner shipping has received historically has
been rather small. Perhaps the reason for this lack of discussion of the container, among
all the other 20th century innovations, is due to the gradual adoption of this costly
technology. In her research, Rua (2012) studies the effects of fixed costs and networks on
the adoption rate of containerized shipping. The author finds that the adoption is driven
by a country’s volume and expected volume of trade, as well as the country’s institutions
and ties to other trading partners utilizing containerized shipments.
Other research revolving around the liner shipping industry has focused on the price
setting behavior of carriers and carrier conferences. Historically, prior to the deregulation
of the industry, the concern of the Federal Maritime Administration has been the price
discrimination imposed by liner conferences acting under antitrust immunity. Heaver
(1973), Lipsey and Weiss (1974) as well as Talley and Pope (1985) all analyze the price
setting behavior of liner conferences to find the main determinants of container freight
rates and evaluate the accusation of price discrimination. One of the general findings
is that freight rates depend on the characteristics of the cargo shipped, which could
19
potentially imply price discrimination. Sjostrom (1992), however, argues that some of the
previous estimations have utilized a reduced form equation that suffers from simultaneity
bias which undermines these findings. In his view, it has not been established whether
liner conferences are price discriminating. Furthermore, the author claims that additional
investigation into independent variables that only affect the demand for transport without
affecting carrier costs, is necessary to remove ambiguity and overcome the simultaneity
bias. In summary, these studies indicate the driving factors behind liner price setting and
demonstrate the difficulty in accurately modeling the liner shipping industry.
The potential price discrimination by carrier conferences is not the only issue that
has sparked interest in container freight rates. Freight rate differentials across varying
market pairs have also drawn attention. The study by Fan et al. (2014) presents a
nonstructural time series analysis of the freight rate differentials and the cointegration
of the joint prices of fronthaul and backhaul. Through this time series analysis of the liner
shipping freight rates, the authors find that cointegration depends on the trade imbalance
and that the imbalance threshold for cointegration changes between regional market pairs.
Generally, a larger trade imbalance reflects a higher probability of an empty backhaul
and is observed to be correlated with disintegrated freight rates. This disintegration
stems from the fact that the joint costs of providing capacity to a transport market pair
serving on a round trip between two countries is no longer shared between the individual
transport markets, but is, instead, completely allocated towards the fronthaul market.
These empirical investigations of liner shipping freight rates are complemented
by several studies that have developed theoretical models that incorporate the specific
dependencies of freight rates on trade imbalances as well as trade volumes. These studies
have started the integration of transportation economics into trade frameworks and are
presented in the following subsection.
20
International Trade and Transportation
As previously discussed, transportation is an integral determinant of international
trade. In addition to that, transportation industries exhibit several interesting dynamics
that lead to asymmetric trade costs and thus, heavily influence overall trade flows.
Samuelson (1952) was one of the first to analyze the effects of transportation costs
between trading regions on the equilibrium level of trade. Since his introduction of the
’ice-berg’ transportation cost21 to the trade literature (Samuelson, 1954), it has been
widely used for its tractability and implicit inclusion of transportation costs.
The renewed focus on trade costs has brought about a few research efforts that have
studied the effects of endogenously determined transportation costs within a given trade
model. Some of these models still rely on the iceberg cost structure, while others explicitly
model the additive rather than proportional nature of transport costs in the determination
of export prices. Luo (2011) as well as Kleinert and Spies (2011) explicitly model the
transport sector within the given trade framework. Luo (2011) continues the traditional
use of iceberg trade and transport costs. He develops a trade cost index (TCI) and
provides empirical evidence that the TCI is a better proxy for trade costs than distance.
Kleinert and Spies establish the simultaneity between trade and transportation costs via
an investment decision faced by transportation firms. In this model higher trade volumes
encourage transportation firms to incur higher fixed costs to invest in cost reducing
technology. Thus, the resulting drop in transportation rates is caused by an increase in
the trade volume, which in turn is positively affected by the reduction in transportation
costs.
21”The simplest assumption is the following: To carry each good across the ocean you must pay some
of the good itself. Rather than set up elaborate models of a merchant marine, invisible item, etc., we can
achieve our purpose by assuming that just as only a fraction of ice exported reaches its destination as
unmelted ice, so will (...) fractions of exports (...) reach the other country as imports.” (Samuelson, 1954,
p. 268)
21
These models suggest that, regardless of the specific assumptions made,
transportation is endogenous to the determination of trade. Due to this endogeneity,
many papers have empirically analyzed the impact of trade flows and imbalances on
transportation rates. Clark et al. (2004), for example, carefully integrate a variety of
determinants of transportation rates into their model. They find empirical evidence that
supports the theoretical predictions. Distance and trade composition increase freight
rates, whereas trade imbalances and economies of scale, captured through vessel size,
have a negative effect on these rates. Furthermore, the authors show that port efficiency,
which is measured by the GCR index22, and infrastructure are strong determinants of
the transportation rates. In particular, the authors’ main finding is that an increase in
port efficiency from the 25th to the 75th percentile leads to a decrease in transportation
rates of 12%. Wilmsmeier et al. (2006) extend this study by additionally controlling for
private sector participation, inter port connectivity, and customs’ delays. However, the
authors fail to instrument for trade volumes, thereby ignoring the endogeneity concerns
suggested by the theory. Despite the issue of endogeneity in the latter study, both of these
research studies point to the transportation sector and ports, in particular, as an excellent
policy avenue that can have real effects on trade costs and thus, trade flows. The empirical
results found in the this study support this general finding and furthermore, suggest that
there is a long-run relationship between trade and transportation that can be exploited
through maritime transportation policy.
Theoretically, Andriamananjara (2004) explores this policy avenue through a
model of the international transport sector (ITS) that lends itself to the theoretical
evaluation of the effects of a variety of ITS oriented policy measures on transport rates.
She demonstrates that increased competition and deregulation are complementary policy
22The Global Competitiveness Report (GCR) was developed by the World Economic Forum in 1999.
22
tools. More importantly, the author shows that multilateral policy efforts can be more
sustainable than unilateral ones due to the positive effects of increased trade flows caused
by a reduction in transportation costs. Francois and Wooton (2001) develop a theoretical
model that complements these results. Testing their model, the authors find that the
gains from trade to consumers and producers are heavily dependent upon the market
structure in the shipping industry. In particular, a more competitive shipping industry
leads to higher pass through gains from trade liberalization. Hummels et al. (2009) present
additional empirical evidence for Adriamananjara’s finding that increased competition
leads to lower transportation rates. Specifically, Hummels et al. (2009) find that the
international shipping industry exhibits market power that allows for mark ups to drive
up transportation rates. The authors identify the market power via uniform tariff changes
that lead to differential price setting behaviors.
Behrens et al. (2006) are the first to endogenize iceberg transportation costs within
a new economic geography model. The authors utilize ’iceberg-like’ trade as well as
transport costs by measuring the cost in terms of the numeraire commodity. Behrens and
Picard (2011) develop an alternative model that incorporates the same joint dependence
of trade flows and transport costs via a capacity constraint. The authors’ main finding
is that endogenous transport costs dampen the Home Market Effect.23 Takahashi (2011)
provides additional support for this finding through another theoretical approach, while
still relying on iceberg transportation costs. In accordance with Behrens and Picard
(2011), the author finds that trade imbalances between two regions lead to freight rate
differentials which in turn, inversely affect trade imbalances and thus, lead to a dispersion
force on economic activity.
23The Home Market Effect(HME) represents the concentration of an industry in the country that
exhibits the largest domestic demand to take advantage of economies of scale and minimize transportation
costs. The HME was incorporated into the New Trade Theory through Krugman (1980).
23
These theoretical findings are complementary to the results obtained by Jonkeren
et al. (2011), who study the effects of trade imbalances between European trading regions
on the freight rates charged on inland waterways. Accounting for the endogeneity of
trade imbalances, the authors find that an increase in the imbalance between region A
and B leads to an increase in the rate charged from A to B. In contrast to this result,
Tanaka and Tsubota (2014) find that a 10% increase in relative exports between Japanese
prefectures leads to a 2.1% decrease in the freight rate charged on that route. The authors
claim that, in this case, the effects of density economies dominate the trade imbalance
effect. Regardless of the dispute over the sign of the impact, all of these studies highlight
the endogeneity of transportation costs in the determination of trade flows.
Following these recent advancements, the theoretical model developed in this
study links international trade and transportation by explicitly deriving the demand
for transport from a trade framework and carefully developing the supply side of the
transport sector accounting for the dynamics of joint production. As in previous studies,
this inclusion of a separate transport sector effectively endogenizes transport and thus,
trade costs. Similar to a few select papers mentioned above, the model I develop next
incorporates the round trip nature of the international transportation industry and
points to the simultaneity of trade and transportation costs. Empirically, I apply this
theoretical model to three market pairs between the major trading regions of the U.S.,
EU and Asia and evaluate it using time series techniques. In particular, I estimate the
panel cointegration equations underlying the demand and supply relations which govern
the long-run structural relationships between trade and trade costs suggested by the
theory. To the best of my knowledge, this analysis is a novel contribution to the literature
offering new insights into the dynamics of this important industry and lends itself for the
evaluation and simulation of the long-run effects of trade and transportation policy on the
24
equilibrium levels of trade and potential spillover effects from these policies across bilateral
export and import markets.
Theoretical Model
In this section, I develop a system of demand and supply equations that apply to
the international maritime transportation markets. The demand for transportation is
derived from a trade framework, and I develop a model of transport supply that reflects
the simple fact that transportation firms typically haul on fixed schedules between two
countries which gives rise to joint production. The result gives a complete system of trade
and transportation from which the effects of unbalanced trade and policy options can be
considered. Furthermore, the equilibrium conditions illustrate that the integration of trade
and transportation accounts for the simultaneity between trade flows and international
transportation costs and depends on the trade imbalance. A comparative statics exercise
demonstrates that this simultaneity may bias traditional gravity estimations and that the
presence of joint production can alter conventional trade theory results, a finding that is
complementary to the analysis of Deardorff (2014).
Demand for Transport
To begin, I derive an expression of the demand for transport from the international
trade theory expressed by Hummels et al. (2009). In this model of trade, each country,
j=1,2,...,M, is composed of one representative consumer. Preferences of each representative
consumer take a quasi-linear form and are expressed over a homogeneous numeraire
commodity and a variety of a good that is differentiated by national origin, as in
Armington (1969). The price elasticity of demand, σ, is assumed to be constant across
representative consumers and greater than one. Given these assumptions, the preferences
25
of the representative consumer in country j can be expressed by the following utility
function
Uj = q0j +
M∑
i=1
q
(σ−1)/σ
ij ∀j = 1, ..,M, (2.1)
where country j’s consumption of the numeraire commodity is given by q0j and the
consumption of a particular variety sourced from country i is given by qij.
The price of the numeraire is normalized to one and it is assumed that this
good can be traded at no cost. In contrast, the domestic sales price of a variety from
country i is represented by pi and taken as given by carriers. Given the fact that each
representative consumer has a taste for variety24 and goods are differentiated by origin,
there is an incentive for trade, and trade between countries introduces the international
transportation markets. Indeed, trade costs in the form of freight rates become a
determinant of the equilibrium. And, a complete model incorporates the transportation
supply to allow the equilibrium transport rates to be endogenously determined along
with trade. In fact, given that each country engages in trade, the import price, pij, of a
variety from country i paid by the representative consumer in country j includes per-unit
transportation costs, fij, and the ad-valorem trade costs, τij ≥ 1, in addition to the sales
price, pi. That is, pij = piτij + fij, where the transport and ad-valorem trade costs are
taken as given by each representative consumer.25
Utility is maximized by each representative consumer with respect to their budget
constraint. The solution to this constrained optimization problem gives the imported
24The marginal utility received from each variety i approaches infinity as consumption of that variety
goes to zero. Therefore, each consumer prefers at least a small amount of each variety to maximize utility
and hence, each consumer has a taste for variety.
25This type of specification, where trade costs include both an iceberg trade cost component as well as
a trade specific cost component, has been utilized by Feenstra and Romalis (2014) and is consistent with
work by Hummels and Skiba (2004). In this study, the authors point out that transport costs are more
accurately modeled as unit-specific rather than ad-valorem or iceberg trade costs, as first introduced by
Samuelson (1954).
26
quantities by country j from each country i. These imports also represent the demand
for transport from each country i to country j and are given by the following expression;
qij =
[
σ
σ − 1(piτij + fij)
]−σ
∀i, j = 1, ..,M, i 6= j. (2.2)
Of course, this expression for the demand of transport holds for any two countries i and
j engaged in bilateral trade and naturally creates the transport market pair ij for each
carrier. However, it is important to make note of the fact that transport demands do
not have to be equal to one another. In fact, trade flows are rarely equal. Most often
country i is a net exporter to country j. This trade imbalance particularly holds for
containerized cargo flows which implies that demands for transportation in the market
pair ij are imbalanced.26 Following common terminology the leg of the round trip facing
higher demand is denoted as the fronthaul, while its counterpart, the leg of the round trip
facing lower demand, is denoted as the backhaul.27
Supply of Transport
In this study, the theoretical and empirical analyses specifically apply to
containerized traffic between regional pairs and focuses on the overall effect of
transportation costs on trade flows and imbalances under the dynamics of joint
production. More formally, each carrier allocates capacity K to transport market pair
ij, in order to offer transport supply, Qij ≤ K, to transport market ij and facilitate
trade from country i to country j. Given that carriers operate on strictly scheduled
26It is possible, of course, to encounter situations where overall trade may be balanced, while
containerized trade flows remain imbalanced due to the varying trade composition.
27Obviously these terms are inaccurate. By definition, fronthaul and backhaul depend on the trade
imbalance between two regions rather than the direction of trade flow or the starting point of a round trip.
27
round trips28, serving transport market ij inadvertently allows the same carriers to offer
transport supply, Qji ≤ K, in transport market ji to facilitate trade from country j to
country i. Due to inseparable joint costs that arise from providing capacity to serve the
transport market pair ij, these transport supplies are joint products.
When the demands for these joint transport supplies are imbalanced, adherence
to strict schedules prohibits the allocation of search and/or waiting time for additional
cargo and forces carriers to adjust transport supplies accordingly. In equilibrium, this
results in transportation costs that adjust to the existing demand imbalance.29 These
market frictions resulting from the joint production and tight schedules introduce the key
dynamics that separate this theoretical model of a transport sector from the majority of
the previous work in the trade literature and extend the model developed by Hummels
et al. (2009).
I model the international shipping industry to exhibit market power. To
accommodate this feature of the industry and following the derivation by Hummels et al.
(2009), the transport sector is modeled as an oligopoly consisting of l = 1, ..., N symmetric
carriers competing in Cournot fashion. Extending the given model, I assume that each
carrier, l, serving the transport market pair ij facilitates a portion of bilateral trade, qij
and qji, between countries i and j and has a round trip cost structure that is twofold.
In particular, similar to Wilson (1994) and Wilson and Beilock (1994), each carrier faces
market specific access costs, aij, such as additional fuel or terminal costs, for shipping
28Interviewing several industry insiders, including port officials and freight forwarders, it was pointed
out that container vessels, with the exception of extreme circumstances, adhere to strict schedules and
that carriers operate on round trips staggering the vessels they deploy, in order to offer more frequent
service.
29Individual carriers may serve multiple locations on a single round trip or take advantage of hub and
spoke shipping networks to reduce the backhaul problem. However, as the cargo flows depicted in Figures
2b, 2d, and 2f reveal, severe aggregate traffic imbalances prevail despite the potential for such strategies.
Thus, for expositional purposes, I assume that each carrier serves only two regions with each round trip.
Generalization to multiple locations is straight forward.
28
one unit of a variety from country i to country j. In addition, each carrier’s technology
is further defined by the previously mentioned joint costs, JC(K l), with JC(0) = 0 and
∂JC(Kl)
∂Kl
> 0. These costs of providing capacity include, for example, labor, maintenance
and repairs, or insurance costs that are not differentiable between the individual transport
markets and can be viewed as, quite simply, the costs of traveling between the two
locations. Therefore, each carrier’s round trip costs can be expressed as follows:
C l = aijQ
l
ij + ajiQ
l
ji + JC(K
l) ∀l = 1, ..., N and i, j = 1, ...,M, i 6= j. (2.3)
Given this cost structure, each carrier chooses the profit maximizing transport
capacity, K l, and optimal supplies of transport, Qlij and Q
l
ji, that are offered to each
market on a given round trip.30 Each carrier’s profit from a given round trip between
country i and country j can be written as
max
Kl,Qlij ,Q
l
ji
Πl = fijQ
l
ij + fjiQ
l
ji − C l ∀l = 1, ..., N and i, j = 1, ...,M, i 6= j,
subject to K ≥ Qij, K ≥ Qji.
(2.4)
Solving each carrier’s constrained profit maximization problem results in three Nx1
vectors of first-order conditions, along with the standard Kuhn-Tucker conditions, that
30Of course, the dimensionality of each carrier’s optimization problem can be extended to include issues,
such as the dependence of carrier costs on the actual port of entry or uncertainty concerning the reliability
of the hinterland transportation network. Although these issues are important, they go beyond the scope
of this paper. Therefore, I abstract from issues of uncertainty and limit the theoretical analysis to a one
port per country model.
29
can be represented as follows;
∂Πl
∂Qlij
= fij +Q
l
ij
∂fij
∂Qlij
− aij − λ1 ≤ 0 with = if Qlij > 0 ∀l = 1, ..., N (2.5a)
∂Πl
∂Qlji
= fji +Q
l
ji
∂fji
∂Qlji
− aji − λ2 ≤ 0 with = if Qlji > 0 ∀l = 1, ..., N (2.5b)
−∂JC(K
l)
∂K l
+ λ1 + λ2 ≤ 0 with = if K l > 0 ∀l = 1, ..., N (2.5c)
K l ≥ Qlij λ1 ≥ 0 (K l −Qlij)λ1 = 0 (2.5d)
K l ≥ Qlji λ2 ≥ 0 (K l −Qlji)λ2 = 0. (2.5e)
The first oder conditions with respect to transport supplies, given by equations
(2.5a) and (2.5b), can be seen as each carrier’s market access conditions indicating that
marginal revenues in either transport market must cover access costs for a given market
to be served. In addition to that, each carrier’s first order condition with respect to
allocated capacity can be interpreted as the service condition. That is, given the fact
that the Kuhn-Tucker multipliers, λ1 and λ2, can be thought of as the shadow prices that
determine the value of an additional unit of transport supply in the respective transport
markets, equation (2.5c) states that the total value of an additional unit of transport
supply in either market must equal the marginal joint costs of the additional capacity to
be allocated and the market pair to be served at all.
In order to solve for the equilibrium transport supplies and capacity allocation, the
transport market clearing conditions that demand for transport must equal the supply
of transport in each market must be imposed. These market clearing conditions can be
30
represented by the following equations;
qij =
N∑
l=1
Qlij (2.6a)
qji =
N∑
l=1
Qlji. (2.6b)
Combining the demand for transport given by equation (2.2), the first order
and Kuhn-Tucker conditions given by (2.5a)-(2.5e) and the market clearing conditions
represented by equations (2.6a) and (2.6b), several equilibrium solutions can be obtained.
While the details of the derivation are fairly standard, there are a few aspects of the set of
solutions that are important to point out. First, given this model, it can be shown that at
least one of the capacity constraints, K l ≥ Qlij and/or K l ≥ Qlji, must be binding in any
equilibrium solution. This implies that any solution to this static model is characterized
by full capacity utilization in at least one of the two transport markets. While additional
considerations, such as the time that it takes to build a container vessel to adjust capacity,
may introduce market frictions that lead to non-binding capacity constraints in the short-
run, full utilization of allocated capacity is a sensible feature for any of the long-run
equilibrium cases derived from this model.
Second, the set of solutions includes equilibrium cases, where optimal transport
supplies and international trade are zero valued when marginal joint and/or access
costs are prohibitively high for the transport market pair or either of the individual
transport markets. For the purposes of this study, the remaining analysis solely focuses
on cases where equilibrium transport supplies and international trade are positive in
both transport markets. Third, equilibrium solutions involving positive unilateral or
bilateral international trade exist and can be derived in symmetric pairs that simply
interchange the i and j notation. Thus, without loss of generality, I treat the transport
31
market facilitating trade from country i to country j as the fronthaul and the transport
market facilitating trade from country j to country i as the backhaul for the remainder of
the analysis.
Equilibrium Considerations
Given non-prohibitive access and marginal joint costs in fronthaul and backhaul
transport markets, the solution to the model has to distinguish between the balanced and
imbalanced trade case. Naturally, the consideration of whether the balanced or imbalanced
trade equilibrium arises, heavily depends on the imbalance concerning the demands for
transport. Given equation (2.2), this demand imbalance can be represented and rewritten
as follows;
qij ≥ qji =⇒ pjτji − piτij ≥ (fij − fji). (2.7)
As equation (2.7) shows, the size of the trade imbalance depends on the difference in
domestic sales prices as well as the endogenously adjusting freight rate differential.
Intuitively, small differences in the bilateral demands for transport, due to small sales price
variations across country i and j, may allow carriers to choose equal transport supplies
that maximize capacity utilization in both transport markets. The resulting equilibrium
freight rate differential must offset any price differences, so that bilateral trade balances.
Large imbalances concerning the bilateral demands for transport caused by substantial
differences in sales prices between two countries, however, may force carriers to choose
asymmetric transport supplies with excess capacity in the backhaul market. The resulting
equilibrium freight rate differential does not offset the sales price variation and leads to
imbalanced bilateral trade. In fact, it can be shown that a balanced trade equilibrium only
arises when pjτji − piτij ∈
(
aij − aji − ∂JC(Kl)∂Kl , aij − aji + ∂JC(K
l)
∂Kl
)
, whereas an imbalanced
trade equilibrium, where exports from country i to country j exceed exports from country
32
j to country i, results when pjτji − piτij >
(
aij − aji + ∂JC(Kl)∂Kl
)
. 31 Both of these two
scenarios can be summarized graphically and are depicted by Figures 1a and 1b.
$
TEU's
DFD
B
ijf
Q = Qij
f ji
ji
a ji
(a) Balanced Trade
$
TEU's
DFDB
ij         f
Qij
         f ji
Q ji
a ji
(b) Imbalanced Trade
FIGURE 1. Linkages between Trade Balance and Transportation Cost Differentials
Figure 1a demonstrates the balanced trade case. In this scenario, the difference
between fronthaul demand, DF , and backhaul demand, DB, is rather small. Given this
small difference in demands for transport, each carrier’s optimal choice leads to symmetric
transport supplies, Qlij = Q
l
ji, which in turn leads to asymmetric equilibrium freight rates,
fij 6= fji. As Figure 1a shows, the size of the potential freight rate differential depends
on the actual imbalance of the demands for transport. Furthermore, Figure 1a illustrates
that this differential between freight rates mitigates the difference in sales prices (inclusive
of ad-valorem trade costs), effectively equalizing the equilibrium demands for transport,
and thus, leading to balanced bilateral trade. If we, instead, maintained the traditional
symmetric trade cost assumption, while allowing sales prices to vary across countries,
31Due to symmetry, trade is also imbalanced when piτij −pjτji > (aji − aij + JC ′). In this case, country
j becomes the net exporter and the transport market ji becomes the fronthaul.
33
Figure 1a shows that this symmetry would impose empty containers in the backhaul
transport market that are inconsistent with balanced trade. This highlights an import
result of the theoretical model, which states that actual trades are only balanced when
freight rates are free to endogenously adjust to the demand imbalances and are allowed to
be asymmetric between two trading countries.
In contrast, Figure 1b demonstrates the unbalanced trade case, where fronthaul
demand, DF , is much larger than backhaul demand, DB. Given such a large difference
in demand stemming from a large sales price variation across countries, each carrier
optimizes by choosing asymmetric transport supplies, Qlij 6= Qlji. This, of course,
results in imbalanced bilateral trade in the presence of potentially asymmetric freight
rates, fij ≥ fji. Differentiating between the balanced and imbalanced trade cases, the
equilibrium solutions are presented more formally next.
Case 1: Balanced Trade
For small transport demand imbalances, each carrier’s equilibrium supplies of
transport for a given round trip between country i and j and the resulting equilibrium
transportation rates can be derived as follows32:
K l = Qlij = Q
l
ji =
1
N
[
σN
2(σN − 1)
σ
σ − 1
(
aij + aji +
∂JC(K l)
∂K l
+ piτij + pjτji
)]−σ
(2.8a)
fij =
σN
2(σN − 1)
[
aij + aji +
∂JC(K l)
∂K l
+ pjτji
]
+
2− σN
2(σN − 1)piτij (2.8b)
fji =
σN
2(σN − 1)
[
aij + aji +
∂JC(K l)
∂K l
+ piτij
]
+
2− σN
2(σN − 1)pjτji. (2.8c)
32Note that the derivation of the optimal supplies of transport and resulting equilibrium freight rates
relies on the symmetry of carriers.
34
Case 2: Imbalanced Trade
Solving the model when the demands for transport are strongly imbalanced yields
the following expressions for each carrier’s equilibrium supplies of transport and capacity
allocation, as well as the respective equilibrium transportation rates for a given round trip
between country i and j:
K l = Qlij =
1
N
[
σ
σ − 1
σN
σN − 1
(
aij +
∂JC(K l)
∂K l
+ piτij
)]−σ
(2.9a)
Qlji =
1
N
[
σ
σ − 1
σN
σN − 1 (aji + pjτji)
]−σ
(2.9b)
fij =
σN
σN − 1(aij +
∂JC(K l)
∂K l
) +
1
σN − 1piτij (2.9c)
fji =
σN
σN − 1aji +
1
σN
pjτji. (2.9d)
Thus, in the balanced trade case, the partial equilibrium,
(
qij, qji, K
l, Qlij, Q
l
ji, fij, fji
)
,
of transport market pair ij facilitating balanced bilateral trade between country i and
j is described by equations (2.2) and (2.8a)-(2.8c). In contrast, in the imbalanced trade
case, the partial equilibrium,
(
qij, qji, K
l, Qlij, Q
l
ji, fij, fji
)
, of transport market pair ij
facilitating imbalanced bilateral trade between country i and j is described by equations
(2.2) and (2.9a)-(2.9d). Both equilibrium cases combined exhibit several key features that
are present when trade is facilitated by an international transportation industry that is
subject to the backhaul problem.
While marginal access costs play a role in the determination of transport supplies
and equilibrium freight rates regardless of the demand imbalance facing carriers, the
allocation of marginal joint costs is heavily dependent upon this imbalance. That is, in
the balanced trade case, equations (2.8a)-(2.8c) show that marginal joint costs matter to
35
the determination of both fronthaul and backhaul equilibrium transport supplies as well as
freight rates. In contrast, equations (2.9a)-(2.9d) demonstrate that, in the imbalanced
trade case, marginal joint costs only matter to the determination of the equilibrium
fronthaul transportation supply and the equilibrium fronthaul freight rate.
Overall, the above system of theoretical equations provides the basis for the
empirical work. It describes the transport market equilibrium facilitating bilateral trade
between two countries and allows for comparative statics that are commonly done in
the trade literature. Some of these comparative statics are highlighted in the following
subsection.
Comparative Statics
One of the key results obtained from various estimations of the gravity equation is
the dependence of trade on aggregate income and trade costs. As Head and Mayer (2013)
point out, standard trade estimations use proxies, such as distance and cultural as well as
geographical ties between trading countries to capture trade costs. However, these trade
cost proxies have limitations. For example, Limao and Venables (2001) point out that
distance is only weakly related to transport costs, while Combes and Lafourcade (2005)
show that it fails to correlate with time-varying transport costs. Thus, in the absence of
a unit-specific trade cost proxy or in cases where common controls do not correlate with
transportation rates, failure to model the endogeneity of international shipping costs
may result in biased coefficient estimates. The existence of such a potential bias can be
shown with the theoretical model above. As equations (2.8a)-(2.8c) as well as (2.9a)-(2.9d)
illustrate, the determination of equilibrium transportation rates partly depends on the
determinants of international trade, namely the domestic sales prices. Therefore, unit-
specific trade costs captured by international shipping rates cannot be held constant given
36
a change in these sales prices. Instead, the simultaneous change in trade costs leads to a
secondary impact on trade that alters the initial response.
Consider, for example, an exogenous shock to country i’s domestic sales price. Given
equation (2.2), the partial derivative that captures the overall response of trade from
country i to country j to a shock in country i’s sales price is given by:
∂qij
∂pi
= −σ
[
σ
σ − 1(piτij + fij)
]−σ−1
σ
σ − 1
[
τij +
∂fij
∂pi
]
(2.10)
If freight rates are assumed to be exogenous to the system, it must be true that
∂fij
∂pi
= 0.
However, if equations (2.8b) and (2.8c) as well as (2.9c) are considered, it becomes clear
that equilibrium freight rates depend on the domestic sales price regardless of the trade
imbalance. To derive the specific response of freight rates, I differentiate between the
balanced an imbalanced trade scenarios.33 In the imbalanced trade case, the responses
of the fronthaul and backhaul equilibrium freight rates to a shock in country i’s sales price
are given by:
∂fij
∂pi
=
1
σN − 1τij > 0 (2.11a)
∂fji
∂pi
= 0. (2.11b)
33I abstract from knives’ edge cases where a change in sales prices causes a switch in the equilibrium
from balanced to imbalanced trade or vice versa.
37
In contrast, an identical shock in the balanced trade case leads to the following responses
of the fronthaul and backhaul equilibrium freight rates :
∂fij
∂pi
=
2− σN
2σN − 2τij (2.12a)
∂fji
∂pi
=
σN
σN − 1τij > 0. (2.12b)
These derivatives highlight several important findings. First, regardless of whether
a balanced or imbalanced trade equilibrium is considered, equations (2.11a) and (2.12a)
show that fronthaul freight rates adjust to an increase in the net exporter’s sales price.
However, the size and direction of the potential bias varies between the balanced and
imbalanced trade cases. If N > 2, the fronthaul freight rate, in the balanced trade
case, decreases in response to the price shock, while the fronthaul freight rate increases
in the imbalanced trade case. Intuitively, while an increase in the sales price causes
the demand for transport to fall in both cases, carriers exercising market power adjust
the fronthaul transport supply differently across the two scenarios. In the imbalanced
trade case, carriers are unconstrained concerning their adjustment of transport supply
viewing fronthaul and backhaul as separate products. In contrast, in the balanced trade
case, carriers are constrained to keep transport supplies symmetric across fronthaul and
backhaul markets leading to a smaller supply adjustment. Overall, this derivation shows
that as long as carriers hold market power, the common estimate of
∂qij
∂pi
clearly depends
on the endogenous adjustment of unit-specific trade costs in response to a change in the
determinants of trade and may bias traditional gravity estimations.
Second, equations (2.11b) and (2.12b) demonstrate that the backhaul freight rate,
in the imbalanced trade case, remains unchanged, whereas the backhaul freight rate,
in the balanced trade case, increases in response to a change in country i’s sales price.
38
This result stems from the presence of joint production in the transportation industry.
When trade is balanced, fronthaul and backhaul supply are joint products which leads
to the integration of fronthaul and backhaul freight rates. Factors that drive a change
in the fronthaul market will trigger a response in the backhaul transport market as
well. Specifically, since the fronthaul transport supply decreased in the balanced trade
case, a shock to the net exporter’s domestic sales price must also cause a reduction in
the backhaul transport supply, so that trade remains balanced. This reduction in the
backhaul transport supply, naturally leads to the increase in the backhaul freight rate
given by (2.12b). Since the elasticity of trade with respect to transport costs is negative,
this, of course, results in a reduction of trade facilitated in the backhaul market and an
unanticipated spillover effect. Therefore, this model, build on a standard trade framework
and accounting for the presence of joint production in the international transportation
industry, predicts that when trade is balanced, a trade shock pertinent to country i’s
exports also leads to an adjustments of country i’s imports. This result compliments the
finding by Deardorff (2014) showing that not only the presence of unit-specific trade costs,
but also their cointegration can distort traditional trade theory results.
Empirical Model
Based on the partial equilibrium conditions (2.2) and (2.8a)-(2.8c) as well as (2.9a)-
(2.9d), I develop the empirical model to test whether the theoretical simultaneity between
trade and transport costs holds in the data. In particular, for any of the given transport
market pairs, the estimation is focused on the demand for transport and fronthaul and
backhaul pricing relations described by the static partial equilibrium framework. Since the
data are quarterly time series observations in three market pairs and the objective of the
empirics is to uncover the static long-run equilibrium relations implied by the structural
39
model, the empirical specifications allow for the use of time series techniques to estimate
the static long-run system of equations. In particular, I employ panel cointegration
methods to estimate the structural equations underlying the theoretical long-run partial
equilibrium model. Indeed, as Hamilton (1994) states:
”Cointegration can be viewed as a structural assumption under which certain
behavioral relations of interest can be estimated (...)” (Hamilton, 1994, p. 589)
First, the demand equation is considered. As equation (2.2) indicates, the demands
for transportation in market pair ij are given by the quantity of containerized bilateral
trade facilitated between region i and region j and are a function of sales prices, ad-
valorem and unit-specific trade costs. Furthermore, the theoretical model suggests that
there are no inherent differences between fronthaul and backhaul transport markets
concerning the dependence of trade on these determinants. Thus, I estimate the demand
for transport via a single equation, where the quantity of transport demanded between any
two regions i and j is denoted by qijt. The cross-sectional dimension of a trade route is
indicated by ij, while the time series dimension of the data is given by t. The domestic
sales price is denoted pit, whereas unit-specific trade costs are given by the container
freight rate, fijt, that is charged to facilitate trade from region i to region j. Although
the theoretical model, due to its partial equilibrium nature, does not indicate aggregate
income to be a determinant of international trade, I follow the vast majority of the trade
literature that suggests that aggregate income plays a central role in the determination
of trade and thus, the international demand for transport. Following standard practice,
aggregate income is given by exporter and importer real GDP and denoted by yit and yjt,
respectively.
To capture the unobservable characteristics and control for the heterogeneity
between trade routes, a transport market specific fixed effect, αij, is included in the
40
model. This follows the fixed effect specification suggested by Cheng and Wall (2005) and
captures time-invariant ad-valorem trade costs.34 Panel cointegration tests developed by
Pedroni et al. (1999) and Pedroni (2004) are used to allow for this heterogeneity across
panels and inform about the necessity of market pair specific time trends. Based on the
test results, I do not integrate a market pair specific time trend in the empirical model of
the demand for transport.
The estimation of the panel cointegration relations in a heterogeneous panel is
based on the Panel Dynamic OLS (DOLS) and Fully Modified OLS (FMOLS) estimators
developed by Pedroni (2001) and Pedroni (2000), respectively. In addition to the
previously discussed variables, the DOLS estimator also includes lagged and lead terms
of the first differences of all the right hand side variables to control for the dynamic
properties of the data. More specifically, these terms, summarized in vector ∆xijt,
control for the endogenous feedback effect that is present between international trade
and unit-specific trade costs as well as the other determinants of trade. Consequently,
the theoretically motivated empirical specification of the demand for transport becomes
qijt = αij + β1fijt + β2pit + β3yit + β4yjt +
S∑
s=−S
Θij∆xijt+s + ijt, (2.13)
where all variables are in logged form, S indicates the maximum number of lags and leads
included in the model and the error term is denoted by ijt. The cointegration relation
and coefficients of interest are described by β1-β4. Accurately estimating the cointegration
34Other empirical trade studies, particularly those estimating gravity models, have included a
variety of ad-valorem trade cost proxies. According to Head and Mayer (2013), the traditional proxies
include dummy variables for contiguity, common official language, colonial linkages and Regional Trade
Agreements (RTA’s) as well as Free Trade Agreements (FTA’s). Due to the fact that the cross-sectional
dimension of the data is at a supranational level, these country-specific effects cannot be separately
included in the empirical model. However, to the extend that ad-valorem trade costs differ across
supranational geographic regions, time-invariant ad-valorem trade costs are captured by the market pair
specific fixed effects.
41
relation underlying the demand equation renders the residual stationary and implies that
any variation from this long-run static equilibrium relationship is only temporary.35
To complete the empirical model and demonstrate the simultaneity between trade
and transportation, I develop the empirical specifications of the theoretical pricing
relations, as suggested by equations (2.8b) and (2.8c) as well as (2.9c) and (2.9d), next.
Following the theoretical model, two equilibrium pricing relations distinguishing between
fronthaul and backhaul transport markets are considered. The left-hand side variables are
given by the international container freight rates, fijt and fjit. These unit-specific trade
costs are each modeled as a function of the respective number of carriers, as well as access
and marginal joint costs. The number of carriers competing in market pair ij is captured
via this market pair’s cumulative shipping capacity, scijt. Access costs are controlled for
via bunker fuel prices, denoted by bfpijt. While it is expected that an increase in the
price of bunker fuel raises access cost and thus, increases the equilibrium freight rates,
an increase in market shipping capacity is associated with intensified competition that
diminishes market power and is, thus, expected to lead to a reduction of international
freight rates.
Lastly, I specify a proxy for marginal joint costs. As equations (2.8b) and (2.9c)
demonstrate, marginal joint costs are a determinant of fronthaul freight rates regardless of
the trade imbalance. However, equations (2.8c) and (2.9d) illustrate that the dependence
of the backhaul freight rate on marginal joint costs varies between the balanced and
imbalanced trade equilibrium scenarios. To capture this switching dependence of the
35In the gravity literature, Santos Silva and Tenreyro (2006) point out that log transformations require
the assumption of a log-normal error term and, furthermore, require the observations with zero trade
flows to be excluded from the estimation sample. Since the sample used in this study includes time series
observations on only three market pairs comprised of trading regions at a supranational level, there are
no zero valued trade flows contained in the dataset. Furthermore, the existence of a cointegration relation
renders the error term stationary which is the critical assumption for the group-mean panel estimators
employed in this study.
42
backhaul freight rate and control for marginal joint costs, the relative trade imbalance,
δijt, is integrated in the fronthaul and backhaul empirical pricing relations. Careful
consideration of the theoretical model suggests that an increase of the trade imbalance
is associated with an increased allocation of marginal joint costs towards the fronthaul
transport market. That is, fronthaul freight rates are expected to increase, given a rise of
the trade imbalance. In contrast, this reallocation of marginal joint costs away from the
backhaul transport market, caused by an increase in the trade imbalance, is associated
with a reduction in the backhaul freight rate which no longer covers the reallocated
portion.
Again, market pair specific fixed effects to account for the heterogeneity across
market pairs and lead and lagged terms of the first differenced right hand side variables,
denoted by vector ∆zijt, to control for the endogenous feedback effects are incorporated.
The cointegration tests reveal that neither of the long-run equilibrium pricing relations
include a market pair specific time trend. Motivated by the theoretical pricing relations
(2.8b) and (2.8c) as well as (2.9c) and (2.9d), this leads to the following empirical pricing
relation specifications;
fijt = αij + γ
ij
1 scijt + γ
ij
2 bfpijt + γ
ij
3 δijt +
S∑
s=−S
Φij∆zijt+s + νijt, (2.14a)
fjit = αji + γ
ji
1 scjit + γ
ji
2 bfpjit + γ
ji
3 δjit +
S∑
s=−S
Φji∆zjit+s + νjit, (2.14b)
where the error terms are given by νijt and νjit and the parameter’s of interest are given
by γij1 -γ
ij
3 and γ
ji
1 -γ
ji
3 for fronthaul and backhaul transport markets, respectively.
Standard panel data estimation of this system is complicated by the potential non-
stationarity of various time series in the data, which may mask the structural equilibrium
relationships implied by the theoretical model. To this end, I proceed with a thorough
43
investigation of each of the time series in the system, including several tests for panel unit
roots as well as panel cointegration. Following these tests, the estimation of the demand
and pricing relations proceeds equation by equation using Pedroni’s group-mean Panel
DOLS and FMOLS estimators. The use of these techniques to estimate the cointegration
relations addresses concerns of endogeneity of right hand side variables. As noted by
Hamilton (1994), the potential for spurious regressions due to unit roots is accounted for
by the existence of a cointegration relation. Furthermore, given cointegration, a system
of equations with i.i.d. errors can, under certain conditions, be estimated via equation by
equation OLS, despite the potential simultaneous equations bias. According to Pedroni
(2001), OLS estimates may still suffer from a second order bias which warrants the use of
group-mean panel estimators. Thus, after checking for the existence of panel unit roots
in each of the time series and establishing the existence of cointegration relations, each
equation is estimated via the DOLS and FMOLS estimators.
Data
The data that are used to estimate the parameters of the empirical model have been
obtained from various sources. Gross Domestic Product and Consumer Price Index (CPI)
data to control for aggregate income and domestic sales prices for the U.S., the Euro-
Area and several Asian countries have been obtained from the OECD Main Economic
Indicators database. Since the cross-sectional dimension considers trade at supranational
levels (except for the U.S.), Asian GDP is controlled for via the cumulative GDP of Japan,
South Korea, India and Indonesia, while the Asian sales price is controlled for via the
average CPI of Japan, South Korea, India, Indonesia, and China. The data on regional
shipping capacity have been obtained from the United Nations Conference on Trade and
Development (UNCTAD) database. Market access costs in the Trans-Pacific and Trans-
44
Atlantic markets are given by the bunker fuel prices in Los Angeles and Philadelphia,
respectively, and have been obtained from the Shipping Intelligence Network.36 Data
on the left-hand side variables, containerized cargo flow and regional freight rates have
been obtained from Drewry and Containerisation International via the annual reports
by UNCTAD, Secretariat (2014), respectively. The majority of the data are observed at
quarterly or annual frequencies and span a time frame from the fourth quarter of 1995 to
the fourth quarter of 2009.37 While all variables used in the estimation of the empirical
model are seasonally adjusted and in logged form38, the seasonally unadjusted level data
on these variables are summarized in Tables 1 and 2.
In accordance with the theoretical model, the data have been categorized into
fronthaul and backhaul transport markets between the various ij market pairs. These
market pairs include the Trans-Pacific Market which is defined as the container cargo
flow between the U.S. and Asia, the Trans-Atlantic Market defined as the container cargo
flow between the U.S. and EU, and the Asia-EU Market including container trade flows
between these two regions. The mean values of container cargo flow and freight rates,
listed in Table 1, indicate large trade imbalances and freight rate differentials in the Trans-
Pacific as well as Asia-EU market. The Trans-Atlantic market, however, exhibits less
distinguished imbalances, on average.
Figures 2a-2f provide additional evidence in support of these initial observations.
The figures depict the unadjusted freight rates and containerized trade flows for each
transport market pair. Imbalances and differentials, present in the Trans-Pacific and
36Bunker fuel prices for the Asia-EU market were unobtainable. Thus, I have chosen for these prices
to be equal to the average between the available two measures. This, however, introduces artificial cross-
sectional dependencies that, if adjusted, destroy all variation of this measure.
37Market shipping capacity data is only available at annual frequency and has been linearly interpolated
to quarterly frequency.
38Seasonal adjustments have been performed via the X11 routine. This standard procedure recognizes
linear interpolation and leaves the interpolated variables unchanged.
45
TABLE 1. Summary Statistics - Trade and Unit-specific Trade Costs
(1) (2) (3) (4) (5)
VARIABLES N mean sd min max
Regional Cargo Flow (million TEUs)
Fronthaul Qty: Trans-Pacific Market 56 2,412 942.4 895.1 3,883
Backhaul Qty: Trans-Pacific Market 56 1,169 352.2 726.0 1,970
Fronthaul Qty: Asia-EU Market 48 1,556 587.4 810.2 2,642
Backhaul Qty: Asia-EU Market 48 843.5 193.2 468.7 1,161
Fronthaul Qty: Trans-Atlantic Market 56 528.2 83.01 348.4 660
Backhaul Qty: Trans-Atlantic Market 56 423.3 76.45 320.8 597
Regional Freight Rates ($ per TEU)
Fronthaul Rate: Trans-Pacific Market 57 1,717 226.3 1,232 2,203
Backhaul Rate: Trans-Pacific Market 57 930.8 217.0 721 1,517
Fronthaul Rate: Asia-EU Market 57 1,491 274.8 897 2,109
Backhaul Rate: Asia-EU Market 57 853.1 167.0 601 1,257
Fronthaul Rate Trans-Atlantic Market 57 1,414 219.9 1,045 1,854
Backhaul rate: Trans-Atlantic Market 57 1,117 255.6 778 1,637
Sources: Containerized Cargo Flow data - Drewry and Freight Rate data - Containerisation International
Asia-EU market pairs, are large with a clearly defined net exporting and net importing
region. That is, the Trans-Pacific market pair between the U.S. and Asia, exhibited by
Figures 2a and 2b, clearly shows that the trade route from Asia to the U.S. constitutes the
fronthaul transport market, ij, for the majority of the sample period, while the route from
the U.S. to Asia constitutes the backhaul, ji, for the majority of the sample. Similarly,
the market pair between Asia and the EU, which is depicted in Figures 2c and 2d, has a
clearly defined fronthaul transport market, ij, where trade is facilitated from Asia to the
EU and a subsequent backhaul, ji, where trade is facilitated from the EU to Asia. The
freight rate differentials in these markets mirror the clear distinction between fronthaul
and backhaul trade flows.
In contrast, Figures 2e and 2f show that the Trans-Atlantic market pair exhibits
switching trade imbalances that roughly coincide with switching freight rate differentials.
46
TABLE 2. Summary Statistics - Trade and Freight Rate Determinants
(1) (2) (3) (4) (5)
VARIABLES N mean sd min max
Regional GDP (trillion U.S.$)
U.S. GDP 57 13.21 1.454 10.41 15.17
Euro-Area (19) GDP 57 11.05 0.914 9.416 12.51
Asia GDP 55 9.696 1.410 7.952 12.31
Regional CPI (2010=100)
U.S. CPI 57 84.36 9.024 70.40 100.6
Euro-Area (19) CPI 57 86.51 7.423 75.20 98.90
Asia CPI 57 76.92 10.42 59.46 97.40
Regional Shipping Capacity (1000 DWT)
Shipping Capacity - Transpacific market 57 20,621 4,812 13,836 34,151
Shipping Capacity - Asia-EU market 57 42,846 14,696 23,065 77,584
Shipping Capacity - Trans-Atlantic market 57 29,243 9,705 14,948 50,082
Bunker Fuel Prices ($ per ton)
Bunker Fuel Price West Coast (Los Angeles) 57 212.8 141.4 60.85 686.6
Bunker Fuel Price East Coast (Philadelphia) 57 210.5 135.8 61.88 659
Containerized Trade Imbalance
Imbalance - Trans-Pacific Market 56 -0.523 0.141 -0.977 -0.346
Imbalance - Asia-EU Market 48 -0.574 0.0973 -0.733 -0.405
Imbalance - Trans-Atlantic Market 56 -0.808 0.112 -1.000 -0.604
Sources: GDP & CPI - OECD, Bunker Fuel Prices - Shipping Intelligence Network, Market Shipping Capacity - UNCTAD
and Trade Imbalance - Drewry
In particular, the figures reveal that initial observations point roughly balanced bilateral
trade between the U.S. and EU, where fronthaul and backhaul transport markets are not
clearly defined until the second quarter of 1997. In line with the theoretical model, the
freight rate differential is initially relatively small and declines over this time period. In
contrast, observations from the second quarter of 1997 until the second quarter of 2007
exhibit a much larger trade imbalance, where westbound EU to U.S. trade is clearly
defined as the fronthaul transport market. During this period, freight rates adjust to
this stark trade imbalance. As the theory predicts, the freight rate charged to facilitate
westbound EU to U.S. trade becomes much larger than the backhaul freight rate charged
47
on eastbound U.S. to EU trade reflecting the reallocation of marginal joint costs towards
the fronthaul transport market. At the end of the sample period, however, the trade
imbalance switches and the eastbound U.S. to EU trade becomes the fronthaul transport
market. Freight rates adjust to this changing trade pattern, so that the freight charged
on eastbound U.S. to EU trade becomes the larger fronthaul freight rate by the end of
the sample. These observations warrant the careful empirical model specification that
describes the pricing relations in terms of fronthaul, ij, and backhaul, ji, transport
markets, rather than directional east and west bound trade flows that potentially disturb
the fronthaul and backhaul distinction.
The remaining data consists of time series observations for the variables that are
used as aggregate income shifters, sales price controls and shipping cost factors in the
estimation. Table 2 presents the summary statistics on the GDP, CPI, market shipping
capacities, bunker fuel prices and trade imbalance data.
Results
In this section, the empirical results are presented. First, I apply multiple panel
unit root tests to all of the individual time series used in the estimation. Several of
these tests point to non-stationarity of various time series in the data and integration
of order one. Given the non-stationarity, the empirical analysis then proceeds with tests
for panel cointegration developed by Pedroni et al. (1999) and Pedroni (2004). The tests
produce supporting evidence of the existence of cointegration. Based on these results, the
cointegration relations are estimated and the structural demand and pricing relationships
of the container shipping industry that govern the long-run equilibrium of containerized
trade are obtained. These estimates point to the simultaneity between trade and unit-
specific trade costs. The estimation is carried out via the group-mean Panel FMOLS
48
1000
1500
2000
1995 1997 1999 2001 2003 2005 2007 2009
Time
Fr
ei
gh
t R
at
es
 in
 $ 
pe
r T
EU
Freight Rates
Asia−US
US−Asia
(a) Trans-Pacific Freight Rates
1000
2000
3000
4000
1995 1997 1999 2001 2003 2005 2007 2009
Time
Ex
po
rts
 (m
il. 
TE
Us
)
Container Flow
Asia−US
US−Asia
(b) Trans-Pacific Cargo Flows
1000
1500
2000
1995 1997 1999 2001 2003 2005 2007 2009
Time
Fr
ei
gh
t R
at
es
 in
 $ 
pe
r T
EU
Freight Rates
Asia−EU
EU−Asia
(c) Trans-Pacific Freight Rates
1000
2000
1995 1997 1999 2001 2003 2005 2007 2009
Time
Ex
po
rts
 (m
il. 
TE
Us
)
Container Flow
Asia−EU
EU−Asia
(d) Trans-Pacific Cargo Flows
750
1000
1250
1500
1750
1995 1997 1999 2001 2003 2005 2007 2009
Time
Fr
ei
gh
t R
at
es
 in
 $ 
pe
r T
EU
Freight Rates
EU−US
US−EU
(e) Trans-Pacific Freight Rates
400
500
600
1995 1997 1999 2001 2003 2005 2007 2009
Time
Ex
po
rts
 (m
il. 
TE
Us
)
Container Flow
EU−US
US−EU
(f) Trans-Pacific Cargo Flows
FIGURE 2. Trade Imbalances and Freight Rate Differentials
49
and DOLS estimators developed by Pedroni (2000) and Pedroni (2001), respectively. I
conclude this section with a discussion and interpretation of the estimated cointegration
relations and use the specific estimates to simulate the long-run equilibrium impact of
several permanent shocks, such as the reduction of carrier market power, on trade and
freight rates distinguishing between fronthaul and backhaul transport markets.
Unit root tests
Several panel unit root tests on each of the time series that are incorporated in the
estimation of the structural model of trade and transportation have been performed.
The panel unit root tests employed, include the Levin-Lin-Chu (LLC) test developed by
Levin et al. (2002) as well as Phillips-Perron (PP) and augmented Dickey-Fuller (ADF)
tests developed by Choi (2001).39 There are a number of tests available to examine the
existence of panel unit roots. The specific tests chosen for this analysis are built on
assumptions that best fit the data employed in this study. Some of these assumptions that
are common among these tests, include the fact that they are devised for a finite number
of cross-sections as well as the ability to allow for cross-section specific fixed effects and
time trends.40
The results of the panel unit root tests are given in Table 3 and are presented for
the levels as well as first differences of each time series. While the adjusted t-statistic is
reported for the LLC test, the Z-statistic, as recommended by Choi (2001), is reported
for the ADF and PP tests. After careful graphical examination of each level time series,
39These tests are, of course, based on the work by Phillips and Perron (1988) as well as Dickey and
Fuller (1979) and Dickey and Said (1981), respectively.
40Other commonalities include the fact that all three tests maintain the null hypothesis that the time
series exhibits a panel unit root. However, the alternative hypotheses vary across tests. That is, for a
finite number of cross-sections the alternative hypothesis of the PP and ADF tests holds that the time
series of at least one cross-section does not exhibit a unit root. In contrast, the LLC test operates under
the alternative hypothesis that none of the cross-sections exhibit a unit root.
50
a time trend has been included in the regression equations of the tests for the transport
quantity demanded, exporter/importer GDP, CPI, shipping capacity, and bunker fuel
prices. Considering the panel unit root tests on the differenced data no time trends were
included in the regression equations. The issue of lag selection has been addressed with
the Hannan-Quinn Information Criterion (HQIC). Furthermore, none of the time series
were demeaned prior to any of the tests.41
As the test-statistics reported in Table 3 suggest, there is strong evidence that
at least two of the time series in the demand and each of the pricing relations are
integrated of order one. On the demand side, the majority of the tests show that the null
hypothesis of a panel unit root for the level of the transport quantity demanded, CPI and
exporter/importer GDP cannot be rejected. In contrast, test results on freight rates are
mixed. Although the PP test fails to reject the null hypothesis of a panel unit root at the
1% significance level, the LLC and ADF test reject the existence of a panel unit root at
the 1% level. Identical tests applied to the first difference of the non-stationary variables
strongly reject the null hypothesis of a panel unit root suggesting that the transport
quantity demanded, CPI and exporter/importer GDP are, in fact, integrated of order one,
while freight rates may be integrated of order one.
Concerning the pricing relations, all of the tests provide strong evidence that the
level of market shipping capacity and the trade imbalance reflect a panel unit root, while
the evidence of a panel unit root concerning freight rates is rather mixed. That is, for the
fronthaul freight rate the Fisher-type PP test fails to reject the null hypothesis of a panel
41Demeaning panel data is used to control for cross-sectional dependencies. However, exporter/importer
GDPs are identical across some of the three market pairs. For example, the U.S. is an exporter to Asia
in the Trans-Pacific market as well as Europe in the Trans-Atlantic market. This creates cross-sectional
dependencies concerning exporter GDP. Furthermore, bunker fuel prices for the Asia-EU market pair are
generated by averaging the available price measures for the other two market pairs in the panel. Thus,
exact cross-sectional dependence for these time series is an artifact of the data generation, and demeaning
these data distorts important variation. The empirical results of the panel unit root tests are generally
robust to demeaning the remaining time series and are available upon request.
51
TABLE 3. Panel Unit Root Tests
Levels 1st Difference
Fisher-type Fisher-type
LLC DF PP LLC DF PP
Demand
Quantity (qij) -2.68*** 2.00 2.11 -13.76*** -5.78*** -14.23***
Freight Rate (fij) -2.46*** -3.58*** -1.18 -6.92*** -7.89*** -8.60***
Sales Price (pi) -0.54 -1.84** 0.34 -7.23*** -6.51*** -10.31***
Exp./Imp. GDP (yi/yj) 1.54 0.85 4.07 -5.74*** -4.09*** -7.19***
Pricing Relations
FH Freight Rate (ffh) -3.54*** -2.59*** -0.89 -6.98*** -5.16*** -7.75***
BH Freight Rate (fbh) -1.32* -2.37*** -1.47* -8.46*** -6.42*** -7.95***
Shipping Capacity (sc) 1.82 1.62 2.31 0.61 -1.63* -1.55 *
Bunker Fuel Price (bfp) -4.21*** -3.38*** -1.42* -5.41*** -4.90*** -10.09***
Trade Imbalance (δ) 0.28 0.32 0.18 -9.88*** -5.87*** -10.53***
Notes: Reported are the adjusted t-statistics obtained from the LLC test and the Z-statistics for both of the Fisher-type
tests as suggested by Choi (2001). Rejection of the null of a panel unit root at the 1% (5%, 10%) significance level
is indicated with *** (**, *).
unit root at any significance level, while both the LLC and Fisher-type ADF test reject
the null at the 1% level. Similarly, for the backhaul freight rate, both the LLC and Fisher-
type PP tests fail to reject the null hypothesis of the presence of a panel unit root at the
5% significance level, whereas the Fisher-type ADF test rejects the null at the 1% level.
Contrary to these mixed findings, with exception of market shipping capacity, all of the
tests suggest that the first difference of all variables is stationary at the 1% significance
level. Even for market shipping capacity, both of the Fisher-type tests reject the null of a
panel unit root at the 10% level. Based on these tests on the pricing relations variables,
I conclude that market shipping capacity and the trade imbalance are also integrated of
order one, while fronthaul and backhaul freight rates may be integrated of order one.
Given the fact that all variables are integrated of order one or less and that at least
two variables of the demand and each pricing relation are integrated of order one, the
52
empirical analysis continues with the panel cointegration tests developed by Pedroni et al.
(1999) and Pedroni (2004).42 That is, I have found evidence of panel unit root processes
of order one, which point to the possible use of cointegration techniques that allow
for super-consistency and unbiased coefficient estimates, despite potential concerns of
endogeneity or non-stationarity of the individual data series. Specifically, I take advantage
of the information contained within the cointegration equation and estimate the long-run
structural relations projected by the static equilibrium model previously developed.
Cointegration Tests
To allow for the heterogeneity across transport markets, the panel cointegration tests
are based on Pedroni’s (1999, 2004) seven test-statistics and critical values. Due to varying
small sample properties, all test statistics are reported. Since the panel employed for the
demand and pricing relations estimations includes only a short cross-sectional dimension
and a medium length time dimension, rejection of the null hypothesis of no cointegration
of any test, gives evidence for the existence of cointegration relations. Consistent with
the panel unit root tests, none of the time series embedded in the empirical model are
demeaned. However, I do allow for cross-section specific fixed effects and use the tests to
determine the potential inclusion of time trends present in the cointegration relations.
The results of these tests are presented in Table 4 and provide supporting evidence
of the existence of cointegration relations between trade, trade costs, trade prices and
aggregate income on the demand side and trade costs, the trade imbalance, as well as
market power and shipping cost factors on the supply side. In particular, on the demand
side, five out of the seven tests excluding a time trend reject the null hypothesis of no
panel cointegration at the 5% level, while the Panel ρ-test rejects the null at the 10% level.
42As Pesaran et al. (2001) point out the existence of ’level relationships’ is not dependent on all
variables being integrated of order one.
53
This rejection rate declines drastically once a time trend is included. I interpret these
findings as strong evidence for the existence of a trend exclusive cointegration relation
that governs international trade, and thus, the demand for transport, as a function of unit-
specific trade costs, the domestic sales price and exporter as well as importer aggregate
incomes.
Concerning the pricing relations, I test for the existence of cointegration
differentiating between fronthaul and backhaul transport markets. While the cointegration
tests excluding a time trend provide only limited evidence of a cointegration relation
concerning the fronthaul pricing relation (only the Panel ν-statistic rejects the null of
no cointegration at the 10% level), the existence of a time trend exclusive cointegration
relation governing the backhaul pricing relation is strongly supported by six out of seven
tests that reject the null in favor of the alternative hypothesis at either the 5% or 10%
significance level. In contrast, the evidence concerning the existence of time trend inclusive
cointegration relations is less convincing for both pricing relations and thus, estimations
are carried out without a time trend.
Although the evidence is relatively weaker for the fronthaul pricing relation, I
interpret these results as overall supporting evidence for the existence of cointegration
relations that describe the long-run equilibrium relationships suggested by the theoretical
model of trade and transportation. Proceeding with the time series analysis, these
cointegration relations are estimated.
Cointegration Relations Estimation
As noted earlier, the existence of cointegration among the unit root processes leads
to super consistency of the OLS estimates. However, as several studies have pointed out,
the endogeneity takes hold in a second order bias that can have a significant influence,
54
TABLE 4. Panel Cointegration Tests
Panel Cointegration Demand Fronthaul Backhaul
Statistics Pricing Pricing
Panel ν-statistic 2.02** 0.83 1.37* 0.37 2.26** 1.39*
Panel ρ-statistic -1.59* -0.78 -0.54 0.12 -1.40* -1.00
Panel pp-statistic -2.02** -1.57* -0.77 -0.31 -1.68** -1.74**
Panel adf-statistic -1.88** -1.78** -0.89 -0.49 -1.44* -1.47*
Group ρ-statistic -0.78 0.8 0.07 0.70 -0.79 -0.42
Group pp-statistic -1.83** -1.08 -0.55 -0.02 -1.55* -1.50*
Group adf-statistic -2.16** -1.88** -0.83 -0.30 -1.42* -1.40*
Trend no yes no yes no yes
Notes: All statistics are normalized to be distributed N(0,1). For the ν-statistic only the right tail of
the normal distribution is considered, while for all others only the left tail of the normal distribution
is considered as the rejection region for the null hypothesis of no cointegration. *** (**,*)
indicates rejection of the null hypothesis at the 1% (5%, 10%) significance level.
despite this super-consistency. Several estimators have been proposed to address this
second-order bias and obtain unbiased estimates of the cointegration relation of interest.
Among these estimators are various versions of the Fully Modified OLS and Dynamic
OLS estimators. Several studies have used Monte Carlo simulations to better understand
the small sample properties of these estimators and have drawn comparison across them.
Kao and Chiang (1999), for example, show that the ’within-dimension’ DOLS estimator
outperforms both the OLS as well as the ’within-dimension’ FMOLS estimators, and can
be applied for both homogeneous and heterogeneous panels. In response to these findings,
Pedroni (2000) develops a ’between-dimension’ FMOLS estimator and demonstrates
that it performs well for small samples. In line with the literature and to provide a
more complete analysis, I employ and report the results of the group-mean Panel DOLS
estimator developed by Pedroni (2001) as well as the group-mean Panel FMOLS estimator
developed by Pedroni (2000).
55
Guided by the empirical model and following the cointegration tests, the estimation
of the cointegration relations includes panel specific fixed effects, but excludes market
specific time trends. The results of the group-mean panel DOLS estimator are given in
Table 5, while the results of the group-mean panel FMOLS estimator are presented in
Table 6. Generally, I find statistically significant coefficient estimates for all cointegration
relations. In fact, across both estimators all but two coefficient estimates are statistically
significant at either the 1% or 5% significance level and all estimates match the expected
signs. That is, on the demand side, unit specific trade costs as well as domestic sales prices
exhibit a negative correlation with the volume of trade, while aggregate incomes exhibit a
positive correlation with international trade. Concerning the determination of the long-run
equilibrium freight rates, despite the variation in coefficient magnitude, both fronthaul and
backhaul pricing relations show that a persistent rise in access costs, due to increases in
bunker fuel prices, leads to permanent increases in freight rates. In contrast, a permanent
increase in competition via larger market shipping capacity is associated with a long-run
decline in fronthaul and backhaul freight rates. Furthermore, both estimators illustrate
that persistent changes in joint costs have a varying effect on fronthaul and backhaul
transport rates.
Considering the estimated demand relation, given by column (1) in Tables 5 and 6,
in more detail, I find consistency of coefficient estimates across the DOLS and FMOLS
estimator. While the coefficient on freight rates turns up insignificant for the DOLS
estimator, its magnitude of -0.044 is very similar to the FMOLS estimate of -0.058 which
is significant at the 1% level. Focusing on the highly statistically significant estimate
of the FMOLS estimator, the demand cointegration relation suggests that in the long-
run, on average, a 1% permanent increase in freight rates permanently reduces trade by
0.058%. This very inelastic response of containerized trade to a change in unit-specific
56
trade costs appears reasonable when considering the fact that container freight rates are
relatively small compared to the total cargo value of a container.43 Similar to this finding,
the DOLS and FMOLS estimators also show an inelastic long-run response of international
trade to a change in sales prices. Coefficient estimates range from -0.340 (FMOLS) to -
0.597 (DOLS) and are both statistically significant at the 5% level. These findings imply
that a persistent 1% increase of domestic goods prices leads to a 0.340%-0.597% long-run
reduction of international containerized exports.
Additionally, the results show that long-run increases in economic mass, measured by
the exporter’s and importer’s GDP, drive international trade. This finding is consistent
across the DOLS and FMOLS estimators and statistically significant at the 1% level.
Specifically, the DOLS estimator finds that a permanent 1% increase in exporter GDP
raises containerized trade by 0.927% in the long-run, while the same persistent increase in
importer GDP permanently raises trade by 1.189% in the long-run. Coefficient estimates
of the FMOLS estimator suggest that a persistent 1% rise in exporter GDP leads to a
permanent 0.632% increase in trade, whereas a permanent 1% increase in importer GDP
raises containerized trade by 1.326% in the long-run. One common prediction of the
gravity model (see, for example, Anderson and van Wincoop, 2003) is that the effects of
exporter and importer economic mass on trade are theoretically equal to unity. In column
(2) of Tables 5 and 6, I test this hypothesis and find strong evidence in support of it.
Specifically, based on the DOLS coefficient estimates, I fail to reject the null hypothesis
that the long-run equilibrium effects of importer and exporter GDP on containerized
international trade are unit elastic at any significance level. The same is true for the
FMOLS estimates at the 1% significance level.
43Actual estimates of the relative size of freight rates to containerized cargo values range from 0.08% for
mid range clothing to 21.5% for assembled furniture according to Rodrigue et al. (2013).
57
TABLE 5. Cointegration Relations - Group-mean Panel Dynamic OLS Estimator
(1) (2) (3) (4) (5)
Demand FH Pricing BH Pricing
Variables qij qij ffh fbh fbh
Freight Rate (fij) -0.044 -0.044 - - -
(-1.350) (-1.350)
Sales Price (pi) -0.597** -0.597** - - -
(-2.374) (-2.374)
Exporter GDP (yi) 0.927*** 0.927 - - -
(3.544) (-0.280)
Importer GDP (yj) 1.189*** 1.189 - - -
(4.874) (0.775)
Shipping Capacity (sc) - - -0.733*** -0.425*** -0.425***
(-4.667) (-4.405) (3.201)
Bunker Fuel Price (bfp) - - 0.449*** 0.121*** 0.121***
(5.543) (2.657) (-7.154)
Trade Imbalance (imb) - - 0.352** -0.307*** -0.307***
(2.322) (-3.515) (-7.550)
Lags and Leads 1 1 1 1 1
Panels 6 6 3 3 3
Observations 320 320 160 160 160
Coefficient Hypothesis [0, 0, 0, 0] [0, 0, 1, 1] [0, 0, 0] [0, 0, 0] [-0.73, 0.45,
-0.35]
Notes: The empirical results were obtained using Pedroni (2001) group-mean panel DOLS estimator. T-statistics are
given in parentheses. Statistical significance of coefficients at the 1%, 5%, or 10% level is indicated via ***, **, or *,
respectively.
Considering the cointegration relations underlying the supply side of the
international transport market in more detail, results are differentiated between fronthaul
and backhaul markets. In Tables 5 and 6, the fronthaul and backhaul pricing relations are
given by column (3) (4), respectively. In particular, the estimates of the fronthaul pricing
relation reveal that, in the long-run, a 1% permanent increase in market shipping capacity
leads to a persistent decline of the fronthaul freight rate ranging from 0.509% (FMOLS)
to 0.733% (DOLS). Both of these estimates are statistically significant at the 1% level. In
58
comparison to these fronthaul estimates, the coefficient estimates of the backhaul pricing
relation, which are also statistically significant at the 1% level, point to a smaller response
of international freight rates charged in backhaul transport markets. While the DOLS
estimator predicts only a 0.425% long-run decline in the backhaul freight rate in response
to a persistent 1% increase in market shipping capacity, the FMOLS estimator suggests
only a 0.342% decline of backhaul freight rates in the long-run.
Observing these differences in magnitude across coefficient estimates, the estimation
of the backhaul pricing relation testing the hypotheses of equality across fronthaul and
backhaul coefficient estimates is repeated. The results are given in column (5) of Tables
5 and 6 and reveal that the differences in fronthaul and backhaul coefficient estimates
are, indeed, statistically significant for all variables. Concerning the coefficient estimates
on market shipping capacity, this finding implies that an increase in competition among
carriers has a statistically significantly smaller effect on the freight rates charged in
backhaul compared to fronthaul transport markets. This finding is intuitive. Since
backhaul markets are by definition subject to excess capacity, an increase in market
shipping capacity should have a smaller effect on transport rates in backhaul markets than
in fronthaul markets where capacity allocation is binding and carriers can exercise larger
market power in setting fronthaul rates.
The coefficient estimates on bunker fuel prices suggest that a permanent increase in
access costs raises both fronthaul and backhaul freight rates in the long-run. Specifically,
it is estimated that the long-run positive effects of a persistent 1% rise in bunker fuel
prices on fronthaul freight rates vary between 0.449% (DOLS) and 0.296% (FMOLS), both
of which are statistically significant at the 1% level. In contrast, the DOLS (FMOLS)
estimator predicts that a 1% permanent increase in bunker fuel prices in backhaul
transport markets leads to only a 0.121% (0.112%) permanent increase in backhaul freight
59
rates. Again, column (5) of Tables 5 and 6 illustrates that these differences in coefficient
estimates are statistically different at the 1% significance level.
TABLE 6. Cointegration Relations - Group-mean Panel Fully Modified OLS Estimator
(1) (2) (3) (4) (5)
Demand FH Pricing BH Pricing
Variables qij qij ffh fbh fbh
Freight Rate (fij) -0.058** -0.058** - - -
(-2.395) (-2.395)
Sales Price (pi) -0.340** -0.340** - - -
(-1.963) (-1.963)
Exporter GDP (yi) 0.632*** 0.632** - - -
(3.397) (-1.975)
Importer GDP (yj) 1.326*** 1.326* - - -
(6.954) (1.710)
Shipping Capacity (sc) - - -0.509*** -0.342*** -0.342**
(-4.413) (4.369) (2.120)
Bunker Fuel Prices (bfp) - - 0.296*** 0.112*** 0.112***
(5.841) (3.494) (-5.749)
Trade Imbalance (imb) - - 0.140 -0.446*** -0.446***
(1.404) (-7.101) (-9.324)
Panels 6 6 3 3 3
Observations 320 320 160 160 160
Coefficient Hypothesis [0, 0, 0, 0] [0, 0, 1, 1] [0, 0, 0] [0, 0, 0] [-0.51, 0.30,
-0.14]
Notes: The empirical results were obtained using Pedroni (2000) group-mean panel FMOLS estimator. T-statistics are
given in parentheses. Statistical significance of coefficients at the 1%, 5%, or 10% level is indicated via ***, **, or *,
respectively.
As previously mentioned, in order to capture simultaneity of trade and
transportation and the role that marginal joint costs play in the determination of
fronthaul and backhaul unit-specific trade costs, the estimation of the fronthaul and
backhaul pricing relations includes the trade imbalance. While the DOLS coefficient
estimate on the imbalance term for the fronthaul pricing relation is statistically significant
at the 5% level, the FMOLS estimate is not statistically different from zero. Inference
60
on the statistically significant DOLS estimate suggests that a persistent 1% increase
in the trade imbalance leads to a permanent 0.352% rise in fronthaul transport costs.
This finding reflects the fact that increases in the trade imbalance correspond to a
reallocation of marginal joint costs towards the fronthaul market. In contrast to this
finding, estimations of the pricing relation in backhaul markets demonstrate that a
permanent 1% increase in the trade imbalance leads to a reduction of the long-run
equilibrium backhaul freight rate that ranges from 0.307% (DOLS) to 0.446% (FOMLS)
depending on the estimator. Again, this finding can be explained by the reallocation
effect. As the trade imbalance grows, the allocation of marginal joint costs predominately
falls onto fronthaul rather than backhaul markets leading to a reduction of the backhaul
freight rate.
Overall, the estimation of the demand and pricing relations establishes the
simultaneity between trade and transportation costs that is often ignored in models of
trade. Furthermore, the estimated pricing relations, which reflect the long-run equilibrium
relationships between trade costs, the trade imbalance, the transport market structure and
shipping cost factors, highlight the potential for maritime transit policy to reduce trade
cost and stimulate the growth of international trade. However, the differences in coefficient
estimates between fronthaul and backhaul markets also suggest that such policies may
have varying effects on trade facilitated in different transport markets. One example of
such a policy might be the StrongPorts initiative by the Maritime Administration of the
Department of Transportation. Based on the results, the long-term effects of a policy, such
as the StrongPorts initiative, as well as shocks to market structure or shipping cost factors
on containerized international trade can be simulated.
61
Simulation
Taking into account the simultaneity of trade and transportation costs introduced by
the transport sector, Table 7 gives the simulated long-term structural responses of trade
and freight rates facilitated in a fronthaul and backhaul transport market pair to a variety
of persistent supply and demand shocks. The results are presented for simulations based
on the DOLS estimates only.
First, I consider the long-run effects of permanent shocks to the supply side of the
international transport industry, such as persistent changes to market shipping capacity
or bunker fuel prices which may result directly from maritime transit policy. Specifically,
the simulations show that a permanent 10% increase in fronthaul and backhaul market
shipping capacity, for example, leads to a persistent 7.284% reduction of freight rates and
0.321% increase of trade in fronthaul transport markets. In contrast, backhaul transport
markets exhibit a 4.290% permanent reduction of freight rates and 0.189% persistent
increase in trade in response to this 10% permanent rise in market shipping capacity.
TABLE 7. Simulation Results
Variables ∆sc (10%) ∆bfp (-10%) ∆yi (1%) ∆pi (10%)
FH Trade (qij) 0.321% 0.195% 0.931% -5.874%
BH Trade (qji) 0.189% 0.055% 1.187% -0.083%
FH Freight Rate (fij) -7.284% -4.441% -0.089% -2.172%
BH Freight Rate (fji) -4.290% -1.253% 0.077% 1.894%
Notes: The simulation results are based on the DOLS cointegration relation coefficient point estimates.
Next, I consider the effects of a persistent 10% decline in bunker fuel prices in
fronthaul and backhaul transport markets. Table 7 reveals that fronthaul freight rates
permanently decrease by 4.441%, while backhaul freight rates permanently decline by only
1.253%. The reductions in unit-specific trade costs coincide with a 0.195% and 0.055%
62
long-run rise in fronthaul and backhaul trade, respectively. Although the estimated long-
run effects on trade are rather small, these simulations point to the potential of maritime
transit policy to stimulate the growth of international trade.
Lastly, I explore the effects of permanent shocks to the demand side of international
transport markets. Specifically, I find that a 1% increase in country i’s income leads to a
0.931% and 1.187% long-run equilibrium increase of fronthaul and backhaul trade. Since
backhaul trade increases by more than fronthaul trade, the trade imbalance shrinks. Due
to the reallocation of marginal joint costs, this permanent reduction of the trade imbalance
causes fronthaul freight rates to permanently decline by 0.089% and backhaul freight rates
to simultaneously increase by 0.077%. This, of course, implies that the long-term response
of fronthaul trade is enhanced by the endogenously adjusting fronthaul freight rate, while
the permanent response of backhaul trade is dampened by the corresponding change in the
backhaul freight rate.
Another simulation that highlights not only this simultaneity between trade and
transportation, but also the potential spillover effects deriving from the joint production
present in the international liner shipping industry, concerns the long-run effects of a
10% increase in country i’s sales price. The naive model of trade, ignoring endogenously
adjusting freight rates, predicts that a change in country i’s sales price should only affect
country i’s exports. Expanding the model to account for the simultaneity between trade
and transport suggests that the freight rate charged on country i’s exports should also
adjust to this change in trade, while transport from country j to i remains unaffected.
However, as illustrated in Table 7, the change in country i’s sales price not only causes an
adjustment in country i’s exports and trade costs, but also leads to a persistent response
of trade and freight rates in the backhaul transport market facilitating trade from country
j to country i.
63
In fact, while fronthaul trade from country i to j permanently decreases by 5.874%,
backhaul trade from country j to i also decreases by 0.083%. This feature is explained
through the effects of joint production present in the liner shipping industry, where
carriers adjust both fronthaul and backhaul supplies in response to a change in fronthaul
demand. Here, the permanent decrease in trade facilitated in the fronthaul transport
market leads to a persistent reduction of the trade imbalance. This reduction of the
trade imbalance causes a 2.172% long-run equilibrium decline of the fronthaul freight
rate and drives carriers to reallocate marginal joint costs away from the fronthaul and
towards the backhaul transport market. This reallocation of marginal joint cost, of
course, simultaneously triggers the 1.894% persistent increases in the backhaul freight
rates, which in turn causes a 0.083% permanent decline in backhaul trade, a previously
unanticipated and unexplainable spillover effect. This simulation strongly supports the
theoretical predictions laid out via the previously discussed comparative statics exercise.
In conclusion, when accounting for the simultaneity of trade and transport costs as well
as the joint production present in the international liner shipping industry, a permanent
shock to one country’s sales prices not only influences its exports, but imports as well.
Conclusion
Naturally, international trade critically depends on transport markets. The majority
of previous models either ignore this dependence or fail to fully capture the unique market
defining characteristics present in the international transportation industry. In this study,
I carefully integrate trade and transportation. Furthermore, I derive the transport sector
recognizing that the international container shipping markets are subject to the key
feature of joint production. Given the model, I evaluate the structural relationships by
obtaining estimates of cointegration relations present in the data. The results demonstrate
64
the existence of long-run equilibrium relations that govern the simultaneous determination
of trade and unit-specific trade costs measured by container freight rates and explain
spillover effects across bilateral export and import markets.
I find that unit-specific trade costs are an integral part to the long-term
determination of trade. More importantly, the existence of the long-run relationships
between trade and trade costs on the demand as well as supply side of the international
transportation markets demonstrate the endogeneity of trade costs. Moreover, the
structural relations between trade costs, market structure and access cost factors on
the supply side of transport create the opportunity for maritime transit policy to have
real impacts on trade costs and thus, facilitate further growth of international trade.
Specifically, simulations show that a permanent 10% increase in market shipping capacity
leads to a 7.284% permanent decline of unit specific trade costs in fronthaul transport
markets and 4.290% decline of unit specific trade costs in backhaul transport markets
which coincide with a permanent 0.321% and 0.189% increase in trade facilitated in
fronthaul and backhaul transport markets, respectively.
Based on these findings, there are various research questions that are of potential
interest. Future studies might examine the nature of the long-run equilibrium relations
between trade and trade cost established in this study at a more disaggregated level.
Of particular interest could be whether these relations differ between countries with
varying trade compositions and levels of development. Of course, any such study hinges
on the development of disaggregated data that reflect trade flows and trade costs at the
product level. Alternatively, an interesting avenue for future research is to focus on the
varying response between front- and backhaul trade given various trade-cost-reducing
policy measures. Further inquiry should delineate between the policy impacts on exports
65
and imports facilitated in fronthaul and backhaul transport markets and deduce policy
implications stemming from these varying responses.
Bridge
Having provided theoretical and empirical evidence of the simultaneity of trade and
transportation, I further explore its consequences in the following chapter. In particular,
the endogeneity and asymmetry of transport costs in the presence of imbalanced bilateral
trade raises the possibility of systematic variation concerning the response of these freight
rates to trade and maritime transport policy. Extending the previously derived theoretical
model, I investigate the elasticities of trade with respect to carrier costs, in the following
chapter. An empirical examination of these elasticities provides supporting evidence for
the theoretical model developed in the current chapter.
66
CHAPTER III
TRADE POLICY OUTCOMES IN THE PRESENCE OF ASYMMETRIC TRANSPORT
COSTS:THEORY AND EVIDENCE
Introduction
Barriers to trade have been a central focus of the international trade literature
for decades. Many of these trade impediments, such as tariffs and border related costs
or cultural differences, have been analyzed by a multitude of studies.1 The insights
gained from historical policy changes and academic research have led to a dramatic
global reduction in tariff rates and a multitude of preferential trade agreements. As the
global reductions of tariffs approach a lower bound, the significance of alternative trade
impediments, such as transport costs, grows (Hummels, 2007).
In recognition of these compositional changes in trade costs, the World Trade
Organization (WTO) has initiated the Trade Facilitation Agreement (TFA) in 2013. This
agreement aims to expedite the movement and clearance of internationally traded goods
and is currently in the ratification process.2 According to the WTO’s World Trade Report
2015, the agreement is expected to decrease total trade costs by 14.5%, on average, and
increase global merchandise exports by up to $1 trillion per year. In light of this and other
transport related trade policies, it becomes of central interest to consider the international
transport sector and the role it plays in the determination of trade policy effectiveness.
While there exists widely accepted knowledge and stylized facts concerning the effects of
1For a comprehensive survey on this literature see, for example, Anderson and van Wincoop (2004).
2Once two thirds of the WTO members have ratified TFA, the agreement will go into affect. As of May
25th, 2016, 79 out of 162 WTO members have ratified TFA.
67
tariff-reducing trade policy, little is known about the efficacy of maritime transport policy
and the potential heterogeneity of its outcomes across bilateral trade relations.
In this study, I analyze the effects that transportation-related commercial and
environmental policy has on international trade and find that the effectiveness of such
policy, indeed, systematically varies across bilateral trade flows. Careful consideration of
the international transport sector points to the backhaul problem3, or in other words, a
potential underutilization of available shipping capacity due to bilateral trade imbalances,
as the source of this variation. More specifically, when carriers operate on strictly
scheduled round trips, allocating fixed transport capacity to facilitate bilateral trade, the
costs of providing this fixed capacity are inseparable between the individual transport
markets and lead to the joint production of bilateral transport services. This joint
production of transport services, in turn, prompts the integration of the equilibrium
freight rates charged to facilitate bilateral trade. When the demands for transport are
imbalanced, round trip cost allocation varies across the resulting fronthaul and backhaul
transport markets4 and causes asymmetric bilateral freight rates that respond differently
to a given policy-induced change in carrier costs. In fact, comparative statics derived from
the proposed theoretical model reveal that the asymmetric response of transport costs
triggers heterogeneous bilateral trade effects that vary systematically across fronthaul and
backhaul transport markets, at different levels of the bilateral trade imbalance and by
product groups differentiated by their respective ad valorem transport costs.
3A significant share of internationally traded goods is facilitated by liner carriers that operate on
strictly scheduled round trips. Naturally, the round trip production process generates at least two
transport markets that can be served. While liner carriers provide a fixed shipping capacity to all
markets of the round trip, the demands for transportation given by bilateral trade may be imbalanced.
The joint allocation of fixed capacity in the presence of imbalanced demands for transport can result in
underutilization of capacity in a given transport market, labeled as the backhaul problem.
4Following common terminology the leg of the round trip facing higher demand is denoted as the
fronthaul, while its counterpart, the leg of the round trip facing lower demand, is denoted as the backhaul.
68
To establish these theoretical predictions, I derive a model of transportation demand
from a standard trade framework as in Hummels et al. (2009) and extend it to incorporate
the supply-side of the transport sector, paying direct attention to the fact that round
trip transport services are provided between pairs of countries under conditions of
joint production. The incorporation of such a transport sector in the model effectively
endogenizes the unit-specific trade costs captured by international freight rates. This
endogeneity, along with the specific round trip structure of transport markets allows
for the integration and asymmetry of bilateral transport costs. I derive the theoretical
elasticities of trade with respect to various carrier costs and show that while these
elasticities vary across fronthaul and backhaul transport markets when bilateral trade is
imbalanced, the difference in elasticities vanishes as bilateral trade becomes balanced. In
addition, these elasticities reveal that imbalanced backhaul trade of products with high
ad valorem transport costs is more responsive to a change in carrier cost than otherwise
identical trade of products with low ad valorem transport costs. Intuitively, these changes
in trade elasticities can be traced back to the asymmetric carrier cost allocation across the
two round trip markets which alters the relative share of transport costs in overall trade
costs making trade more or less responsive.
The empirical evaluation of these theoretical predictions rests on a difference-in-
differences approach commonly used to estimate the treatment effects of exogenous policy
changes. In specifying the estimation model, I incorporate the traditional gravity equation
framework. Identification of the theoretically derived heterogeneous trade elasticities is
achieved through the estimation of the negative U.S.-EU trade externalities imposed
by the EU low sulfur fuel requirement of 2010 that was enacted as part of EU Council
Directive 2005/33/EC. According to this Directive, as of January 1st, 2010, liner carriers
are forced to switch from low cost heavy fuel oils to high cost low sulfur fuels while at
69
berth at any EU port to reduce air pollution from shipping. Anecdotal evidence suggests
that this requirement imposed a 70% to 100% premium on in-port fuel costs, an estimated
aggregate annual fuel cost increase of $1.3 billion (Ivanov, 2010) and a significant hike in
international freight rates (Notteboom et al., 2010). The estimation of the potentially
varying effects of this exogenous policy-induced increase in carrier costs proceeds by
integrating the standard difference-in-differences technique into the Poisson Pseudo-
Maximum Likelihood estimator developed by Santos Silva and Tenreyro (2006). The
identification strategy exploits the exogenous rise in trade costs for U.S.-EU bilateral trade
flows relative to all other U.S. trade and differentiates the average treatment effects along
various dimension, including U.S. containerized exports and imports, U.S. fronthaul and
backhaul transport markets, across balanced and imbalanced bilateral trade flows and
across U.S. trade of disaggregated product groups with high or low ad valorem transport
costs.
The empirical analysis provides robust results that are consistent with the theoretical
predictions. In particular, I find that the low sulfur fuel requirement, on average, causes
a statistically-significant 9.9% reduction in U.S.-EU containerized backhaul trade,
while U.S.-EU trade facilitated in fronthaul transport markets exhibits no statistically
significant response.5 Based on the patterns of U.S. trade, these findings translate into a
statistically significant 8.0% reduction of U.S. exports to EU countries and a statistically
insignificant effect on U.S. imports. Further differentiation between rather balanced
and imbalanced bilateral U.S. trade flows provides strong evidence that the differences
in policy outcomes decreases as trade becomes more balanced. Lastly, the empirical
results show a statistically significant difference in treatment effects across products
5For the purposes of the empirical analysis, a transport market represents a fronthaul for a given
month when the current value of bilateral trade facilitated in this market exceeds the current value of
bilateral trade facilitated in the opposing direction.
70
with high and low ad valorem transport costs and point to additional heterogeneity
concerning policy effectiveness at the disaggregated product level. In particular, I find
that trade in products with high ad valorem transport costs declines by 17.9% and 7.7%
in backhaul and fronthaul transport markets, respectively. In contrast, trade in low ad
valorem transport cost products exhibits small and statistically insignificant reductions in
fronthaul and backhaul transport markets in response to the low sulfur fuel requirement.
Given the differences in current trade imbalances across low to high income countries and
geographical regions, the derived implications are quite significant. In conjunction with
aggregated World Bank data, my findings suggest that trade from low income countries,
particularly in South Asia, will experience considerably disproportionate effects from
transport-related global policies.
The present study contributes to the literature in several ways. First, the theoretical
model provides an extension to a standard trade framework by integrating a transport
sector that accounts for the backhaul problem.6 Comparative statics illustrate that the
effects of trade policy vary across international transport markets and that this variation is
due to the reflection of the backhaul problem in the carrier cost structure. Furthermore,
the empirical analysis offers novel results that exhibit the heterogeneous responses of
international trade to an exogenous shock in carrier costs. Hence, the empirical findings
provide evidence in support of the theoretical predictions that maritime transport policy
leads to trade outcomes that vary across international transport markets, balanced and
imbalanced bilateral trade flows and at the product level leading to otherwise unexplained
asymmetric effects across U.S. containerized exports and imports. Overall, the empirical
results presented in this study are the first to provide quantitative evidence concerning
6The theoretical model complements derivations by Behrens and Picard (2011) or Ishikawa and Tarui
(2015), for example.
71
the importance of the consideration of the backhaul problem in the formulation of trade
policy.
The remainder of this paper is organized as follows: In section 2, I describe the
institutional background of the liner shipping industry. This background section is
complemented by the literature review of trade, transportation and the integration of the
two subject areas presented in section 3. In section 4, I develop the theoretical model.
The empirical model builds on these foundations and is presented in section 5. Section
6 summarizes the data employed, while section 7 gives the empirical results. Section 8
provides a summary as well as conclusion and points to potential areas of further inquiry.
Institutional Background
Prior to the development of a theoretical model that integrates trade and
transportation via the incorporation of a carefully drafted international transport sector, it
is imperative to gain a basic understanding of its workings. Since the theoretical model
and empirical analysis are centered on containerized trade, this section provides some
of the most relevant information pertaining to the liner shipping industry7, its historical
development and significance to international trade.
In 2006, container shipping had its 50 year anniversary. During the previous 50
years, the industry that was sparked by Malcolm McLean’s historic innovation of a
vessel carrying 58 trailer-truck bodies from Newark, New Jersey to Houston, Texas has
rapidly grown and revolutionized international trade. As the demand for international
transport of goods has risen dramatically in recent decades, world shipping capacity8,
7According to the World Shipping Council, liner shipping encompasses all modes of high capacity
transport services. Moreover, it is pointed out that liner vessels are primarily constituted of containerships
justifying representation of liner shipping via the container shipping industry.
8By 2010, the global container shipping fleet was comprised of 4,677 vessels with a maximum capacity
of 12.8 million TEU’s (UNCTAD, Secretariat, 2014).
72
as well as the capacity of individual vessels has grown to match this demand.9 Based
on predictions by the International Transport Forum (ITF) at the OECD (2015), this
rapid growth of containerized trade is expected to continue and lead to a 400% increase
in international freight transport by 2050. However, the ITF also predicts that this growth
will be unevenly distributed across North Pacific and North Atlantic transport markets
shifting the global patterns of trade and increasing transport distances by 12% by 2050
as well. These anticipated changes in the volume and patterns of international trade will
continue to put tremendous pressure on international liner shipping and motivate research,
like the present study, pertaining to this industry central to global economic development.
Just as the international transport sector has had to adjust to the ever increasing
demands from trade, its regulation had to adapt as well. Recent deregulation via the
Shipping Act of 1984 and Ocean Shipping Reform Act (OSRA) of 1998 has led to the
restructuring of the global liner shipping industry.10 Historically, container freight rates
on various trade routes have been set by the associated conferences of carriers serving
the respective trade route. But in the aftermath of OSRA, these conferences have lost
some of their control over the separate markets of this industry. In 1999, for example,
the Asia North America Eastbound Rate Agreement (ANERA) and the Japan U.S.
Eastbound Freight Conference (JUEFC), which previously controlled freight rates charged
on transpacific trade routes, were suspended. In 2008, the EU repeal of the competition
law block exemption, previously granted to the liner shipping conferences in EU trades,
led to the termination of the Transatlantic Conference Agreement (TACA). Based on
these recent regulations, it is clear that the container shipping industry has lost some of its
9While it was not uncommon for vessels to carry over 9,000 twenty-foot equivalent units (TEUs) in
2006 (Transportation Research Board, 2006), this capacity has been more than doubled by the largest
container vessel, the triple E-class, capable of transporting over 18,270 TEUs per ship (Port Finance
International, 2013).
10Fusillo (2013) finds that post-OSRA market shares have been much less stable.
73
market power. However, container carriers continue to seek global alliances that influence
the level of market power in international shipping and determination of international
freight rates.
Recent data concerning the present state of the international shipping industry
reveal that maritime transport accounts for the facilitation of over 70% of the total
value of international trade (UNCTAD, Secretariat, 2014). Within this overarching
transportation industry, liner shipping represents one of the most important modes of
transportation. In fact, according to the World Shipping Council, liner vessels transport
about 60% of the total value of all seaborne trade. Across countries, the U.S. was the
world’s largest importer and second largest exporter of containerized cargo, as of 2010.
However, these data also show that this industry is subject to the backhaul problem
stemming from the presence of the joint round trip production by container carriers. Table
8 illustrates the magnitude of this issue pertaining to the container shipping industry. The
depicted average freight rates and container flows, measured in the number of Twenty-Foot
Equivalent Unit (TEU) containers, facilitated on the major trading routes between Asia,
the EU and the U.S., point to large trade imbalances that coincide with sizable average
freight rate differentials in these markets.
Figures 2a-2f in Chapter II further suggest that trade imbalances and freight rate
differentials are highly correlated and persistent over time. The transpacific market,
depicted in Figure 2b, shows large and growing trade imbalances between U.S. imports
from Asia and U.S. containerized exports to this region that correspond to growing
freight rate differentials in this market, as illustrated by Figure 2a. In contrast, the
transatlantic market between the EU and U.S., illustrated by Figure 2f, reveals switching
trade imbalances between the two regions that largely coincide with the switching freight
rate differentials depicted by Figure 2e.
74
TABLE 8. Avg. Containerized Trade and Freight Rate Imbalances
Fronthaul Backhaul
Markets Mean Sd Mean Sd
Regional Cargo Flow (million TEUs)
Transpacific Market 2,412 (942.4) 1,169 (352.2)
Asia-EU Market 1,556 (587.4) 843.5 (193.2)
Transatlantic Market 519.1 (79.51) 432.5 (90.52)
Regional Freight Rates ($ per TEU)
Transpacific Market 1,717 (226.3) 930.8 (217.0)
Asia-EU Market 1,491 (274.8) 853.1 (167.0)
Transatlantic Market 1,385 (224.0) 1,146 (281.8)
Sources: Containerized Cargo Flow data - Drewry and Freight Rate data - Containerisation International
Overall, it is clear that the liner shipping industry plays an integral role in the
facilitation of international trade and U.S. trade, in particular. Aggregated data provide
supporting evidence that much of this containerized international trade is subject to
the backhaul problem. To the extent that freight rates matter to the determination
of international trade, the correlation of freight rate differentials and trade imbalances
and the resulting integration of bilateral trade costs suggests that the backhaul problem
plays an important role in the determination of international bilateral trade as well. The
asymmetry of these unit-specific trade costs and their dependence on trade imbalances
allows for the possibility of varying responses of international trade to a given change in
commercial or maritime policy reforming the liner shipping industry and motivates the
theoretical and empirical analyses conducted in this study.
Literature Review
The following section provides a basic review of the trade literature paying particular
attention to studies that focus on trade costs. Within the area of trade costs, I provide
a detailed summary of those studies concentrated on the international transport sector
75
and its integration with international trade. In conclusion of this review, I illustrate the
remaining gaps in the literature and point to the contributions this study offers to the
field.
International Trade & Trade Costs
The driving forces underlying international trade, its welfare implications, as well
as the development and analysis of policy instruments that may stimulate or hinder
international trade have been a central focus of the economic literature for decades.
As a result a plethora of models of international trade have been developed. While the
assumptions underlying these models vary greatly11, a common point of emphasis is the
role of trade costs. In fact, regardless of the specific model of international trade, trade
costs consistently manifest themselves as one of its integral determinants.12
While early studies have simply used geographic distance as a proxy for trade costs,
many additional determinants, such as free trade agreements or colonial and cultural
ties, have been identified in the more recent literature.13 Anderson and van Wincoop
(2004) survey this literature and establish that trade costs can be divided into three main
categories which include border related costs, local distribution costs and transportation
costs.14 Specifically, the authors show that about 21% of all trade costs are attributable
11Some of the more prominent theoretical models are based on assumptions, such as absolute cost
advantages (Smith, 1776), Ricardian comparative cost advantages (Dornbusch et al. (1977), Eaton
and Kortum (2002) , Bernard et al. (2003)), varying factor endowments (Heckscher and Ohlin, 1991),
differences in incomes and trade costs (Samuelson, 1952), economies of scale and a taste for variety
(Krugman, 1979), and varying productive efficiencies (Melitz, 2003), among others.
12Moreover, Obstfeld and Rogoff (2001) claim that trade costs are also a critical key to solving the
remaining macroeconomic puzzles identified in their study.
13A brief summary of these factors is presented by Head and Mayer (2013)
14Hummels (2001) points out that explicit trade costs such as tariffs and freight rates are more
significant contributors to trade costs than implicit determinants such as common language and colonial
linkages, while Hummels (2007) illustrates that in 2004, for example, transportation costs on U.S. imports
far outweighed the costs imposed by tariffs.
76
to transportation costs. While the authors argue that is a rough estimate, there is some
debate over the true effects of transport costs and trade liberalization on trade flows and
growth. Baier and Bergstrand (2001), for example, find that the reduction of transport
costs accounts for only 8% of total growth in global trade, compared to 33% due to trade
liberalization. In contrast, a more recent study by Bernhofen et al. (2016) finds that the
reduction in transport costs due to containerization has had a much more significant
impact on trade growth than trade liberalization efforts. In particular, the authors find
that the cumulative average treatment effect (ATE) of containerization on ’North-North’
trade 15 years after treatment is 1240%, compared to free trade agreements and GATT
which have a cumulative ATE of 68% and 194%, respectively.
International Trade & Transport Costs
Despite the importance of transport costs in the determination of the overall barriers
to trade, the subject has received surprisingly little attention historically. Recent research
studies, however, have underlined the significance of the international transport sector
to the determination of trade and have given the subject more careful consideration.15
Hummels and Skiba (2004), for example, argue that transport costs are more accurately
modeled as unit specific, rather than ad valorem trade costs, as introduced by Samuelson
(1954). Contrary to this intuitive argument, the ad valorem specification has been adopted
by the majority of the literature16, and, despite its shortcomings17, distance has been
used as the main variable to capture these trade costs. Few studies have focused on the
15Behar and Venables (2011) offer a concise summary of the recent literature revolving around trade and
transportation.
16One exception is the study by Feenstra and Romalis (2014) who adopt a hybrid specification of trade
costs that includes both the traditional ad valorem and unit-specific component.
17Studies by Limao and Venables (2001), Combes and Lafourcade (2005), and Martinez-Zarzoso and
Nowak-Lehmann (2007), for example, have uncovered a variety of issues with the geographic distance
proxy and concluded that it is an overall poor instrument for transport costs.
77
direct integration of trade and transportation. Exceptions include studies by Behrens et al.
(2006), Behrens and Picard (2011), Takahashi (2011), Kleinert and Spies (2011), Friedt
and Wilson (2015), or Ishikawa and Tarui (2015) who have developed theoretical models
that incorporate a transport sector into a variety of trade frameworks to account for the
simultaneity between trade and transportation and analyze the effects of endogenous
transport costs on trade.
Empirically, there are many studies that have analyzed a variety of facets concerning
the relationship between trade and transportation. Friedt and Wilson (2015), for example,
provide empirical evidence of the simultaneity between trade and transportation by
estimating panel co-integration relations that govern the long-run structural demand
and supply equations of the international transport sector. Other studies explore the
dependence of international transportation costs on the volume of trade (Mart´ınez-Zarzoso
and Sua´rez-Burguet, 2005; Martinez-Zarzoso and Nowak-Lehmann, 2007), or investigate
its dependence on bilateral trade imbalances (Demirel et al., 2010; Jonkeren et al., 2011).
Alternatively, some studies focus on the effect of transport costs on trade (Mart´ınez-
Zarzoso et al., 2003; Mart´ınez-Zarzoso and Sua´rez-Burguet, 2005; Martinez-Zarzoso and
Nowak-Lehmann, 2007). While specific coefficient estimates vary18, the general finding
of this literature is that transport costs are a significant deterrent to trade. In addition
to these findings, some studies have provided reduced form estimations that show that
changes in transport costs have varying effects on trade across different product categories
(Mart´ınez-Zarzoso et al., 2008), the extensive and intensive margins of trade, and across
exports and imports (Bensassi et al., 2014).
18Possible reasons for the varying elasticity estimates include short-run versus long-run considerations
(Egger, 2002), differences in the specific trade flows examined in the individual studies, or differences
in the specific measures of transport costs. The measures used in the literature range from aggregate
c.i.f./f.o.b. ratios to micro level survey data. As Hummels and Lugovskyy (2006) have shown, these
differences may cause variability in the elasticity estimations.
78
However, the reduced form empirical models underlying these findings of the varying
effects of transport costs on trade provide no theoretical explanation for their causes.
That is, the influence of the transport market structure and its unique characteristics are
unexplored in the reduced form specifications. Only a very small subset of the literature
has considered the specific characteristics that are unique to the international transport
sector and theoretically analyzed their impact on the distribution of economic activity and
international trade, as well as their importance to trade policy (e.g. Hummels et al., 2009;
Behrens and Picard, 2011; Friedt and Wilson, 2015; Ishikawa and Tarui, 2015). Perhaps
the most important one of these characteristics, as pointed out in the previous section,
is the backhaul problem experienced by the liner shipping industry. As an abundant
literature in the field of transport economics has pointed out, the backhaul problem is
a significant issue for various transportation industries and affects not only transport
pricing, market service and market access, but also regulation outcomes (see, for example,
Nicholson, 1958; Basemann and Daugherty, 1977; Rietveld and Roson, 2002; Wilson, 1994;
Wilson and Beilock, 1994).
Given these findings in the transportation economics literature, the presence of
the backhaul problem in the international liner shipping industry suggests that it may
alter the effectiveness of maritime transport policy as well. The potential of such policies
has been documented by several studies, including Bougheas et al. (1999), Clark et al.
(2004), or Blonigen and Wilson (2008), which have considered the effects of infrastructure
investments and port efficiency gains on international trade. In general, these studies find
that increases in either infrastructure or port efficiency lowers transport costs and leads to
increases in international trade. However, none of these studies considers the issues and
effects arising from the backhaul problem present in the international transport sector that
may lead to a variation in policy outcomes across trade flows.
79
In summary, the international trade literature has established the importance of
transport costs to the determination of international trade and points to the international
transport sector as a potential commercial policy instrument. However, up to this point
this literature has failed to empirically consider the issue of the backhaul problem and
its potential influence on commercial policy outcomes. Given the current state of the
literature, this study contributes in several ways. First, I provide a theoretical extension
to a standard trade framework by integrating a transport sector that accounts for the
backhaul problem. Comparative statics based on this model illustrate that the effects of
trade policy vary across international transport markets and that this variation is due to
the reflection of the backhaul problem in the carrier cost structure. Second, this study
offers an empirical analysis that estimates the varying responses of international trade
to an exogenous shock pertaining to these carrier costs. The empirical findings provide
supporting evidence for the theoretical results and show that trade policy outcomes vary
across international transport markets leading to otherwise unexplained asymmetric
effects exhibited by U.S. exports and imports. That is, the empirical results presented
in this study are the first to provide quantitative evidence concerning the importance of
the consideration of the backhaul problem in the formulation of commercial policy and
highlight the suggested implications for the global patterns and composition of trade.
Theory
The theoretical model developed in this section integrates trade and transportation.
The solution to this model is given by an equilibrium in both of the transportation
markets that facilitate bilateral trade to and from each country under conditions of the
backhaul problem. The purpose of the model is to derive the response of trade to a
shock in transportation costs, while accounting for the simultaneity between trade and
80
transportation as well as the integration of bilateral transport costs. More formally, the
primary focus of this section is to examine whether this response of trade is equivalent
for all bilateral trade flows or whether the theory suggests systematic variation across
fronthaul and backhaul transport markets. I find that maritime transport policy
stimulates trade in almost all cases, albeit the magnitude of this increase in trade varies
across fronthaul and backhaul transport markets and across balanced and imbalanced
bilateral trade flows.
To derive these theoretical predictions, I follow the model of trade developed by
Hummels et al. (2009) and provide an extension to the transport sector. In particular,
I integrate a model of the international liner shipping industry that accounts, unlike
Hummels et al. (2009), for the joint production by liner carriers that operate on round
trips between trading countries and thereby, offer transport capacities in two transport
markets that are linked by inseparable joint costs, such as expenses on crew, maintenance
and repairs, as well as port and cargo dues.19
Demand for Transport
To begin, I derive an expression for the demand of transport from the international
trade theory expressed by Hummels et al. (2009). In this model of trade, each country,
j=1,2,...,M, is composed of one representative consumer. Preferences of each representative
consumer take a quasi-linear form and are expressed over a homogeneous numeraire
commodity and a variety of a good that is differentiated by national origin, following
Armington (1969). The price elasticity of demand, σ, is assumed to be constant across
representative consumers and greater than one. Given these assumptions, the preferences
of the representative consumer in country j can be expressed by the following utility
19This model of the transport sector follows theoretical derivations by Wilson (1994) and Wilson and
Beilock (1994).
81
function
Uj = q0j +
M∑
i=1
q
(σ−1)/σ
ij , (3.1)
where country j’s consumption of the numeraire commodity is given by q0j and the
consumption of a particular variety sourced from country i is given by qij.
The price of the numeraire is normalized to one and it is assumed that this good can
be traded at no cost. In contrast, the sales price of a variety from country i is represented
by pi and taken as given by carriers. The import price, pij, of a variety from country i
paid by the representative consumer in country j includes per-unit transportation costs,
fij, and ad valorem trade costs, τij ≥ 1, in addition to the sales price, pi. That is,
pij = piτij + fij, where the transport and ad valorem trade costs are taken as given by
each representative consumer. Given these prices, the representative consumer’s budget
constraint can be formulated as follows
Yj = q0j +
M∑
i=1
pijqij, (3.2)
where national income of country j is given by Yj and the price of the local variety is
expressed as pjj = pj, since it is assumed that there are no intra-national trade costs.
Each representative consumer maximizes utility with respect to their budget
constraint. The solution to this constrained optimization problem gives the imported
quantities by country j from each country i. These imports, of course, also represent the
demand for transport from each country i to country j and are given by the following
expression;
qij =
[
σ
σ − 1(piτij + fij)
]−σ
. (3.3)
82
Supply of Transport
This expression for the demand of transport, given by equation (3.3), holds for any
two countries i and j engaged in bilateral trade and naturally creates the transport market
pair ij for each carrier. However, it is important to make note of the fact that given local
sales price differences, transport demands do not have to be equal to one another. In fact,
as previously noted, trade flows are rarely equal. Most often country i is a net exporter to
country j. This trade imbalance particularly holds for containerized cargo flows which are
facilitated on strictly scheduled round trips20 that prohibit any search and/or wait time
and implies that the demands for transportation in such a market pair are imbalanced.21
Following common terminology, the leg of the round trip facing higher demand is denoted
as the fronthaul, while its counterpart, the leg of the round trip facing lower demand, is
denoted as the backhaul.22
To facilitate bilateral trade between country i and country j, each carrier allocates
capacity, K, to transport market pair ij and offers transport supplies, Qij ≤ K and
Qji ≤ K, to each transport market, respectively. The provision of capacity to the
market pair results in available capacity in both transport markets. As such, the costs
of allocating capacity are inseparable joint costs leading to the joint production concerning
these transport supplies. In line with the current market structure of the liner shipping
industry, I follow the example by Hummels et al. (2009) and model the international
shipping industry to exhibit market power in each market pair ij. To accommodate
20Interviewing several industry insiders, including port officials and freight forwarders, it was pointed
out that container vessels, with the exception of extreme circumstances, adhere to strict schedules and
that carriers operate on round trips staggering the vessels they deploy, in order to offer more frequent
service.
21It is possible, of course, to encounter situations where overall trade may be balanced, while
containerized trade flows remain imbalanced due to the varying trade composition.
22Given this definition, fronthaul and backhaul depend on the trade imbalance between two trading
countries rather than the direction of trade flow or the starting point of a given round trip.
83
this feature of the industry, the transport sector is modeled as an oligopoly consisting of
l = 1, ..., N symmetric carriers competing in Cournot fashion. Extending the given model,
I assume that each carrier, l, facilitates a fraction, Qlij and Q
l
ji, of total bilateral trade,
qij and qji, between countries i and j and has a round trip cost structure that is twofold.
In particular, similar to Wilson (1994) and Wilson and Beilock (1994), each carrier faces
market specific access costs, aij, such as additional fuel and terminal costs, for shipping
one unit of a variety from country i to country j. In addition, each carrier’s technology
is further defined by the previously addressed joint costs, JC(K l), with JC(0) = 0 and
∂JC(Kl)
∂Kl
> 0, that are not differentiable between the individual transport markets and
depend on the shipping capacity, K l, that is allocated to the market pair served on a given
round trip. Intuitively, these costs can be viewed as, quite simply, the costs of operating
a vessel of capacity K l ≥ max(Qlij, Qlji) between two locations. Therefore, each carrier’s
round trip costs can be expressed as follows:
C l = aijQ
l
ij + ajiQ
l
ji + JC(K
l) ∀l = 1, ..., N and i, j = 1, ...,M, i 6= j. (3.4)
This cost structure, stemming from the joint production present in the liner shipping
industry, is a key factor in the derivation of varying commercial policy effectiveness that
is illustrated through some interesting comparative statics in the following subsection.
Given this cost structure, each carrier chooses the profit maximizing capacity, K l, and
optimal supplies of transport, Qlij and Q
l
ji, that are offered to each market on a given
round trip. Specifically, each carrier’s round trip profit from transporting bilateral trade
between country i and country j is comprised of revenues earned in each transport market
84
netting out the incurred access and joint costs and can be written as
max
Kl,Qlij ,Q
l
ji
Πl = fijQ
l
ij + fjiQ
l
ji − C l ∀l = 1, ..., N and i, j = 1, ...,M, i 6= j
subject to K l ≥ Qlij, K l ≥ Qlji.
(3.5)
Solving each carrier’s constrained profit maximization problem results in three Nx1
vectors of first-order conditions, along with the standard Kuhn-Tucker conditions, that
can be represented as follows;
∂Πl
∂Qlij
= fij +Q
l
ij
∂fij
∂Qlij
− aij − λ1 ≤ 0 with = if Qlij > 0 ∀l = 1, ..., N (3.6a)
∂Πl
∂Qlji
= fji +Q
l
ji
∂fji
∂Qlji
− aji − λ2 ≤ 0 with = if Qlji > 0 ∀l = 1, ..., N (3.6b)
−∂JC(K
l)
∂K l
+ λ1 + λ2 ≤ 0 with = if K l > 0 ∀l = 1, ..., N (3.6c)
K l ≥ Qlij, λ1 ≥ 0, (K l −Qlij)λ1 = 0 (3.6d)
K l ≥ Qlji, λ2 ≥ 0, (K l −Qlji)λ2 = 0. (3.6e)
The first-order conditions with respect to transport supplies, given by equations
(3.6a) and (3.6b), can be seen as each carrier’s market access conditions indicating that
marginal revenues in either transport market must at least cover access costs for a given
market to be served. In addition to that, each carrier’s first-order condition with respect
to the allocated capacity can be interpreted as the service condition. Given the fact that
the Kuhn-Tucker multipliers, λ1 and λ2, can be thought of as the shadow prices that
determine the value of an additional unit of transport supply in the respective transport
markets, equation (3.6c) states that market pair ij is served only if the marginal joint
costs of providing capacity, K l, do not exceed the cumulative value of an additional unit of
transport supply in either market.
85
In order to solve for the equilibrium transport supplies and capacity allocation,
the transport market clearing conditions must be imposed. These conditions state that
the demand for transport equals the supply of transport in both markets and can be
represented by the following equations;
qij =
N∑
l=1
Qlij (3.7a)
qji =
N∑
l=1
Qlji. (3.7b)
Combining the demand for transport given by equation (3.3), the first-order and
Kuhn-Tucker conditions given by (3.6a)-(3.6e) and the market clearing conditions
represented by equations (3.7a) and (3.7b), an equilibrium solution with multiple cases
can be obtained. While the details of the derivation are fairly standard, there are a few
aspects of the different equilibrium cases that are important to point out.
First, given the model, any feasible equilibrium solution requires at least one binding
capacity constraint, K l ≥ Qlij and/or K l ≥ Qlji. This implies that any solution to
this static model is characterized by full capacity utilization in at least one of the two
transport markets.
Second, the set of solutions includes equilibrium cases, where optimal transport
supplies and international trade are zero valued. These scenarios arise when marginal joint
and/or access costs and the resulting freight rates are prohibitively high.
Third, equilibrium solutions involving positive unilateral or bilateral international
trade exist and can be derived in symmetric pairs that simply interchange the i and j
notation. Thus, without loss of generality, I treat the transport market facilitating trade
from country i to country j as the fronthaul and trade from country j to country i as the
backhaul for the remainder of the analysis.
86
As a matter of tractability, the remaining analysis solely focuses on cases where
equilibrium transport supplies and international trade are positive in both transport
markets. This limits the analysis to two potential solutions that differentiate between
balanced and imbalanced bilateral trade. Naturally, the realization of a particular solution
simply depends on the imbalance of the demands for internationally traded goods between
two countries. Given equation (3.3), this demand imbalance can be represented and
rewritten as follows;
qij ≥ qji =⇒ pjτji − piτij ≥ (fij − fji), (3.8)
As equation (3.8) shows, the size of the trade imbalance depends on the difference in
domestic sales prices, as well as the endogenously adjusting freight rate differential.
Intuitively, small differences in the bilateral demands for transport, due to small sales price
variations across countries i and j, may allow carriers to choose equal transport supplies
that maximize capacity utilization in both transport markets and lead to asymmetric
freight rates that offset the sales price differential. In fact, it can be shown that a balanced
trade equilibrium arises only if the difference in sales prices is restricted to the following
interval:
pjτji − piτij ∈
(
aij − aji − ∂JC(K
l)
∂K l
; aij − aji + ∂JC(K
l)
∂K l
)
, (3.9)
whereas an imbalanced bilateral trade equilibrium, with country i as the net exporter to
country j, results when pjτji − piτij >
(
aij − aji + ∂JC(Kl)∂Kl
)
. 23
Case 1: Balanced Trade
For small transport demand imbalances, each carrier’s equilibrium supplies of
transport for a given round trip between countries i and j and the resulting equilibrium
23Due to symmetry, bilateral trade is also imbalanced when piτij−pjτji >
(
aji − aij + ∂JC(K
l)
∂Kl
)
. In this
case, country j becomes the net exporter and the transport market ji becomes the fronthaul.
87
transportation rates can be derived as follows:
K l = Qlij = Q
l
ji =
1
N
[
σN
2(σN − 1)
σ
σ − 1
(
aij + aji +
∂JC(K l)
∂K l
+ piτij + pjτji
)]−σ
(3.10a)
fij =
σN
2(σN − 1)
(
aij + aji +
∂JC(K l)
∂K l
+ pjτji
)
+
2− σN
2(σN − 1)piτij (3.10b)
fji =
σN
2(σN − 1)
(
aij + aji +
∂JC(K l)
∂K l
+ piτij
)
+
2− σN
2(σN − 1)pjτji. (3.10c)
Case 2: Imbalanced Trade
Solving the model when the demands for transport are strongly imbalanced yields
the following expressions for each carrier’s equilibrium capacity allocation and transport
supplies, as well as the resulting bilateral equilibrium transportation rates:
K l = Qlij =
1
N
[
σ
σ − 1
σN
σN − 1
(
aij +
∂JC(K l)
∂K l
+ piτij
)]−σ
(3.11a)
Qlji =
1
N
[
σ
σ − 1
σN
σN − 1 (aji + pjτji)
]−σ
(3.11b)
fij =
σN
σN − 1
(
aij +
∂JC(K l)
∂K l
)
+
1
σN − 1piτij (3.11c)
fji =
σN
σN − 1aji +
1
σN − 1pjτji. (3.11d)
Discussion
Thus, in the balanced trade case, the partial equilibrium,
(
qij, qji, K
l, Qlij, Q
l
ji, fij, fji
)
,
of transport market pair ij facilitating balanced bilateral trade between countries i and j
is described by equations (3.3) and (3.10a)-(3.10c). In contrast, in the imbalanced trade
case, the partial equilibrium,
(
qij, qji, K
l, Qlij, Q
l
ji, fij, fji
)
, of transport market pair ij
88
facilitating imbalanced bilateral trade between countries i and j is described by equations
(3.3) and (3.11a)-(3.11d). A comparison between both equilibrium cases reveals several
key features that are present when trade is facilitated by an international transportation
industry that is subject to the backhaul problem. Specifically, each carrier’s supply
of transport depends on marginal access costs, regardless of whether a given route is
considered a fronthaul or backhaul, or whether trade is balanced or imbalanced. The
allocation of marginal joint costs, however, heavily depends on the given trade imbalance.
That is, if trade is balanced, marginal joint costs play a role in the determination of both
fronthaul and backhaul equilibrium transport supplies and freight rates and therefore, lead
to the integration of bilateral trade costs. In contrast, if trade is imbalanced, marginal
joint costs only matter to the determination of fronthaul transportation supply and the
fronthaul freight rate leading to the disintegration of bilateral trade costs.
Comparative Statics
Based on these partial equilibrium scenarios, the response of trade to a shock
in carrier costs can be evaluated. Given each carrier’s cost structure, two alternative
transportation supply shocks can be considered. That is, both a change in marginal
access costs, as well as a change in marginal joint costs can have an impact on trade. For
notational convenience, I express marginal joint costs with JC ′ for the remainder of the
analysis.
First, I consider a shock to marginal access cost. In the balanced bilateral trade case,
the elasticity of trade with respect to a change in marginal access costs is given by
∂qij
∂aij
aij
qij
= −σ aij
piτij + pjτji + aij + aji + JC ′
< 0 if Qij = Qji, (3.12)
89
while in the imbalanced trade case, this elasticity can be represented as follows:
∂qij
∂aij
aij
qij
= −σ aij
(piτij + fij)
∂fij
∂aij
=
 −σ
aij
piτij+aij+JC′
< 0 if Qij > Qji (fronthaul)
−σ aij
piτij+aij
< 0 if Qij < Qji (backhaul)
.(3.13)
Equations (3.12) and (3.13) illustrate one of the key points of this study. When trade
is imbalanced, equation (3.13) demonstrates that the elasticity of trade with respect to
marginal access costs varies across fronthaul and backhaul transport markets. That is,
while both effects depend on the elasticity of import demand with respect to import
prices, marginal access costs, the local sales price and ad valorem trade costs, marginal
joint costs only contribute to the elasticity of trade facilitated on fronthaul transport
markets. Because of this critical distinction, one can show that fronthaul trade is more
inelastic with respect to marginal access costs than otherwise identical backhaul trade.
This result implies that when trade is imbalanced, access cost related policy outcomes are
larger in backhaul relative to fronthaul transport markets. This finding is intuitive. Since
marginal access costs represent a larger share of total trade costs in backhaul relative to
fronthaul transport markets, backhaul trade should be more responsive than fronthaul
trade to an identical change of these costs.
Proposition 1 If equilibrium trade is imbalanced, the effect of an identical
change in marginal access costs is larger for trade facilitated in backhaul
transport markets than for trade facilitated in fronthaul transport markets,
ceteris paribus. (A proof of this proposition is provided in Appendix A)
In contrast, when trade is balanced, equation (3.12) shows that the elasticity of
trade with respect to marginal access costs depends on additional terms, including the
foreign sales price, pj. Since otherwise identical fronthaul and backhaul transport markets
90
are distinguished by this foreign sales price, its inclusion complicates the comparison of
the effects on otherwise identical trade facilitated in fronthaul and backhaul transport
markets. Nevertheless, several key results can be derived. First, when equilibrium trade
is balanced and foreign sales prices, pj and pk, for example, are identical, the response
of trade facilitated in the ij and ik transport markets is identical as well. Second, when
trade is balanced and the foreign sales prices simultaneously approach the respective upper
and lower bound of expression (3.9), the difference in the elasticities of trade concerning
backhaul and fronthaul transport markets is smaller than the difference in elasticities
derived from the imbalanced trade equilibrium. This suggests that the difference in
fronthaul and backhaul policy outcomes should decrease, as the demands for transport
become more balanced.
In addition to these comparisons across fronthaul and backhaul transport markets in
the balanced and imbalanced bilateral trade cases, variations of the trade elasticity with
respect to marginal access costs across different product varieties can be considered as
well. Naturally, one would expect trade in product varieties with large values and thus,
low relative transport costs to exhibit smaller responses to an identical change in marginal
access costs than trade in low valued product varieties with high relative transport costs.
Indeed, based on equation (3.13), it can be shown that when imbalanced backhaul trade
is considered, the elasticity of trade with respect to marginal access costs becomes more
inelastic as the sales price of a given variety increases.24 Intuitively, this theoretical result
can be explained as follows; since transport costs represent a larger barrier to international
trade for bulky and heavy goods, such as metals or assembled furniture, trade in these
24An extension of this result to trade in imbalanced fronthaul transport markets or the balanced trade
case requires an additional nontrivial assumption on the size of the second derivative of joint costs with
respect to allocated capacity.
91
products is more responsive to change in carrier access costs than trade of small and
valuable products, such as electrical machinery.
Proposition 2 If equilibrium trade is imbalanced, the absolute value of the
elasticity of trade with respect to marginal access costs in backhaul transport
markets is decreasing in the sales price of any given variety. (A proof of this
proposition is provided in Appendix A)
Next, I consider the effects of a change in marginal joint costs. Again, I differentiate
between the balanced and imbalanced trade cases. In the balanced case, I obtain the
following expression for the elasticity of trade with respect to marginal joint costs:
∂qij
∂JC ′
JC ′
qij
= −σ JC
′
piτij + pjτji + aij + aji + JC ′
if Qij = Qji, (3.14)
whereas in the imbalanced trade case this elasticity of trade can be represented as:
∂qij
∂JC ′
JC ′
qij
= −σ JC
′
(piτij + fij)
∂fij
∂JC ′
=
 −σ
JC′
piτij+aij+JC′
if Qij > Qji (fronthaul)
0 if Qij < Qji (backhaul).
(3.15)
The interpretation of these results is very similar to the previous comparative statics.
That is, the elasticity of trade with respect to marginal joint costs is equal to the share of
marginal joint costs relative to all trade costs, including the sales price, ad valorem trade
costs, access costs, as well as marginal joint costs and is scaled by the price elasticity of
trade. In the imbalanced trade case, equation (3.15) shows that the response of trade
to shock in marginal joint costs strongly depends on whether a given transport market
is characterized as a fronthaul or a backhaul. In particular, when trade is imbalanced, a
rise in marginal joint costs triggers a reduction of trade facilitated in fronthaul transport
markets, while backhaul trade is unaffected. This variation concerning the elasticity of
92
trade with respect to marginal joint costs in the imbalanced trade case stems from the
carrier cost allocation. Given sufficient imbalances concerning the demands for transport,
carriers provide asymmetric transport supplies and allocate joint costs solely to the
fronthaul transport market. Due to this cost allocation, a change in marginal joint costs
has no impact on trade facilitated in backhaul markets.
Overall, these results show that when equilibrium bilateral trade is imbalanced,
maritime transport policy outcomes strongly depend on the identification of fronthaul
and backhaul transport markets, regardless of whether access or joint costs are affected.25
However, the specific responses of trade in each of these transport markets heavily depend
on the type of carrier cost affected by the given policy. Interestingly, none of the derived
trade elasticities depend on the number of carriers serving a given transport market. This
implies that the established variation in policy-induced trade effects is robust to changes in
the transport market structure.
The significance of these theoretical results is derived from the fact that trade
policy, aimed at improving infrastructure to lower the costs of international carriers and
reduce the unit-specific barriers to trade, can have very different effects depending on
the type of carrier cost that is affected and depending on whether the change in cost
applies to a fronthaul or backhaul transport market of a balanced or imbalanced bilateral
trade relation. Consider, for example, U.S. containerized trade. While U.S. containerized
exports typically represent a backhaul to a carrier’s round trip in recent decades, U.S.
containerized imports typically represent a fronthaul. The theoretical results suggest
that trade policy, such as the StrongPorts26 initiative by the Maritime Administration
25The theoretical findings presented in this subsection are focused on policy outcomes with respect
to the volume of trade, qij . All of these results continue to hold when the value of trade is considered
instead. The derivation of the trade elasticities in the value case are provided in Appendix B.
26The basic goal of the Maritime Administration’s StrongPorts initiative is to provide support for the
development of projects that increase port freight efficiencies. Port efficiency gains can certainly effect
93
or the Trade Facilitation Agreement27 by the WTO, will lead to very different outcomes
concerning U.S. containerized exports and imports. In fact, a reduction in marginal access
costs that applies to both U.S. containerized exports and imports is expected to have a
larger effect on U.S. exports than imports, while a reduction in marginal joint costs is
expected to mainly affect U.S. containerized imports.
Empirical Model
Policy Shocks
To test the theoretical propositions concerning the heterogeneity of trade policy
outcomes in the presence of the backhaul problem, I build on the standard empirical
model of trade, the gravity equation. This model is ubiquitous in the trade literature and
has been heavily used to not only analyze the determinants of trade, but also evaluate
the effects of commercial policy. Identification of the theoretically predicted systematic
variation in maritime transport policy outcomes relies on an exogenous environmental
regulation by the EU.
More specifically, identification of the potentially varying responses of trade to a
given change in carrier costs is achieved through the low sulfur fuel requirement imposed
on liner carriers through a revision of the EU Council Directive 1999/32/EC. Initial
European regulation of fuel sulfur contents was imposed by the European Council via
Directive 93/12/EC in March of 1993. While some of the provisions of this Directive were
later repealed, the low sulfur requirements, that were in line with regulations set by the
carrier costs, although it depends on the specific project to determine whether joint or access costs are
affected.
27The central focus of the Trade Facilitation Agreement concerns the simplification and standardization
of customs practices to expedite the movement of goods. Improved release and clearance times could
potentially lower carrier access costs by reducing port handling times and terminal costs.
94
International Maritime Organization (IMO) on heavy fuel and other marine oils, were
subsequently instated via Council Directive 1999/32/EC, drafted in April of 1999. The
key revision of the low sulfur regulations set forth in this Directive was initiated in July
of 2005 via EU Directive 2005/33/EC. This latest revision was created in response to the
European Commission’s strategy to reduce atmospheric emissions from seagoing ships
(European Commission, 2015).28 While there are several revisions set forth in Directive
2005/33/EC to improve air quality for the protection of human health, the one of interest
to this study concerns the requirement that as of January 1, 2010, all liner carriers must
use fuels containing no more than a maximum level of up to 0.1% sulfur once at berth or
anchorage in EU ports.29
This requirement marked a significant reduction in the allowable fuel sulfur content.
At the time this revision went into effect, restrictions set by IMO regulations required
carriers to use fuels not exceeding a maximum level of 4.5% sulfur globally and 1.5%
in specific Emission Control Areas (ECAs) around Europe (Cullinane and Bergqvist,
2014). In an assessment of various impact studies, the European Maritime Safety Agency
predicted that a change from 1.5% to 0.1% sulfur fuel content would result in an average
fuel premium of 74% ((EMSA), 2010). In line with this prediction, Notteboom et al.
28The strategy discusses the effects of ship emissions in the EU and proposes a variety of policies to
reduce the shipping emissions and their contribution to acidification, ground-level ozone, eutrophication,
health, climate change and ozone depletion.
29There are a number of exceptions to the Directive’s requirement. Most notably, if the scheduled
duration of a vessel anchored at an EU port does not exceed two hours, the low sulfur fuel requirement
does not apply. However, this provision tends to only be relevant to ferry services and other smaller ships,
rather than large container vessels facilitating transatlantic trade. Alternatively, if a carrier agrees to
completely shutdown a vessel’s engines while at berth in an EU port, fuels do not have to be switched
over. However, the feasibility of this alternative depends on whether a given port provides shore-side
electricity from the national grid. While EU ports are encouraged to provide this service, they are not
required to do so, as of January 2010. Since these exclusions reduce the burden of the low sulfur fuel
requirement on the average carrier, their existence may create a potential downward bias in the estimation
of treatment effects. Thus, the results can be interpreted as conservative estimates of the actual policy
impact.
95
(2010) show that the long run price difference between Intermediate Fuel Oil (IFO 380)
and Marine Gas Oil (MGO 0.1% sulfur) averaged around 93% between 1990 and 2008.
The authors claim that, due to increasing demand and the cost of the desulfurization
process, the cost of marine distillate fuels is roughly twice that of residual fuels and causes
a significant increase in container freight rates. Further anecdotal evidence suggests that
these increases in required fuel prices would equate to a cumulative increase of fuel costs of
around $1.3 billion per year (Ivanov, 2010).30 As such, the low sulfur fuel requirement
meets the necessary condition of significantly influencing carrier costs pertaining to
international shipments of U.S.-EU trade.
The exogeneity of this change to the existing sulfur content regulations stems from
the fact that its origination and implementation has been motivated by the European
Commission’s environmental strategy to reduce pollution from maritime shipping, rather
than trade related issues. There are a few studies that have analyzed a variety of issues
related to the low sulfur fuel requirement of Directive 2005/33/EC (see, for example,
Endresen et al., 2005; Schrooten et al., 2008; Bosch et al., 2009). However, to the best
of my knowledge, there are no studies that have estimated the requirement’s effects on
international trade or considered the potential heterogeneity of its impact across trade
facilitated in fronthaul and backhaul transport markets.
Of course, the considerable gap between the publication of EU Directive 2005/33/EC
and its implementation may have caused a few issues to properly identify the resulting
treatment effects on U.S.-EU trade. An immediate concern leading to the possible inflation
of the estimated treatment effects may be the potential anticipatory changes in trade
immediately prior to the implementation of the low sulfur fuel requirement in an attempt
30Although there were several alternative emissions abatement technologies approved as of January
2010, all of these technologies placed a significant burden on liner carriers involved in EU trade as well
(P&O Ferrymasters).
96
to avoid rising trade costs. To address this potential issue, Figures 3a and 3b display
October, November, and December growth rates of U.S.-EU trade by year. The figures
show, that for both U.S. exports and imports with treated European countries, there is
no evidence of extraordinary average trade growth in 2009, indicated by the vertical line,
immediately before the treatment implementation. That is, in comparison to previous
years, there is no evidence of a build-up or rush to get shipments in or out prior to the
implementation of the cost raising low sulfur fuel requirement annulling this potential
concern.
Another issue associated with the sizable time lapse between publication and
implementation of this policy may be early adjustments by international carriers that
could potentially put a downward bias on the estimated treatment effects. Anecdotal
evidence, however, suggests that the majority of carriers avoided pre-implementation
adjustment costs by playing a waiting game in hopes of a last minute EU repeal or
postponement of the low sulfur fuel requirement (Ivanov, 2010). Refusing to yield to
carrier opposition, on December 21, 2009, the EU published a statement re-enforcing the
timely implementation of the low sulfur fuel requirement (European Commission, 2009)
forcing the majority of non-compliant shipowners to immediately adjust their practices
and nullifying this concern as well.
Empirical Specification
Given this identification strategy and data availability, a difference-in-differences
(henceforth referred to as DID) estimator is the natural choice to evaluate the varying
trade outcomes of commercial and related environmental policy. Of course, various DID
specifications are available and have been used in the literature. These specifications range
from indicator variables that capture the average treatment effect, to the inclusion of
97
−
.
1
−
.
05
0
.
05
.
1
.
15
G
ro
wt
h 
Ra
te
s
2000 2005 2010 2015
Year
Oct. growth rates Nov. growth rates
Dec. growth rates
(a) U.S. Exports
−
.
1
0
.
1
.
2
G
ro
wt
h 
Ra
te
s
2000 2005 2010 2015
Year
Oct. growth rates Nov. growth rates
Dec. growth rates
(b) U.S. Imports
FIGURE 3. Average End-of-the-Year Monthly Growth Rates of U.S.-EU Trade
treatment and control group, or even panel-specific, time trends (e.g. Friedberg, 1998),
which are intended to control for a violation of the parallel paths assumption. Other
specifications use time varying post-treatment indicator variables in conjunction with
group-specific time trends to capture the dynamic response to policy changes (e.g. Wolfers,
2006) or exclude time trends altogether and instead use time-varying pre- and post-
treatment dummies for a more flexible specification (e.g. Mora and Reggio, 2012). In this
study, I employ the standard DID estimator and include a treatment indicator variable,
δjt, that captures the difference in trade across treatment and control groups post-policy
implementation.31 As Bertrand et al. (2004) have shown, the standard DID estimator can
suffer from significant bias in the presence of serial correlation in both the dependent and
dummy variable.32 Following suggestions by Bertrand et al. (2004), standard errors are
31In the following section, I present data plots that provide evidence in support of the parallel paths
assumption and dissuade the use of time trends in conjunction with the DID estimator. Nevertheless, as
part of the robustness analyses, I estimate the model including treatment and control group, as well as
country-specific time trends. In general, the treatment effect estimates and their statistical significance are
robust to the inclusion of these time trends and the results are presented in Panels 2 and 3 of Table 23 in
Appendix C.
32Moreover, Donald and Lang (2007) show that biased coefficient estimates can also occur when the
number of panels in a given dataset is small. This, however, is not a concern for analysis conducted in this
study as the dataset includes over 1400 U.S. port-foreign country pairs.
98
clustered at the state, rather than state-time level. For the application in this study, this
implies that standard errors are, in fact, clustered at the U.S. port-foreign country level.33
In-line with the majority of the empirical trade literature, the DID estimator
employed in this study is incorporated into the standard gravity equation framework.
In addition to the indicator variable, δjt, which captures the exogenous shock to unit-
specific transport costs involving EU trade, all other trade determinants are accounted for
following standard practices. In particular, economic mass is captured via total exporter
and importer employment, Lit and Ljt, while the multilateral resistance terms introduced
by Anderson and van Wincoop (2003) are controlled for by means of various fixed effects,
including exporter and importer, ai and aj, as well as time fixed effects, at.
34 Since the
empirical analysis is focused on monthly containerized trade flowing to and from a variety
33Some concerns may arise due to the potential cross cluster correlation at more aggregated state levels.
These include potential bilateral trade correlations across the U.S. port-foreign country pairs that respond
to a common shock at the U.S. state level, as well as potential trade correlations across foreign countries
that experience a common shock at the U.S. port or state level. To address these concerns, I have re-
estimated the model clustering standard errors at the U.S. state-foreign country level, U.S. port level, and
U.S. state level. In general, the statistical significance of coefficient estimates is robust to these variations
in clustering and results are presented in Table 24 of Appendix C.
34Within the trade literature focused on the estimation of the gravity equation, the use of fixed effects
to capture multilateral trade cost differences has rapidly evolved in recent years. The use of fixed effects
commenced with the inclusion of exporter/importer dummy variables (see, e.g., Harrigan, 1996; Egger,
2000; Feenstra, 2002), and advanced to more sophisticated specifications also including bilateral fixed
effects (e.g. Egger and Pfaffermayr, 2003). The progression culminates in the use of time varying exporter
and importer fixed effects, in addition to bilateral fixed effects, as in Baldwin and Taglioni (2006), for
example. In this study, I estimate the average treatment effects of the low sulfur fuel requirement on U.S.-
EU trade. The estimation sample includes monthly observations of U.S. containerized exports and imports
to and from OECD and APEC countries. Since the U.S. is the common trade partner for all bilateral
trade pairs, exporter and importer fixed effects capture not only country-specific, but also bilateral-specific
unobservables. While the low sulfur fuel requirement varies across foreign OECD and APEC countries,
it is implemented at one specific date. Identification off of this policy, thus, prohibits the use of time-
varying exporter and importer fixed effects. However, since the U.S. is the common trade partner for
all trade observations, the inclusion of time fixed effects controls for the time variation of U.S. related
multilateral trade cost differences. To ensure the robustness of the primary findings against this fixed
effects specification, I also estimate the model including time-varying regional, state, or port level fixed
effects that capture more disaggregated U.S. trade related time varying unobservables. The results are
presented in Table 25 of Appendix C and illustrate that the variation in trade policy outcomes is largely
consistent against various fixed effects specifications.
99
of foreign countries through U.S. ports of entry, the empirical specification also includes
port fixed effects, ap, to control for the heterogeneity across these ports.
Given the fact that the data sample only includes U.S. trade, the use of these
fixed effects not only controls for the standard time-varying national trends and time-
invariant bilateral trade specific ad valorem trade costs, but also for systematic differences
between pre- and post-treatment periods and between treatment and control group
countries. To further ensure the accurate estimation of the treatment effect, I include
various control variables summarized in vector Zrijt. This collection of variables includes
indicator dummies for U.S. free trade agreements with OECD and APEC countries and
an interaction term capturing any variation in the response of U.S.-EU trade to the
Great Trade Collapse (GTC) relative to all other U.S. trade. In addition to these dummy
variables, the vector Z also includes regional real U.S. retail diesel fuel prices to control
for changes in heavy fuel oil and low sulfur fuel prices, which vary over time, t, and are
available at the U.S. regional level, r.
This gives rise to the following empirical specification;
xijpt = exp
(
β0 + β1ln(Lit) + β2ln(Ljt) + β3δjt + γln
(
Zrijt
)
+ ai + aj + at + ap
)
ijpt, (3.16)
where the dependent variable, xijpt, reflects the value of trade facilitated from country
i to country j through U.S. port p at time t, ln
(
Zrijt
)
represents the natural log of each
element of the vector Z, and the random component is given by ijpt. The key parameter
of interest is given by β3. A negative and statistically significant estimate of β3 would
indicate a decrease in U.S. trade with an EU country relative to U.S. trade with non-
member countries due to the implementation of the low sulfur fuel requirement.
100
Data
The data employed in the estimation of this empirical model are comprised of a
number of variables obtained from several different sources. The main time series of
interest, and dependent variable in the empirical model, is given by U.S. containerized
maritime bilateral trade with the majority of OECD and APEC countries35 at the seaport-
of-entry level. The data were obtained from the U.S. Census Bureau, USA Trade Online
database and include monthly observations from January of 2003 until February of 2015.
This dataset provides the unique opportunity to closely identify fronthaul and backhaul
transport markets facilitating U.S. trade; an identification that would be lost at national
or yearly aggregation levels. The USA Trade Online dataset includes a variety of ports
with vastly different trade volumes. The selection of ports included in this analysis is
based on economic significance. That is, only ports of entry with an annual import volume
of over $100 million and simultaneous annual export volume of over $50 million in 2014
have been included in the sample. This restricts the sample to the forty-three largest
exporting and importing container ports of entry in the U.S. Over the sample period,
these ports account for roughly 98.70% and 97.75% of total U.S. container imports and
exports, respectively.
35The selection of U.S. trade with OECD and APEC countries rests on the data availability concerning
economic mass control variables at monthly frequency. Even among these OECD and APEC countries,
the unavailability of macroeconomic data leads to the exclusion of several members including Brunei
Darussalam, Chinese Taipei, Estonia, Israel, Papua New Guinea, and Slovenia. In addition to these
sample restrictions, Austria, the Czech Republic, Hungary, Luxembourg, Slovakia and Switzerland are
excluded from the sample because of an unclear treatment or control group status. All of these countries
are landlocked and located in Europe. It is unclear whether U.S. trade with these countries is subject to
the low sulfur fuel requirement, as these international transactions may be facilitated by non-EU ports.
Canada and Mexico are excluded from estimation sample because the liner shipping industry serving
the associated transport markets faces a unique market structure, where carriers compete with external
transportation options, such as rail or trucking, that are unavailable for alternative bilateral U.S. trade
relations. Generally, the inclusion of the later two groups of countries does not alter the primary empirical
findings or their statistical significance. The specific results are reported in Panels 2 and 3 of Table 26 in
Appendix C.
101
The extensive trade data are complemented by U.S. and international total
employment observations. Monthly data on total non-farm employment by U.S. state
have been procured from the U.S. Bureau of Labor Statistics. International employment
data for OPEC and APEC countries have been obtained from the International Labour
Organization (ILO).36 These employment data are used as proxies for importer and
exporter income. Although Gross Domestic Product (GDP) at the national level is the
common proxy for economic mass in the gravity literature, monthly observations of these
data are unavailable for the majority of countries included in the sample. Due to this
unavailability, highly correlated employment statistics at the foreign country and U.S.
state-level are used to control for economic mass variation instead. The application of
state-level, rather than national, U.S. employment data controls for local income variations
that may vary from national trends and could potentially influence the local port-of-entry
trade flows.37
As indicated in the previous section, the additional control variables include U.S. free
trade agreements and monthly observations on regional U.S. diesel fuel prices. The diesel
fuel price data have been obtained from the Energy Information Administration (EIA).
These prices are intended to control for the time variation in costs of heavy fuel oil (HFO)
and low sulfur fuels that may change access costs, aside from the EU directive.38 Data on
36ILO statistics vary in frequency and are a compilation of employment time series from various national
sources. The ILO sources of data used in this analysis include the EU Labour Force Survey, the Labour
Force Survey, the Population survey on employment problems, the General Household Survey, National
Labour Force Survey, the Economically active Population Survey, the Household Labour Force Survey,
the Nueva Encuesta Nacional de Empleo, the Encuesta Nacional de Ocupacio´n y Empleo, the Encuesta
Especializada de Niveles de Empleo, and Official Estimates.
37As part of the robustness analysis, I test whether the empirical findings are sensitive to the exclusion
of state-level U.S. employment. The results, presented in Panel 2 of Table 27 in Appendix C, demonstrate
that the evidenced variation in policy-induced trade effects is robust against the exclusion of state-level
U.S. employment.
38Due to the limited price data availability of HFO used in the container shipping industry, the No. 2
distillate retail sales prices by refiners are included in this dataset instead. According to the EIA, residual
102
U.S. free trade agreements, enacted during the 2003-2015 sample period considered in this
study, were obtained from the Office of the United States Trade Representative (USTR)
and involve the sample countries of Australia, Chile, Peru, Singapore and South Korea.
The dummy variables indicating these free trade agreements are intended to control for
trade liberalization efforts that result in the time variation of U.S. ad valorem trade costs,
such as tariffs or quotas. The inclusion of the diesel fuel prices and data on free trade
agreements completes the unique data set employed in this study.
To summarize and provide a description of these data, as well as justification
for the appropriateness of the difference-in-differences estimator, I consider various
dimensions of the data. In Table 9, I summarize the U.S. trade data along the cross-
sectional dimension and provide summary trade statistics for each of the EU member
and non-member countries included in the sample. The summary statistics reveal that
U.S. containerized trade varies by the direction of trade (exports vs. imports), across
countries, and treatment and control groups. While some countries, like China or Japan,
hold considerable shares of total U.S. containerized imports, the majority of countries
command shares of less than 1%. Similarly, the U.S. container export market exhibits
few countries with large market shares, such as China, Japan, or South Korea, and many
countries holding export shares around 1%. However, the U.S. container export market
appears to be less concentrated, with market shares ranging from 0.03% to 13.02%, than
the U.S. container import market, with shares ranging from 0.02% to 38.97%. Despite
these differences, there are also some commonalities between exports and imports and
between the treatment and control groups. In fact, for both exports and imports, the ten
largest U.S. trade partners (measured in average trade value) command over 50% market
fuel oils may contain No. 2 distillate in order to meet specifications (Wallace, nd) and thus, warrants its
use as a proxy for HFO and low sulfur fuel prices.
103
share, respectively, are comprised of both EU member and non-member countries and
share five common members including China, Japan, Germany, South Korea, and Taiwan.
A more detailed comparison between the sample EU member and non-member
countries reveals that, with the exception of the largest U.S. trade partners, including
China and Japan, the treatment and control groups are very similar concerning average
trade values. Specific to this sample, EU member countries hold 16.59% and 21.73%
market share of U.S. containerized imports and exports, respectively, while non-member
countries excluding China and Japan account for 19.96% and 27.42% of these markets
(68.42% and 49.00% including China and Japan).39 These observations illustrate the
importance of controlling for systematic differences in trade values between the treatment
and control groups, as well as individual countries and support the use of a DID estimator
which accounts for these level differences that would otherwise bias the treatment effect
estimation.
TABLE 9. U.S. Containerized Trade by Foreign Country
(1) (2) (3) (4) (5) (6)
Imports Exports
Country Trade Trade BH Trade Trade BH
($ Mil.) Share (%) Share (%) ($ Mil.) Share (%) Share (%)
EU Members
Germany 2,520 5.44 19.6 612 3.76 80.4
Italy 1,230 2.66 16.78 252 1.55 83.22
France 960 2.07 24.81 293 1.8 75.19
United Kingdom 895 1.93 41.41 666 4.1 58.59
Netherlands 465 1 61.02 580 3.57 38.98
Spain 368 0.8 34.95 167 1.03 65.05
Ireland 331 0.72 35.57 65 0.4 64.43
Continued on next page
39Since Japan and, in particular, China hold such market dominant positions concerning U.S. bilateral
trade, I re-estimate the empirical model excluding both of these countries. The empirical results are given
in Panel 4 of Table 26 in Appendix C and illustrate that estimated trade effects and their statistical
significance are largely insensitive to the exclusion of these two countries.
104
Table 9 – Continued
(1) (2) (3) (4) (5) (6)
Imports Exports
Country Trade Trade BH Trade Trade BH
($ Mil.) Share (%) Share (%) ($ Mil.) Share (%) Share (%)
Sweden 275 0.59 25.71 82 0.51 74.29
Belgium 246 0.53 70.35 662 4.07 29.65
Denmark 146 0.32 29.22 44 0.27 70.78
Finland 122 0.26 34.4 70 0.43 65.6
Poland 119 0.26 29.09 55 0.34 70.91
Portugal 86 0.18 20.54 16 0.1 79.46
Greece 39 0.08 41.15 24 0.15 58.82
Total 7,680 16.59 - 3,530 21.73 -
Non-Members
China 18,000 38.97 11.62 2,120 13.02 88.38
Japan 4,390 9.49 24.09 1,390 8.56 75.91
Korea, South 1,560 3.37 32.62 821 5.05 67.38
Taiwan 1,500 3.24 18.74 535 3.29 81.26
Thailand 994 2.15 16.76 203 1.25 83.24
Vietnam 962 2.08 10.6 129 0.8 89.4
Indonesia 926 2 13.43 197 1.21 86.57
Malaysia 672 1.45 14.27 143 0.88 85.73
Australia 383 0.83 56.76 598 3.68 43.24
Philippines 356 0.77 28.94 128 0.79 71.06
Chile 298 0.64 40.52 217 1.34 59.48
Russia 254 0.55 60.33 205 1.26 39.67
Hong Kong 244 0.53 67.66 422 2.6 32.34
Singapore 241 0.52 64.45 364 2.24 35.55
Turkey 236 0.51 35.37 182 1.12 64.63
New Zealand 204 0.44 41.24 73 0.45 58.76
Peru 180 0.39 37.72 140 0.86 62.28
Norway 105 0.23 36.74 39 0.24 63.26
Iceland 9 0.02 53.6 5 0.03 46.4
Total 31,700 68.42 - 7,960 49.00 -
Source: U.S. Census Bureau USA Trade Online database
Another interesting feature of the data concerns the share of backhaul transport
markets for U.S. containerized exports and imports by foreign country. Recall, that a
105
transport market is defined as a backhaul when current bilateral trade flows facilitated
in this market are less than the current bilateral trade flows facilitated in the transport
market of opposite direction.40 The country-specific backhaul share data are presented
in columns (3) and (6) of Table 9 and demonstrate that while the share of backhaul
transport markets exhibits large heterogeneity across countries, it systematically varies
across U.S. exports and imports.41 In particular, as columns (3) and (6) illustrate, U.S.
containerized exports tend to be facilitated in backhaul transport markets, while U.S.
containerized imports typically represent a fronthaul transport market. That is, on a
round trip facilitating bilateral trade between the U.S. and an OECD or APEC country,
the majority of U.S. containerized imports from these countries reflect a fronthaul market
for international carriers, while the majority of U.S. containerized exports to these
countries reflect a backhaul market for these carriers.
In fact, given this sample, only 28.7% of the total U.S. containerized imports from
EU member countries and 17.8% from non-member countries are facilitated in backhaul
transport markets. In contrast, over 58% of the sample U.S. containerized exports to EU
member countries and 70.7% of the sample U.S. exports to non-member countries are
facilitated in backhaul transport markets. This shows that there is no inherent difference
40Average shipment durations between the U.S. and Asia or Europe can range from 8 to approximately
30 days depending on the port of origin and destination. To address the possible one month lag between
the fronthaul and backhaul route on a given round trip between the U.S. and an Asian or European
country, I perform several robustness checks where the backhaul transport market is defined by comparing
this month’s U.S. imports(exports) to last month’s U.S. exports(imports) for the same U.S. port-foreign
country pair. Estimations based on these redefined fronthaul and backhaul routes yields consistent and
statistically significant average treatment effects that are presented in Table 28 of Appendix C.
41While the primary estimates rely on the backhaul identification at the U.S. port-foreign country level,
actual shipping routes may include several stops at a few major U.S. ports and foreign countries. The
existence of these primary transatlantic and transpacific shipping routes may alter the fronthaul and
backhaul definitions assigned to each U.S. port-foreign country observation. In an attempt to match these
shipment patterns and address potential concerns, I aggregate the bilateral U.S. port trade data at state,
supranational, as well as state-supranational levels and re-estimate the empirical model. The results reflect
point estimates of similar magnitude and statistical significance and are presented in Panels 2 through 4 of
Table 29 in Appendix C.
106
between the backhaul shares of treatment and control groups, but systematic variation
across U.S. containerized imports and exports. Since bilateral U.S. trade tends to be
imbalanced, the theoretical findings suggests that maritime transport policy applied
to both U.S. containerized imports and exports should, in fact, produce very different
responses in U.S. trade regardless of whether carrier joint or access costs are affected.
To further ensure the appropriateness of the DID estimator, the time dimension
of the data is considered next. In line with the theoretical and empirical model, I
differentiate the data between treatment and control groups, as well as the fronthaul
and backhaul transport markets, rather than between the traditional export and import
perspectives. Figures 4a and 4b illustrate the time path of logged U.S. containerized trade
facilitated in fronthaul and backhaul transport markets comparing the trends involving
U.S. trade with EU member and non-member countries over the entire sample period.
Both figures present supporting evidence of the initial observations from Table 9. That is,
U.S. containerized trade facilitated in fronthaul and backhaul transport markets is much
larger for non-EU member countries, than EU members. Despite these stark differences
in the levels of trade, Figures 4a and 4b suggest that both fronthaul and backhaul trade
exhibit very similar long-run growth patterns across treatment and control groups. This
provides supporting evidence for the validity of the parallel paths assumption required by
the standard DID estimator.
Treatment is indicated via the vertical line. However, due to large short-run
variations, Figures 4a and 4b provide no immediate insight concerning the potential
treatment effects. To alleviate this issue, Figures 4c and 4d present the same data, but
restrict the sample period around the treatment date of January 1st, 2010, when the
EU low sulfur fuel requirement went into effect. Based on Figures 4c and 4d, it appears
that fronthaul and backhaul U.S. trade with either the treatment or control group
107
19
.5
20
20
.5
21
21
.5
Lo
g 
of
 A
vg
. T
ra
de
 V
al
ue
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1 2016m1
Time
Type = Control Type = Treatment
(a) Fronthaul
18
.5
19
19
.5
20
Lo
g 
of
 A
vg
. T
ra
de
 V
al
ue
2002m1 2004m1 2006m1 2008m1 2010m1 2012m1 2014m1 2016m1
Time
Type = Control Type = Treatment
(b) Backhaul
20
20
.5
21
21
.5
Lo
g 
of
 A
vg
. T
ra
de
 V
al
ue
2008m1 2009m1 2010m1 2011m1 2012m1
Time
Type = Control Type = Treatment
(c) Fronthaul
19
19
.5
20
Lo
g 
of
 A
vg
. T
ra
de
 V
al
ue
2008m1 2009m1 2010m1 2011m1 2012m1
Time
Type = Control Type = Treatment
(d) Backhaul
FIGURE 4. U.S. Trade Flows across Treatment & Control Groups by Transport Market
declines around the treatment date. However, Figures 4c and 4d also indicate that the
magnitude of these declines is larger for the treatment than the control group and varies
across fronthaul and backhaul transport markets. The use of the DID estimator should
delineate the specific responses of U.S. trade from seasonal and otherwise noisy variation
and provide clear quantitative insights into the potential heterogeneity of trade policy
outcomes across fronthaul and backhaul transport markets.
108
Results
In this section, I first present the empirical findings obtained from the gravity
equation estimations of bilateral U.S. containerized trade with OECD and APEC
countries. An extension of the analysis further differentiates trade effects across the
balanced and imbalanced bilateral trade samples as well as different product groups.
I conclude the analysis with a series of robustness checks and a discussion of its
implications. For the majority of these estimations, I use the more recently developed
Poisson pseudo-maximum likelihood (PPML) estimator, as suggested by Santos Silva
and Tenreyro (2006). The primary results point to systematic variation in trade policy
outcomes across fronthaul and backhaul transport markets and demonstrate that this
variation can explain the otherwise surprising difference in trade effects across U.S. exports
and imports. The findings of the extended analysis further complement the theoretical
predictions. While the differentiation across various trade imbalances illustrates that the
difference in fronthaul and backhaul treatment effects decreases as trade becomes more
balanced, the disaggregated analysis reveals that trade in high ad valorem transport
cost products is more responsive to a shock in carrier costs than trade in low ad valorem
transport cost products.
Solidifying the primary empirical findings, I present and discuss the results of various
robustness checks involving standard sensitivity analyses, multiple model modifications
and various backhaul identifications, among others. In general, this secondary analysis
points to the consistency of the empirical results demonstrating systematic heterogeneity
of trade policy outcomes across fronthaul and backhaul transport markets. Concluding
this section, I emphasize the significance of these findings presenting current trade
imbalance data and deducing the suggested implications across developing versus
developed countries and geographic regions. Overall, these data suggest that transport-
109
related trade effects are expected to be consistently larger for low rather than high income
countries, but to vary dynamically across geographic regions.
Primary Results
The empirical results for the gravity equation estimation using the PPML estimator
are presented in Table 10. The estimations differentiate potentially varying policy effects
between the full sample including U.S. containerized exports and imports, as well as each
of these trade categories separately. The full sample estimation results, given in column
(1) of Table 10, reveal a negative, yet statistically insignificant, effect of the low sulfur fuel
requirement on bilateral U.S.-EU trade, relative to all other bilateral U.S. containerized
trade with OECD and APEC countries. This finding, however, is without distinction
between U.S. exports and imports. Differentiation between the two, presented in columns
(2) and (3) of Table 10, unmasks a small and statistically insignificant effect on U.S.
containerized imports, but large and statistically significant export treatment effect. That
is, U.S. containerized imports from EU countries experience no statistically significant
reduction in response to the implementation of EU Directive 2005/32/EC, while U.S.
containerized exports to EU members decrease by 7.99%[≈ (exp(−0.0833)− 1) ∗ 100] post
treatment; an estimate that is both economically and statistically significant. Without
further consideration of the transport sector this finding is surprising. The low sulfur fuel
requirement applies to all container vessels at berth in an EU port. Since container vessels
are subject to this increase in costs while unloading U.S. exports as well as loading U.S.
imports, one would expect both U.S.-EU exports and imports to be negatively affected by
the EU Directive.
However, as the theoretical model shows, carriers allocate costs according to
fronthaul and backhaul transport markets, rather than import-export specific routes.
110
Furthermore, the model predicts that, in imbalanced trade cases, increases in marginal
access costs affect trade in backhaul transport markets more than in fronthaul markets.
Since U.S. exports and imports are generally imbalanced and U.S. exports are mainly
facilitated in backhaul transport markets, this finding suggests that U.S. exports should
experience a larger decline in response to the low sulfur fuel requirement compared
to U.S. imports, which are mainly facilitated in fronthaul transport markets. To test
this hypothesis, I re-estimate the model differentiating between fronthaul and backhaul
transport markets, rather than U.S. imports and exports. The results are given in columns
(4) and (5) of Table 10 and provide supporting evidence for Proposition 1. Specifically, the
treatment effect estimates show that, albeit a negative coefficient point estimate, there is
no statistically significant impact on fronthaul transport markets. In contrast, backhaul
U.S.-EU trade experiences a statistically significant 9.87% reduction in response to the
EU environmental policy relative to all other U.S. containerized backhaul trade with other
OECD and APEC countries.
TABLE 10. ATE - PPML
(1) (2) (3) (4) (5)
VARIABLES Full Sample Imports Exports Fronthaul Backhaul
ATE, (δ) -0.030 -0.010 -0.083** -0.029 -0.104**
(0.033) (0.041) (0.040) (0.040) (0.045)
Observations 410,134 205,067 205,067 205,067 205,067
R-squared 0.896 0.918 0.679 0.903 0.764
Port FE Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
There are a few questions that may arise naturally. One concern may be the overall
magnitude of the estimated trade effects. Statistical as well as anecdotal evidence suggests
111
that compliance with the low sulfur fuel regulation requires carriers to use high quality
in-port fuels that command an approximate 100% premium and results in an estimated
$1.3 billion increase in annual fuel costs (see, for example, (EMSA), 2010; Ivanov, 2010;
Notteboom et al., 2010). Furthermore, Notteboom et al. (2010) have estimated that a
similar rise in fuel costs due to European Emission Control Areas will lead to freight rate
increases ranging from 8% to 40% for traditional and fast short sea services, respectively.
Based on this evidence, significant increases in transport costs and the resulting estimated
trade effects appear reasonable. Another concern that may arise, pertains to the difference
in magnitudes of the statistically significant treatment effects on U.S. exports to EU
countries compared to U.S.-EU trade facilitated in backhaul transport markets. A priori,
one might expect the treatment effect on backhaul transport markets to be equal to the
effect on U.S. exports. However, as the summary statistics in Table 9 reveal, not all U.S.
containerized exports are transported in backhaul markets. Since fronthaul markets show
no statistically significant effect, the overall reduction of U.S. exports, which represents a
partial blend of the fronthaul and backhaul treatment effects, is, in fact, expected to be
smaller than that exhibited by trade in backhaul transport markets. Overall, the estimates
provide consistent empirical evidence in support of Proposition 1, which states that trade
effects are larger in backhaul relative to fronthaul transport markets when marginal access
costs are affected.
Another potential source for the heterogeneity of commercial policy outcomes
concerns the differences in trade effects across balanced and imbalanced bilateral
equilibrium trade flows. The theoretical model predicts that the difference in fronthaul
and backhaul average treatment effects in response to an identical shock to marginal
access costs should decrease as more balanced bilateral trade observations are considered.
To test this hypothesis, I restrict the sample to rather balanced bilateral trade cases and
112
compare the estimated fronthaul and backhaul trade effects.42 Columns (1) and (2) of
Table 11 present the average fronthaul and backhaul treatment effects restricting the
sample to observations where the value of backhaul U.S. trade is at least half of the value
of fronthaul U.S. trade (η = min(xijpt/xjipt)/max(xijpt, xjipt) > 0.5). In contrast, Columns
(3) and (4) of Table 11 illustrate the results of identical estimations that restrict the
sample to even more balanced bilateral trade observations where the difference between
fronthaul and backhaul trade is no larger than 20% (η > 0.8). While Panel 1 considers all
balanced bilateral trade flows, including zero-valued trade, Panel 2 presents the results of
identical estimations, but limiting the sample non-zero valued trade observations.
TABLE 11. ATE - Varying Trade imbalance
(1) (2) (3) (4)
Fronthaul Backhaul Fronthaul Backhaul
η > 0.5 η > 0.5 η > 0.8 η > 0.8
Panel 1 -0.059 -0.080 -0.026 -0.036
ATE, (δ) (0.085) (0.091) (0.116) (0.119)
Panel 2 -0.068 -0.091 -0.061 -0.072
ATE, (δ) (0.085) (0.091) (0.118) (0.120)
Port FE Yes Yes Yes Yes
Time FE Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
In contrast to the previous results, restricting the sample to rather balanced bilateral
trade observations yields statistically insignificant treatment effects in both fronthaul and
backhaul transport markets. However, the point estimates of the individual treatment
effects, presented in columns (1)-(4) of Table 11, suggest nearly identical fronthaul and
42Perfectly balanced trade observations are rare and treatment effects cannot be identified in this case.
113
backhaul policy outcomes when trade is rather balanced. This finding nicely contrasts the
drastic variation in estimated fronthaul and backhaul trade effects when rather imbalanced
trade observations are included as well. In fact, comparing the coefficient estimates across
fronthaul and backhaul transport markets when zero-valued balanced trade is excluded,
displayed in Panel 2 of Table 11, it can be shown that the trade effects in the respective
transport markets vary by only 25%(η > 0.5) to 15%(η > 0.8) depending on the sample
restriction. Importantly, these results indicate that the variations in policy outcomes
further decrease as trade becomes more balanced. While these findings provide suggestive
evidence in support of the theoretical predictions, they lack in statistical significance.
To strengthen this empirical evidence, I relax the previous sample restrictions and
consider the variation in trade effects across all levels of bilateral trade imbalances via
the estimation of marginal effects. These marginal treatment effects are derived from
an ordinary least squares estimation including the interaction of the treatment indicator
with a measure of bilateral U.S. trade imbalances. Figure 5 displays the fronthaul and
backhaul marginal trade effects, along with the respective 90% confidence intervals, of
the low sulfur fuel requirement. The graphs are consistent with the previous results and
reveal negative marginal backhaul trade effects that become statically significant at the
10% level when the difference between fronthaul and backhaul trade is no larger than
70%(=1-Trade Balance). In contrast, the estimated marginal fronthaul treatment effects
are largely insignificant at any conventional level of significance. More importantly, the
graphs illustrate that the point estimates of marginal fronthaul and backhaul trade effects
are converging as bilateral trade becomes more balanced and are nearly indistinguishable
at a trade balance of 90% to 100%. These findings further substantiate the theoretical
results.
114
Product Level Results
In addition to the results obtained from the aggregate data analysis, evidence
of heterogeneous policy outcomes at the product level may be of considerable interest
to policy-makers as well. To give insight into the theoretically suggested variation of
trade policy outcomes across various product groups, I re-estimate the empirical model
using disaggregated data that combines two digit HS code level products43 according to
their average ad valorem transport costs.44 Dividing the data into product groups with
above and below average ad valorem transport cost yields empirical results that support
the initial findings at the aggregate level. That is, the low sulfur fuel requirement is a
larger deterrent to U.S.-EU trade facilitated in backhaul rather than fronthaul transport
markets. More importantly however, the analysis provides novel evidence in support of
Proposition 2 showing that the trade effects from maritime transport policy are, indeed,
increasing in ad valorem transport costs.
Recent research by the working party of the OECD Trade Committee has shown
that ad valorem transport costs exhibit large variation at the product group level and
averaged 6.7% for the top twenty product groups of U.S. imports from China, for example
(Korinek, 2011). In fact, the author reveals that ad valorem transport costs range from
3.7% to 15.7% for the most traded product groups, but also points out that some specific
goods are subject to much higher rates. Variation in these ad valorem transport costs
across products stems not only from differences in the unit value of each product, but
also from differences in container capacity utilization. Meaning, bulky products, such
as assembled furniture, or heavy goods, like wood and metal for example, cannot take
43These codes coincide with the product category levels as defined in the Harmonized Tariff Schedule
(HTS) that is administered by the U.S. International Trade Commission, (USITC).
44In the given context, ad valorem transport costs can be thought of as the ratio of transport cost
relative to associated container values in terms of percentages.
115
−
.
4
−
.
2
0
.
2
Ef
fe
ct
s 
on
 P
re
di
ct
io
ns
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
Trade Balance
Fronthaul Backhaul
FIGURE 5. Marginal Treatment Effects by Transport Market and Trade Imbalance
advantage of the low cost transport capacity provided by containers and thus, hold a
low per-container value. In contrast, light products, such as clothing or footwear, or high
value added products, such as electrical or mechanical machinery for example, can achieve
higher per-container values and thus, lower ad valorem transport costs by maximizing
container capacity utilization.
For the given analysis in this study, trade is disaggregated into two product
groups with either high or low ad valorem transport costs where the threshold is set at
6.7%, the average of the ad valorem transport cost data presented by Korinek (2011).45
The categorization of products also follows Korinek (2011) and is therefore, limited to
45The robustness of the results to variations in this threshold ranging from 6% to 8% has been evaluated
and, in general, the disaggregated results are consistent across these variations.
116
twenty-five product groups.46 The theoretical hypothesis to be tested is summarized in
Proposition 2 and states that an identical shock to carrier costs is expected to have a
larger effect on products with high ad valorem transport costs compared to those with
low ad valorem transport costs.
The primary PPML estimation results of the disaggregated analysis are presented
in Table 12 and display the differentiated average treatment effects of the low sulfur fuel
requirement on the low and high ad valorem transport cost product groups, respectively.
In general, the estimated trade effects for low ad valorem transport cost products, with
the exception of U.S. imports from EU members, are small and statistically insignificant.
In contrast, with the exception of U.S. exports to EU members, the EU Directive has
large, negative, and statistically significantly different treatment effects on all U.S.-EU
trade in products with high ad valorem transport costs. Intuitively, this implies that equal
increases in freight rates due to the low sulfur fuel requirement are, in fact, more taxing on
bulky and heavy products that have lower total container values, than small or high value
added products that command relatively high total container values.
Moreover, the disaggregated estimates reveal that fronthaul U.S.-EU trade of high
ad valorem transport cost products experiences a statistically significant decline of 7.72%.
This provides additional evidence that the low sulfur fuel requirement did, in fact, raise
marginal access costs for both U.S. exports and imports to and from EU members and
has led to a decline in U.S.-EU trade facilitated in both fronthaul and backhaul transport
markets. A comparison of the fronthaul and backhaul high ad valorem transport cost
product treatment effects, given in columns (4) and (5) of Table 12, illustrates that U.S.-
46Product groups with high ad valorem transport costs exceeding 6.7% include Plastics (39), Rubber
(40), Wood (44), Paper and related articles (47-49), Ceramic products (69), Iron and Steel (72-73),
Miscellaneous articles of base metal (83), Vehicles (87) and Furniture (94), whereas product groups with
ad valorem transport costs below 6.7% include Organic chemicals (29), Articles of leather, etc. (41-42),
Knitted clothing (61), Non-knitted clothing (62), Other textile articles (63), Footwear (64), Tools (82),
Mechanical machinery (84), Electrical machinery (85), Photo/Cinema equipment (90) and Toys (95).
117
TABLE 12. ATE - Disaggregated Product Group Level
(1) (2) (3) (4) (5)
VARIABLES Full Sample Imports Exports Fronthaul Backhaul
ATE, (δ) (low ad 0.047 0.098* -0.061 0.049 -0.013
valorem transport costs) (0.045) (0.052) (0.052) (0.052) (0.059)
Differential ATE (high ad -0.122*** -0.136** -0.073 -0.129** -0.185***
valorem transport costs) (0.045) (0.060) (0.061) (0.055) (0.063)
Port FE Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
EU trade of these products experienced a larger decline in backhaul transport markets.
In fact, the point estimates of column (5) show that backhaul U.S.-EU trade declined
by 17.93% in response to the EU Directive relative to the previously indicated 7.72%
decline of fronthaul U.S.-EU trade. This finding provides additional evidence in support of
Proposition 1 that trade facilitated in fronthaul markets exhibits a smaller elasticity with
respect to marginal access costs than trade facilitated in backhaul transport markets.47
In summary, all of these empirical findings compliment the theoretical results that the
elasticity of trade with respect to marginal access costs is, indeed, increasing in ad valorem
transport costs, heterogeneous across fronthaul and backhaul transport markets and
displaying vanishing transport market differences for rather balanced bilateral trade
observations.
47All of these results are largely insensitive to variations concerning the dividing threshold. That is,
empirical findings obtained when the data are restricted to products with the highest and lowest 25% ad
valorem transport costs are qualitatively similar to those presented in Table 12.
118
Robustness Analyses
To test the consistency of the primary empirical findings presented above, a
multitude of robustness checks have been performed. These analyses include alterations
in the DID specification, variations in standard error clustering, alternative fixed
effects specifications, variations in sample restrictions, changes in the empirical model
specification and alterations concerning the backhaul identification. The respective results
are reported in Tables 23-29 of Appendix C.48
DID Specifications
At a fundamental level, the first robustness check tests the appropriateness of the
DID estimator. Panels 2 and 3 of Table 23 present the results obtained from the inclusion
of treatment and control group as well as country-specific time trends, respectively.
The inclusion of these time trends in the estimation tests the appropriateness of the
parallel paths assumption. The obtained results are consistent across the various trend
specifications and support the use of the DID estimator.
Clustering
Through the second robustness analysis, I scrutinize the statistical significance of
the primary empirical results re-estimating the model using various levels of clustered
standard errors. While the statistical significance of the primary results is based on
standard errors clustered at the the route specific U.S. port-foreign country level, this
robustness analysis involves clustered standard errors at the U.S. state-foreign country,
U.S. port and U.S. state levels. These changes in the level of cluster aggregation explore
potential cluster correlations across ports and foreign trade partners. The respective
48For the ease of comparison, each of these tables first reports the primary empirical findings.
119
results are reported in columns 3 through 8 of Table 24 and illustrate that the statistical
significance of the primary results is generally robust to these estimation adjustments.
Fixed Effects
Following standard practices in the trade literature, I also test the robustness of the
empirical results against the inclusion of alternative sets of fixed effects. Evaluating the
importance of time-varying unobservable trade costs at a more local level, I incrementally
include time-varying fixed effects at the U.S. region, state and port level. The respective
results are reported in columns 3 through 8 of Table 25 and point to relatively stable
coefficient estimates and statistical significance in fronthaul and backhaul transport
markets.
Sample Restrictions
While alternative fixed effects and cluster specifications are common practice in a
variety of empirical applications, the following robustness checks are rather specific to this
study. In Panels 2 through 4 of Table 26, I report the estimation results obtained from
various alterations of the sample restrictions. In particular, I test whether the inclusions of
landlocked European countries as well as the North American U.S. trade partners, Canada
and Mexico, or exclusion of market dominant foreign countries, such as China and Japan,
alter the primary finding of heterogeneous trade policy outcomes. The corresponding
results illustrate that none of these exclusions drive the primary empirical findings and
ease the concerns of potential sample selection bias.
Model Alterations
In addition to these sample restrictions, I also investigate the sensitivity of the
empirical results to alterations of the empirical model specification and aggregation of
120
the data to quarterly rather than monthly frequency. Panels 2 and 3 of Table 27 report
the empirical findings obtained from estimations excluding U.S. state level employment as
well as U.S. free trade agreements, respectively. The results indicate that the estimated
treatment effects and their statistical significance are not reliant on the inclusion of these
control variables. In panel 4 of Table 27, I present the results obtained from the inclusion
of six month lagged explanatory variables. Once again, the coefficient estimates and their
statistical significance are robust and do not depend on the inclusion of these lagged
control variables.
Contrary to these model alterations, Panel 5 of Table 27 provides the estimation
results based on aggregated data at quarterly, rather than monthly frequency. This
aggregation is used to evaluate the potential issues arising from the lumpiness of trade
(see, for example, Hornok et al., 2011) that may obscure the precise estimation of average
treatment effects. The reported results for the quarterly aggregated data demonstrate
that the negative average treatment effects of the low sulfur fuel requirement vary by only
a small margin compared to primary results based on monthly data and continue to be
statistically significant at a 5% level for estimations involving U.S. exports and U.S. trade
facilitated in backhaul transport markets.
Backhaul Identification
Lastly, I explore whether the empirical results are consistent across various changes
to the identification of fronthaul and backhaul transport markets. These changes are
intended to capture alternative transportation patterns of bilateral U.S. trade and are
performed along the time and cross-sectional dimensions of the data. Given the fact
that transatlantic as well as transpacific shipping between the U.S. and Asian as well as
European countries can take anywhere from eight to over thirty days depending on the
121
port of origin and destination, the original identification of backhaul markets based on
current month bilateral trade comparisons may be distorted. To address this concern, I
redefine a backhaul transport market comparing current month’s U.S. imports(exports)
with last month’s U.S. exports(imports) for the same U.S. port-foreign country pair.
Estimations based on these redefined fronthaul and backhaul transport patterns yields
robust results both in terms of the magnitude and statistical significanceF and are
reported in Table 28.
In addition to these variations in shipment durations, it is possible that actual
transatlantic and transpacific shipping routes are more complex than the simple U.S.
port-foreign country trade observations available in the data used for this study. General
shipping patterns, in fact, suggest that most of the bilateral U.S. trade considered in
this study is handled by only a few major ports located in the U.S. and other foreign
countries in Asia and Europe. Of course, this added layer of complexity may mask the
actual fronthaul and backhaul shipment structure. To capture the potential changes in
fronthaul/backhaul patterns created by these multi-stop shipping routes, I aggregate
the data at the international supranational, U.S. state-supranational and U.S. region-
supranational levels and re-estimate the model. In general, the results presented in
Panels 2 through 4 of Table 29 reflect point estimates of similar magnitude relative to the
primary results given in Panel 1. While the statistical significance of these estimates varies
for U.S. exports, backhaul transport markets continue to reflect a trade reduction that is
statistically significant at either the 10% or 5% level. Interestingly, the potentially most
realistic representation of actual shipping routes, aggregating the data at the U.S. region-
supranational level, given in Panel 4, shows statistically significant trade reductions for
overall trade and fronthaul transport markets, in addition to the consistently significant
estimates for exports and backhaul transport markets.
122
Implications
In summary, the empirical analyses conducted in this study provide novel and
consistent evidence in support of the theoretically suggested heterogeneity in trade policy
outcomes. While all of the aggregate findings provide evidence in favor of Proposition
1, the results based on disaggregated product level data substantiate the theoretical
hypothesis manifested in Proposition 2. That is, when trade is imbalanced, trade policy
reducing marginal access costs exhibits treatment effects that are larger in backhaul
relative to fronthaul transport markets and increasing in ad valorem transport costs.
Overall, these findings accentuate the relevance of the international transport sector to
the determination of trade and point to potentially large differences in policy outcomes
pertaining to U.S. containerized trade. That is, commercial policy, such as infrastructure
investments or deregulation of customs directives that lead to lower marginal access costs,
may be much more effective for U.S. containerized exports than imports, which are mainly
facilitated in backhaul, rather than fronthaul transport markets.
The implications of this study, however, are not only domestic in scope. As the
results presented in Tables 10 through 12 as well as Figure 5 suggest, variations in
transport-related policy outcomes are closely connected to bilateral trade imbalances and
the overall trade composition. Merchandise trade statistics, obtained from the World Bank
Database and presented in Figures 6a and 6b, reveal that aggregated trade imbalances
exhibit large and fairly permanent fluctuations across high to low income countries,
yet they exhibit dynamic changes across geographic regions. More specifically, Figure
6a shows that trade imbalances experienced by low income countries are significantly
and consistently larger than those displayed by middle to high income countries. In
conjunction with the empirical result that the difference in trade effects is increasing in
the trade imbalance, this observation suggests that low income countries are exposed to
123
−
.
3
−
.
2
−
.
1
0
.
1
Tr
ad
e 
Im
ba
la
nc
e
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
High Income Middle Income
Low Income
(a) Income Level
−
.
2
−
.
1
0
.
1
Tr
ad
e 
Im
ba
la
nc
e
2013 2014 2015
East Asia & Pacific Europe & Central Asia
Latin America & Caribbean North America
South Asia Sub−Saharan Africa
Middle East & Northern Africa
(b) Geogrpahic Region
FIGURE 6. Aggregate Trade Imbalance by Income Level and Geographic Region
much larger export and import volatility considering transport-related policy outcomes
compared to middle to high income countries. Moreover, Figure 6a shows that low income
countries tend to be net importers suggesting that exports of these countries are mainly
facilitated in backhaul transport markets. Based on the empirical results, exports of low
income countries are therefore subject to disproportionate trade effects from transport-
related policies that affect carrier access costs - a conjecture that emphasizes both the
potential and risk of transport-related policies in developing countries.
In contrast, Figure 6b offers a geographical comparison of aggregated trade
imbalances. The data demonstrate stark geographical differences that are subject to
considerable short-run fluctuations for some regions. Sub-Saharan Africa, for example,
progresses from a slight net exporting region in 2013 to a considerable net importing
region by 2015. Given the empirical findings, this observation suggests dynamic variation
in transport-related trade effects that add significant complexities concerning the
effectiveness of commercial and maritime policies.
Complementing the differences in the patterns of trade, trade composition displays
considerable geographic variation as well. As indicated by Figures 7a and 7b, the
124
0
5
10
15
20
Sh
ar
e 
of
 E
xp
or
ts
(%
)
2013 2014 2015
East Asia & Pacific Europe & Central Asia
Latin America & Caribbean North America
South Asia Sub−Saharan Africa
(a) Export
0
2
4
6
8
10
Sh
ar
e 
of
 Im
po
rts
 (%
)
2013 2014 2015
East Asia & Pacific Europe & Central Asia
Latin America & Caribbean North America
South Asia Sub−Saharan Africa
(b) Import
FIGURE 7. Export and Import Share of Ore and Metal by Geographic Region
shares of heavy ore and metal relative to overall merchandise exports and imports are
asymmetrically distributed across various global regions. While Sub-Saharan Africa
and Latin America as well as the Caribbean display considerable export shares in
bulky products, such as ore and metal, South and East Asia as well as the Pacific
region command larger import shares of these products. Based on the qualitative and
quantitative evidence that high ad valorem transport cost products display larger trade
effects in response to transport-related policies, this composition of trade suggests
disproportionate export volatility for the former regions and disproportionate import
fluctuations for the latter. The suggested influence of the patters and composition of
trade on policy effectiveness via the channel of international transportation raises a host
of research questions with considerable merit for economic development and pertaining to
a variety of international policies.
Conclusion
In this study, I extend a model of international trade by integrating a transport
sector that subsumes the key feature of joint round trip production present in the
125
international container shipping industry and allows for asymmetric and integrated
bilateral transport costs to be endogenously determined in equilibrium. Given this
theoretical model, I demonstrate that trade policy implications may vary across
different types of trade flows and develop two specific propositions about this systematic
heterogeneity of trade policy outcomes.
The empirical findings presented in this study provide supporting evidence for
these theoretical hypotheses. Based on the results, I conclude that the low sulfur fuel
requirement enacted as part of EU Directive 2005/33/EC raised marginal access costs
for both U.S.-EU exports and imports and caused a significant reduction in U.S.-
EU containerized trade facilitated in backhaul transport markets; a finding that is
consistent across a host of robustness analyses, including trade imbalance variations, data
aggregation and various estimation sample restrictions, among others. In contrast, the
response of U.S.-EU trade facilitated in fronthaul transport markets has been markedly
smaller and becomes statistically significant only when more balanced trade cases or high
ad valorem transport cost products are considered. Further differentiation of the data
reveals that this heterogeneity in treatment effects is decreasing as trade becomes more
balanced. In conjunction with aggregate data on the patterns of international trade, these
findings suggests that the effects of trade policy are rather symmetric for middle to high
income countries with nearly balanced trade but rather volatile for low income countries
with considerable merchandise trade deficits.
The additional analysis conducted at a more disaggregated product group level
suggests that responses to trade or environmental policy, such as the low sulfur fuel
requirement, are not only asymmetric across fronthaul and backhaul transport markets,
but also idiosyncratic across product groups. The results show that U.S.-EU trade in
products subject to higher relative transportation costs, such as plastics, metals, vehicles
126
or furniture, is more responsive to changes in carrier costs than U.S.-EU trade of product
groups with comparably low relative transportation costs, such as apparel and footwear, or
electrical and mechanical machinery.
Overall, these findings point to the relevance of the international transport sector in
the determination of the patterns and composition of trade. The differences concerning
the estimated trade effects provide supporting evidence for the theoretically suggested
heterogeneity in commercial and related policy outcomes and identify the backhaul
problem as the source of this variation. Naturally, these findings have considerable
implications for U.S. policy intended to stimulate U.S. exports or imports in general
and are of particular importance when specific products or bilateral trade relations are
targeted. However, the results also point to the fact that foreign maritime and related
policy can have negative externalities for U.S. trade as well. This feature of the analysis
demonstrates that commercial policy intended to stimulate U.S. trade, via carrier cost
reductions, must not be limited to the domestic and unilateral scope, but could involve
international multilateral efforts.
Future research may focus on identifying the varying responses of trade to changes
in marginal joint costs or test whether policy outcomes vary for alternative commercial
policies, such as preferential trade agreements. These inquiries could further delineate
between the effects on individual products or differences in policy implications across
developing and developed countries.
Bridge
Similar to the analysis conducted in this chapter, which has focused on the variation
in trade effects across various transport markets, products and trade imbalances, the
research provided in the following chapter considers trade disruptions and their variations
127
across U.S. ports. In following chapter, however, I investigate the response of trade to a
natural disaster, rather than a policy induced change in carrier costs. Again, the results
point to systematic variation in trade effects. The dimension of this variation, however, is
spatial in nature and dynamic over time.
128
CHAPTER IV
THE RESILIENCE OF INTERNATIONAL TRADE: AN EMPIRICAL EXAMINATION
OF THE DYNAMIC SPATIAL TRADE EFFECTS OF NATURAL DISASTERS
Introduction
Natural disasters pose a constant threat to human life and economic activity, as
evidenced by the recent earthquakes in Japan, Ecuador and Italy. According to the
Centre for Research on the Epidemiology of Disasters (CRED), the global community has
experienced an average of 384 natural disasters per year, over the last decade. As a result,
close to 200 million people have been victimized yearly and annual average damages are
estimated around $162 billion and increasing over time (CRED, 2015).
A variety of case studies indicate that this devastation not only encompasses the
tragic loss of human life, but also the impairment of entire regional economic structures
(Vigdor, 2008; Grenzeback et al., 2008). Upon the strike of a largely unanticipated
natural disaster, housing, employment, and infrastructure, among others, are found
to be in complete disarray. As international trade has grown and gained economic
significance, its global presence has exposed it to the destruction and tragedy originating
from natural disasters. The displacement of workers, destruction of product and capital,
and impairment of infrastructure paramount to the facilitation of international trade
can lead to substantial delays and/or rerouting of traded products. As such, natural
disasters represent infrequent and uncertain trade costs, but yet have the potential to
be immensely taxing, particularly at the local level. While the majority of commercial
policies and academic research on trade is geared towards more common trade barriers,
such as tariffs and transportation costs, relatively little attention has been paid to the
129
linkages between trade and natural disasters. The existing studies suggest that natural
disasters cause relatively small and short lived disruptions of aggregate international trade
and heterogeneous responses across countries, industries, firms and products (Gassebner
et al., 2010; Oh and Reuveny, 2010; Ando and Kimura, 2012; Martincus and Blyde, 2013).
In this study, I build on this small strand of the economic literature and provide a
novel analysis of the natural disaster induced trade effects at the regional and port level.
To this end, I build on the traditional gravity model of trade and extend it to capture the
dynamic and spatial variation of natural disaster induced trade effects and account for the
spatial correlation of local trade flows. The resulting spatial econometric model is applied
to U.S. port level trade data from August 2003 to August 2013. Identification of the
spatial heterogeneity in natural disaster induced trade effects is based on the exogenous
variation in trade caused by Hurricane Katrina. Evaluations of the treatment effects at the
aggregate or regional level produce statistically insignificant estimates that are consistent
with the findings of the previous trade literature (Parsons, 2014).
In contrast to these aggregate estimates, I also investigate trade disruptions at
the disaggregated local port level. The empirical results provide novel evidence of
statistically significant trade disruptions. The port-specific treatment effects vary over
time and strongly depend on the port’s distance to Katrina’s epicenter as well as various
characteristics, such as harbor type and entry restrictions. While ports of New Orleans,
Louisiana, and Gulfport, Mississippi, exhibit economically and statistically significant
reductions in trade, the nearest ports of Mobile, Alabama, and Panama City, Florida,
experience substantial and statistically significant increases in trade; a finding that
is true both across the value of trade and the number of traded products. Regardless
of whether the positive or negative trade disruptions are considered, I find that the
impact of Hurricane Katrina exponentially vanishes as distance to its epicenter rises.
130
As expected, disruptions in exports are estimated to be more sensitive to distance than
those of imports. Driving the resilience of international trade, this spatial distribution of
counteracting trade effects leads to profound short-run disruptions at the local port level,
but negligible effects in aggregate.
In addition to this static spatial analysis, I also consider the dynamic changes
of these natural disaster induced trade effects over time. The case study of Hurricane
Katrina provides important insights into the evolution of the spatially heterogeneous trade
effects. Differentiating the monthly impacts across first to sixth order contiguous ports, I
find that the duration of the experienced trade effects is largely port specific. While some
ports recover fairly quickly, others are estimated to have experienced a permanent long-
run change to their trade volumes and growth. Trade facilitated through the port of New
Orleans, for example, recovers from the drastic short-run reductions within the first one to
two years following treatment, while the port of Gulfport exhibits only a partial recovery
that results in permanent reductions of both exports and imports relative to pre Hurricane
Katrina levels. The adjacent ports reflect similar heterogeneous dynamics. While the port
of Panama City experiences an immediate and persistent increase in trade, the port of
Mobile shows a long-run increase in the growth rates of trade triggered by Hurricane
Katrina.
The estimation of these spatially heterogeneous and dynamic trade effects delivers
novel evidence in support of the static and dynamic resilience of international trade as
defined by Rose (2007). In his study, the author describes static resilience as the ability
of the economic system to maximize output based on the remaining, after-shock resources
and dynamic resilience as the speed of recovery of the economy post natural disasters.1
1Rose and Wei (2013) develop an input-output type model to simulate the macroeconomic effects of a
hypothetical port shutdown and illustrate several mechanisms of resiliency, most important of which is the
ability to reroute international and domestic trade.
131
Based on these definitions and the empirical results obtained in this study, the static
resilience of trade is founded in the ability of international carriers to use the remaining
local infrastructure to provide alternative channels of trade facilitation - a finding that is
of particular interest to policy makers in developing countries which continue to experience
significant reductions in output growth due to natural disasters. The dynamic resilience
of trade is shown to be driven by port-specific recovery as well as permanent port choice
alterations.
Conducting these analyses, my research contributes to the existing literature in
several ways. To the best of my knowledge, this study is the first to identify the spatial
heterogeneity of natural disaster induced trade effects and consider their short-run and
long-run spatial distribution. The empirical findings offer insights into the dynamic
response of the international transport sector and domestic infrastructure network to local
trade disruptions and point to the importance of these mechanisms in mitigating aggregate
repercussions. As such, the estimated spatial and dynamic variation in trade effects
presented in this study provide supporting evidence of the static and dynamic resilience
of international trade and identify the specific channels that empower this pliancy.
The remaining parts of the paper are organized as follows: Section 2 presents an
overview of the evolution of natural disasters and their devastating consequences, while
also providing detailed background information on Hurricane Katrina and its specific
effects. Section 3 offers a literature review focused on research pertaining to trade cost, its
linkages to natural disasters and the resulting trade disruptions. To analyze the dynamic
spatial variation in trade effects, a theoretical gravity model is developed in Section 4 and
the resulting empirical specification is presented in section 5. The U.S. port level trade
data employed in this study are summarized in Section 6, while the empirical results are
132
discussed in section 7. Section 8 concludes this study and points to the significance of the
empirical results as well as areas of further inquiry.
Institutional Background
As Blonigen and Wilson (2013) point out, international trade has been growing for
decades and has exhibited a growth rate much larger than that of world Gross Domestic
Product (GDP). The growing importance and global presence of international trade have
led to its exposure to the destruction and challenges originating from a host of natural
disasters in all corners of the world. According to CRED, natural disasters frequently
occur across all continents and cause significant human losses and economic damages.
In fact, CRED reports that over the last 20 years 6,457 weather-related disasters were
recorded worldwide and that these natural disasters have claimed over 600,000 lives in
total (CRED, 2015). Table 13 is based on the data presented in CRED’s annual disaster
statistical report of 2014 and provides continental averages concerning the frequency,
number of overall victims and economic damages caused by all types of natural disasters
over the time period from 2004 through 2013.
These data demonstrate that natural disasters are, indeed, frequent and global
events. More importantly, the statistics show that recent natural disasters have caused
substantial human and economic losses with roughly 200 million people affected annually
and economic damages reaching a staggering $162 billion per year, on average. However,
the data presented in Table 13 also reveal that the human and economic impacts of
natural disasters vary greatly across continents. While an annual average of 69 natural
disasters affected over 27 million people in Africa, natural disasters of similar average
frequency in America and Europe affected only 9.82 and 0.64 million people, respectively.
In contrast, average annual economic damages due to these natural disasters range from
133
$67.97 billion in America and $13.45 billion in Europe to $0.58 billion in Africa. Out of all
continents, Asia is most affected with an annual average of over 160 million victims and
over $75 billion in economic damages caused by an average 156 natural disasters per year
(Guha-Sapir et al., 2015).
TABLE 13. Average Continental Disaster Impact
Continent Frequency Victims (mil.) Damages (2014 $ bil.)
Africa 69 27.86 0.58
America 91 9.82 67.97
Asia 156 160.71 75.27
Europe 54 0.64 13.45
Oceania 14 0.19 5.26
Global 384 199.23 162.53
The data presented mark the 2004 through 2013 averages across all types of disasters.
Sources: Centre for Research on the Epidemiology of Disasters, Annual Disaster
Statistical Review 2014
In addition to these average impacts of natural disasters, their historical trends in
frequency and the evolution of the resulting human losses and economic damages are of
considerable interest as well. When considering these dynamic developments of natural
disasters, several key aspects, such as changes in exposure or destructive force, come
into play. While Kunkel et al. (1999) report that recent demographic trends have led
to an increasing population and property density in heavily disaster stricken regions,
research by Emanuel (2005) points out that the power dissipation of tropical cyclones,
for example, has doubled over the last century. Data collected by CRED and published
in the International Disaster Database (EM-DAT), give insight into the efficacy of these
observations. Figures 8a-8d display the annual global frequency of natural disasters and
the resulting global economic damages, the overall number of victims affected and the
number of deaths caused by these catastrophes.
134
Figure 8a shows that the period from 1960 to about 2000 saw a sharp ten-fold
increase in natural disasters, whereas the more recent history suggests a downward trend
concerning their frequency. Matching the historical increase in natural disaster frequency,
Figures 8b and 8c illustrate that annual global economic damages and the number of
people affected by these disasters are also increasing over the sample period from 1960 to
2000. However, these positive trends are more gradual and exhibit much larger volatility
compared to the steep and rather smooth increase concerning the frequency of natural
disasters. The recent reduction in disaster frequency is accompanied by a decline in the
number of affected people, whereas economic damages do not reflect this downturn. This
finding supports the arguments made by Kunkel et al. (1999) and Emanuel (2005) that
even less frequent disasters can cause significant overall losses due to increases in economic
vulnerability and a rise in the destructive force of the most recent natural disasters.
Despite these dispiriting findings, the lethality of natural disasters, depicted in Figure 8d,
encouragingly does not match the historic rise in their frequency, but appears to be rather
disaster-specific instead.
The combination of larger populations subjecting themselves to the potential havoc
of natural disasters, the growth of and increasing dependence on international trade and
the rise in the destructive force of these natural disasters suggest potentially intensifying
disruptions of international trade and supply chains. While some empirical studies
consider the average effect of natural disaster on aggregate trade, this study identifies
the dynamic and spatially heterogeneous trade effects via the variation caused by a
single event, Hurricane Katrina. Hurricane Katrina is widely recognized for its immense
devastation that caused tremendous hardship in human life and economic outcomes.
According to Grenzeback et al. (2008), Katrina was the costliest and most destructive
135
0
10
0
20
0
30
0
40
0
50
0
G
lo
ba
l F
re
qu
en
cy
1960 1970 1980 1990 2000 2010 2020
year
(a) Frequency
0
10
00
00
20
00
00
30
00
00
40
00
00
G
lo
ba
l E
co
no
m
ic 
Da
m
ag
e 
(in
 $ 
mi
l.)
1960 1970 1980 1990 2000 2010 2020
year
(b) Damages
0
20
00
00
40
00
00
60
00
00
80
00
00
G
lo
ba
lly
 A
ffe
ct
ed
 V
ict
im
s 
(in
 10
00
)
1960 1970 1980 1990 2000 2010 2020
year
(c) Reach
0
50
00
00
10
00
00
0
15
00
00
0
G
lo
ba
l D
ea
th
s
1960 1970 1980 1990 2000 2010 2020
year
(d) Lethality
FIGURE 8. Global Disaster Trends (1960-2015)
natural disaster ever experienced by the U.S.2 causing over 1,800 deaths and an estimated
$149 billion in direct and indirect economic losses (Hallegatte, 2008).
As depicted in Figure 9, Hurricane Katrina originated around the Bahamas and
made its first landfall as a Category 1 hurricane in Florida on August 25th of 2005. After
causing modest disruptions in Florida, the windstorm moved to the Gulf of Mexico, where
it rapidly intensified and developed into a Category 5 hurricane at its peak. Its second
landfall occurred in the state of Louisiana on August 29th, 2005, with sustained winds of
2While Grenzeback and Lukman’s assessment is based on nominal values, Pielke Jr et al. (2008) show
that in normalized terms Hurricane Katrina actually caused the second largest losses in U.S. history
behind the Great Miami storm of 1926.
136
FIGURE 9. Geographical Movement and Strength of Hurricane Katrina
125 mph. Upon this second landfall, the havoc caused by Hurricane Katrina was felt along
the majority of the U.S. Gulf coast severely affecting the coastal regions of Louisiana and
Mississippi.
In addition to the tragic loss of human life, Hurricane Katrina’s wreckage extended
across the entire regional economic structure and even to the national level. Based on
early estimations, Holtz-Eakin (2005) argued that the effects of Hurricane Katrina were
expected to lower U.S. output growth by 0.5 percentage points in the short-run, whereas
recovery efforts were expected to reverse this effect by 2006. The underlying causes for
this initial decline in the growth of aggregate income range from extensive reductions
137
in employment and housing due to the flooding of New Orleans3, the destruction of
physical capital4, and trade disruptions caused by the severe impairment of the regional
infrastructure. According to Grenzeback et al. (2008), the infrastructure of the U.S.
coastal region of Louisiana, Mississippi and Alabama experienced substantial ruination
encompassing damages to road, rail and port networks. Taking account of the specific
damages, the authors point to the destruction of bridges, specific railways, the ports
of New Orleans and Gulfport and Interstate 10, as well as the loss of electricity and
closing of major waterways as the main factors determining this wreckage of the coastal
infrastructure and potentially causing severe local trade disruptions.
Despite this detriment to the regional infrastructure, Parsons (2014) finds that
aggregate U.S. imports were unaffected by the destructive force of Hurricane Katrina in
the long-run. Upon providing supporting evidence of this aggregate finding, I evaluate
the significance of trade disruptions at the regional and local levels. I find statistically
significant local trade effects that are offsetting in aggregate and evidence the substantial
resilience of international trade to natural disasters, even to those as monumental as
Hurricane Katrina.
Literature Review
Within the international economics literature it has been widely recognized that
trade costs are an integral determinant of international trade. In fact, regardless of
3Studies by Dolfman et al. (2007) and Vigdor (2008) show that Hurricane Katrina resulted in
significant reductions of employers and employment (between 70,000 and 95,000 lost jobs) and the long-
term displacement of over 150,000 people. Elliott and Pais (2006) as well as Masozera et al. (2007) find
significant heterogeneity across the individually experienced losses and illustrate that this heterogeneity
systematically varied by socioeconomic factors, such as race, class and income. In addition to these labor
market effects, Vigdor (2008) reports that the New Orleans’ availability of housing declined from 215,000
units in 2000 to 106,000 units in the aftermath of Hurricane Katrina.
4Holtz-Eakin (2005) estimate physical capital damages to total between $70 billion and $130 billion.
138
whether Krugman’s ’New Trade Theory’ (Krugman, 1980), the ’New-New Trade Theory’
initially introduced by Melitz (2003), the gravity model (see, for example, Anderson
and van Wincoop, 2003) or recent work on the classical model by Deardorff (2014) is
considered, most theoretical derivations point to the significance of trade costs in the
determination of the level and composition of international trade. In this section, I present
a literature review concerning the micro- and macroeconomic effects of natural disasters,
linkages to trade costs and resulting trade disruptions.
Trade costs manifest themselves in variety of ways and many of their facets
have been analyzed in the trade literature. Brander and Spencer (1984), for example,
study the effects of tariffs on international trade, whereas Clausing (2001) and Frankel
and Rose (2002) focus on the impact of preferential trade agreements and monetary
unions, respectively. Other studies have considered the trade effects of national borders
(McCallum, 1995; Anderson and van Wincoop, 2003), cultural and linguistic differences
(Egger and Lassmann, 2012) or transportation costs (Hummels and Skiba, 2004; Hummels,
2007; Friedt and Wilson, 2015; Friedt, 2016), for example.
While these types of barriers are relatively constant factors in the determination of
international trade, trade disruptions caused by natural or man-made disasters represent
rather irregular elements of overall trade costs. Nevertheless, these events can have
significant long-term impacts on trade. Glick and Taylor (2010), Li and Sacko (2002),
or Anderton and Carter (2001), for example, study the impacts of war and militarized
conflict on international trade and find that the effects are generally long-lasting, lower the
level and growth of international trade, impose large externalities on impartial countries,
and vary by the level of uncertainty, duration and hostility. Alternative causes of man-
made trade disruptions include acts of terrorism (see, for example, Egger and Gassebner,
2015) as well as economic sanctions (Caruso, 2005). In general, the empirical evidence
139
concerning the trade effects of these alternative disasters varies by the severity of the event
and the time horizon under consideration. Despite the fact that there is a large volume
of studies providing theoretical and empirical analyses of the effects of trade costs on
international trade, the consequences of natural disasters on trade have received very little
attention.
Within the literature on natural disasters, many non-trade, microeconomic and
macroeconomic aspects have been considered. On the microeconomic side, these issues
include, for example, the natural disaster induced effects on labor markets (Belasen and
Polachek, 2008), housing prices (Hallstrom and Smith, 2005), consumption volatility
(Auffret, 2003) or supply chains (Altay and Ramirez, 2010). The general findings of
this strand of the literature provide evidence of large economic distortions that exhibit
significant heterogeneity across economic agents and over time. This variation in economic
impacts is a reoccurring theme throughout the natural disaster’s literature (see, for
example, a survery by Cavallo and Noy, 2009) and applies to macroeconomic and trade
related studies as well.
On the macroeconomic side, several studies have focused on primary issues, such
as the effect of natural disasters on inflation (Noy, 2009), financial flows (Rasmussen,
2004; Yang, 2008) or output growth. While Yang (2008) finds a statistically significant
increase in international financial flows for developing countries in response to natural
disasters, evidence concerning their impact on output growth is rather mixed. Skidmore
and Toya (2002), for example, find that a rise in the frequency of natural disasters causes
an increase in the growth of aggregate income due to the substitution of investment
towards human capital. Contrary to this finding, Strobl (2011) provides evidence of a
very localized increase in output growth that is canceled out at the state level and leads
to a negligible effect of natural disasters on the growth of aggregate output. In response to
140
this variation concerning the effects on output growth, the literature has turned towards
a more disaggregated analysis differentiating between developing and developed countries
(see, for example, Stro¨mberg, 2007; Crespo Cuaresma et al., 2008; Noy, 2009; Noy and Vu,
2010; Strobl, 2012) and identifying a variety of key factors determining the macroeconomic
impact of natural disasters. Kahn (2005), Toya and Skidmore (2007), Raschky (2008) and
Noy (2009), for example, show that countries with higher levels of democracy, government
stability and education, greater openness, a more complete financial system, better
investment climate and less inequality, in addition to higher income, experience fewer
losses from natural disasters, on average.5
Of course, a potential explanation for this variation in disaster-induced aggregate
economic outcomes may be the underlying heterogeneity of impacts at the regional level.
While research by Burrus Jr et al. (2002) illustrates that output, employment and indirect
business taxes in directly affected regions decline in response to hurricanes, Xiao and
Nilawar (2013) demonstrate that income and employment in directly neighboring regions
experience a short-run increase. Rose et al. (1997) and Lin et al. (2012) add that the
effects of natural disasters vary not only geographically, but also across local industries.
Although international trade and its disruptions due to natural disasters can act
as a significant catalyst for the discussed regional and aggregate variation in economic
outcomes, research in this area is very limited. In a seminal study, Gassebner et al. (2010)
analyze the effects of natural disasters on aggregate international trade. In general, the
authors find that governance and economic size matter to the degree of devastation
5The definition of losses varies across these studies. Kahn (2005), Toya and Skidmore (2007) and
Raschky (2008) measure the effects on deaths and damages due to natural disasters, whereas Noy (2009)
captures losses by estimating the effects on output growth via an interaction term with the disaster
variable. Although these studies yield very similar results overall, their conclusions pertaining to the
effects of governments vary. Specifically, Noy (2009) finds that larger governments dampen the reduction
in output growth, whereas Toya and Skidmore (2007) provide evidence indicating that larger governments
increase the lethality of disasters.
141
on the import side, while exports experience a negative shock regardless of country
characteristics. Research by Oh and Reuveny (2010) complements these findings showing
that political risk is another important factor in the determination of trade effects caused
by natural disasters. In contrast to these general analyses, studies by Andrade da Silva
and Cernat (2012) and Meng et al. (2015) distinguish between the trade effects of natural
disasters on developing versus developed countries. In general, the authors show that the
estimated trade effects vary by economic and geographical size of the affected country
as well as across imports and exports. Further disaggregation of these trade effects has
provided evidence that the resulting trade disruptions, in fact, vary across the time
horizons under consideration (Ando and Kimura, 2012; Parsons, 2014), the type of trade
flow (Chang, 2000), industries and firms (Ando and Kimura, 2012; Martincus and Blyde,
2013) as well as product groups (Martincus and Blyde, 2013) and can lead to a change in
trade composition (Ando and Kimura, 2012; Pelli and Tschopp, 2013).
Although this relatively small strand of the economic literature on natural disasters
has provided substantial insights into the variation of natural disaster induced impacts
on trade, little is known about the spatial heterogeneity of these trade effects.6 The
spatial econometric analysis presented in this study provides supporting evidence of the
static and dynamic resilience of international trade by estimating the short-run and long-
run spatial distribution of trade disruptions caused by Hurricane Katrina. The results
offer an intuitive explanation for the generally small or even positive aggregate effects
of natural disasters on trade and output. In addition, the empirical results point to
the significance of infrastructure networks to dampen the economic devastation caused
by natural disasters. Since, developing countries continue to experience reductions in
6The study by Martincus and Blyde (2013) exploits geo-referenced data on Chile to estimate the short-
run impact of local infrastructure disruptions on trade and is perhaps most closely related to the present
study.
142
output growth due to natural disasters, the specific resiliency channels identified in
this research are of particular interest to international policy makers and one of the
important remaining gaps in the literature as established by Cavallo and Noy (2009).
Based on these findings, my research contributes to both the trade and natural disaster
literature in several ways. First, to the best of my knowledge, this study is the first to
identify the spatial heterogeneity of natural disaster induced trade effects. Second, the
empirical findings offer insights into the dynamic responses of the international transport
sector and domestic infrastructure network to these local disruptions. Third, the results
concerning the spatial and dynamic variation in trade effects presented in this study
provide supporting evidence of the static and dynamic resilience of international trade
and identify the specific channels that empower this resiliency.
Theoretical Model
As the CRED data indicate, natural disasters inflict substantial damages to human
life and economic activity. However, the empirical findings in the literature suggest that,
while these devastations appear to be locally intensive in the short-run, they are rather
negligible at the national level and insignificant in the long-run (Parsons, 2014). To test
the validity of these findings with regards to international trade, I extend the standard
model of trade to allow for time-varying natural disaster induced trade effects at the local
port level. In particular, I follow the majority of the empirical trade literature and build
on the conventional gravity equation extending it to include a spatially varying trade cost
function that allows for the potential static and dynamic resilience of international trade
to be estimated.
To begin, I base these extensions on the multi-country general equilibrium model
derived by Anderson and van Wincoop (2003). To develop their model, the authors
143
assume that the global economy consists of J trading countries. Each country, j,
specializes in the production of a unique good which implies that goods are differentiated
by origin. The supply of each of these goods is fixed and consumers in each country
maximize utility based on their identical, homothetic, variety-loving CES preferences.
Grounded on these assumptions, the theoretical derivation by Anderson and van Wincoop
(2003) culminates in a structural gravity equation that expresses the value of country j’s
imports sourced from country i at time t, xijt, as a function of each countries aggregate
income, Yit and Yjt, relative to world income, Ywt, ad valorem trade costs, τijt and
multilateral resistance terms, Pjt and Πit. Specifically, Anderson and van Wincoop (2003)
present a static version of the following gravity equation:
xijt =
YitYjt
Ywt
(tijt)
1−σ
P 1−σjt Π
1−σ
it
, (4.1)
where the multilateral resistance terms are given by
P 1−σj =
∑
i
Yi
Y w
(
tij
Πi
)1−σ
(4.2a)
Π1−σi =
∑
j
Yj
Y w
(
tij
Pj
)1−σ
. (4.2b)
Following Egger and Gassebner (2015), I assume that ad valorem trade costs, τ 1−σijt ,
can be described by a multiplicative function of K individual elements, indexed by k,
and represented as τ 1−σijt = Π
K
k=1τ
1−σ
kijt . Further drawing on the notation introduced by
Egger and Gassebner (2015), the kth element of the multiplicative trade costs function is
captured by the kth observable trade-cost variable hkijt, such that τ
1−σ
kijt = h
βkt
kijt. To reflect
the spatial variation in trade costs due to natural disasters, I expand the dimension of the
trade cost function and introduce the spatial notation, p, to indicate the dependence of
144
trade costs on the location of the specific port of entry or exit, τ 1−σkpijt = h
βkpt
kpijt. Intuitively,
one of these trade costs elements, for example, captures the natural disaster induced trade
costs that are allowed to vary across ports at various distances from the given epicenter.
Specific to the present study, the kth trade-cost variable indicates the timing of
Hurricane Katrina for each of the given ports at various distances to its epicenter. The
β coefficient on this variable is intended to capture the resulting port-specific trade
effects. Comparisons of these coefficients across ports and time give insights into the
dynamic spatial heterogeneity of the disaster induced trade effects and provide a test for
the hypothesized resilience of international trade. Log-linearizing the resulting gravity
equation leads to the final theoretical expression of international trade flowing from
country i to country j through port p at time t.
ln(xijpt) = ln(Yit) + ln(Yjt)− ln(Ywt) +
K∑
k=1
βkptln(hkijpt)− ln(P 1−σjpt )− ln(Π1−σipt ). (4.3)
Empirical Model
Based on equation (4.3), I derive the stochastic specification of the previously
described gravity model accounting for the dynamic and spatially varying trade costs.
Moreover, I expand the econometric model to control for potential presence of spatial
autocorrelation across U.S. ports. For the purposes of this study and simplifying
the theoretical expression, I aggregate U.S. port-level trade across foreign countries.
The resulting dependent variable, the log of U.S. aggregate trade through port p at
time t (monthly frequency), ln(xpt), is modeled as a function of port-varying natural
disaster induced trade costs and various fixed effects that control for the aforementioned
multilateral resistance terms. More specifically, the set of fixed effects include port, ap, and
145
time, at, specific indicators that capture time-invariant port characteristics and national
trends in macroeconomic and trade cost variables, such as national income or changes
in free trade agreements and tariffs. The main trade cost component of interest includes
a set of interacted dummy variables, τp,t∗±s, each indicating a specific month before or
after Hurricane Katrina for a given port of entry or exit. The specific timing of Hurricane
Katrina’s landfall in August of 2005 is indicated via t∗. Intuitively, τp,t=t∗+1 indicates the
disaster induced trade disruption at port, p, one month following Hurricane Katrina’s
landfall.
Since U.S. port-level trade flows and the unobservables influencing these
transactions, upt, are potentially correlated across ports of entry and exit, I adopt the
flexible spatial econometric model presented by LeSage and Pace (2009). This specification
nests the spatial autoregression (SAR), spatial error (SEM) and spatial autocorrelation
(SAC) models which allow for a spatially correlated dependent variable, ρWln(xpt),
and/or error term, upt = λWupt + pt. Given this specification of the stochastic component,
pt is a normally and independently distributed random error, while ρWln(xpt) and λWupt
consist of the spatial correlation coefficients, ρ and λ, as well as spatial weight matrix W .
Of course, there are various weight matrices available when considering the final spatial
econometric specification (Ord, 1975; LeSage and Pace, 2009). Since the distances between
ports are non-uniform, a natural choice for the spatial weights may be an inverse distance
measure which proposes that the spatial correlation across ports declines exponentially
with the distance between a pair of ports (Cliff, 1969; Griffith, 1996; Getis, 2009). An
alternative may be a row normalized contiguity matrix indicating each port’s neighboring
facilities irrespective of their distance.7
7The empirical results presented in section 7 and Appendix C are consistent across all weight matrix
specifications including inverse distance based on nautical distances as well as contiguity matrices of order
one through four.
146
Combining this stochastic structure with the original gravity specification yields the
following empirical model;
ln(xpt) = β0 + ρWln(xpt) +
R∑
r=1
P∑
p=2
βprτp,t∗−r +
S∑
r=1
P∑
p=2
βpsτp,t∗+s + ap + at + upt (4.4)
upt = λWupt + pt,
where 2 years of port-specific pre-treatment, R = 24, and 8 years of port-specific post
treatment, S = 96, effects are considered. Given this specification, the SAR model is
obtained when ρ > 0 and λ = 0, while the SEM model is nested via ρ = 0 and λ > 0.
Alternatively, the SAC model assumes ρ > 0 and λ > 0. The parameters of interest
capturing the dynamic and spatially distributed trade effects caused by Hurricane Katrina
are given by βps. These parameters, along with pre-treatment indicators, βpr, are evaluated
in relation to the month of Hurricane Katrina’s landfall, t∗, and a given port of reference,
p = 1.
Data
The data used to estimate the empirical model specified by equation (4.4) have been
obtained from various sources. The main variable of interest is given by U.S. containerized
trade concerning both exports and imports at the container seaport of entry and exit level.
These data are available through the USA Trade Online database by the U.S. Census
Bureau and cover the entirety of U.S. bilateral trade facilitated through U.S. ports of
entry and exit at monthly frequency. The time period considered in this study extends
from August of 2003 to August of 2013. While USA Trade Online includes a variety of
ports with vastly different trade volumes, the selection of ports included in this analysis
is based on economic significance. That is, only the largest forty ports of entry and exit
147
have been included in the sample.8 At the time of Hurricane Katrina’s landfall, these forty
ports account for roughly 98% and 96% of total U.S. containerized imports and exports,
respectively.
The key variables of interest distinguishing the systematic variation in trade
disruptions caused by Hurricane Katrina are based on longitudinal and latitudinal
coordinates obtained from the World Port Index (WPI) compiled by the National
Geospatial-Intelligence Agency and nautical port-to-port distances published by the U.S.
Department of Commerce in collaboration with the National Oceanic and Atmospheric
Administration (NOAA) and the National Ocean Service.9 The distinction in local
treatment effects is based on the spatial distribution of U.S. ports and their nautical
distances from the epicenter of Hurricane Katrina’s landfall, which has been approximately
located around Waveland and Bay St. Louis, Mississippi. Based on the selection of ports
under consideration and their geographic locations, the ports of New Orleans, LA, and
Gulfport, MS, have been identified as those closest to the epicenter in the western and
eastern direction, respectively, while the port of Mobile, AL is the only other sample port
still within the hurricane warning zone. Other second or higher order contiguous ports are
located in Florida and Texas or more remote U.S. states. The respective nautical epicenter
distances of ports located in these and other states are presented in Tables 14 and 15.
To gain preliminary insights into the spatial and dynamic distribution of trade
disruptions caused by Hurricane Katrina, a summary detailing the cross-sectional, spatial
and time dimensions of the data is provided in Tables 14 and 15 as well as Figures
10 through 12f. In particular, Tables 14 and 15 provide the average port throughput
of total exports and imports over a two year span pre and post Hurricane Katrina.
8Due to the need to control for spatial correlations and their unique locations, the ports of Honolulu,
HI, and Ranier-Falls, MN, have been excluded from this sample.
9The port-to-port nautical distance matrix is presented in Table 30 in Appendix C
148
Column (1) of each table presents the nautical distances between a given port and the
estimated epicenter of Hurricane Katrina, whereas columns (2) and (3) present the two
year average trade flows pre and post its landfall. These data reveal that the majority of
ports experienced an increase export and import throughput over this time period. The
exceptions to this rule are the first order contiguous ports of New Orleans and Gulfport
on the import side and Gulfport on the export side which exhibit substantial reductions
in trade. Another irregularity that stands out from the general trend is given by the port
of Panama City which experienced a twenty fold increase in imports and immense 198
fold increase in exports. Indicatively, as depicted by Figures 10 and 11, this port is a third
order contiguous port just east of the hurricane warning zone, fortunately spared from its
devastation and clearly benefiting from its proximity to the negatively disrupted ports.
TABLE 14. U.S. Port of Entry - Export Summary Pre & Post Hurricane Katrina
(1) (2) (3) (4) (5) (6) (7)
Ports Dist. to Pre Post Share Share Rank Rank
Epicenter ($ mil.) ($ mil.) Pre (%) Post (%) Pre Post
Gulfport, MS 15.66 85.59 49.24 0.824 0.363 18 24
New Orleans, LA 40.44 222.06 251.18 2.137 1.850 14 15
Mobile, AL 113.67 9.41 17.09 0.091 0.126 28 29
Panama City, FL 237.67 0.16 31.79 0.002 0.234 41 27
Tampa, FL 428.67 4.53 7.65 0.044 0.056 31 31
Port Arthur, TX 481.44 1.25 3.41 0.012 0.025 40 35
Galveston, TX 486.44 6.53 6.79 0.063 0.050 30 32
Freeport, TX 529.44 15.38 18.61 0.148 0.137 27 28
Houston, TX 530.44 936.62 1,402.18 9.012 10.329 4 4
Corpus Christi, TX 595.44 1.61 1.91 0.015 0.014 36 40
Miami, FL 703.00 459.73 487.85 4.424 3.594 9 10
Port Everglades, FL 718.00 247.39 378.39 2.380 2.787 12 12
West Palm Beach, FL 759.00 60.84 87.74 0.585 0.646 21 20
Jacksonville, FL 1015.00 118.33 167.22 1.139 1.232 17 17
Brunswick, GA 1041.00 3.61 2.52 0.035 0.019 33 39
Savannah, GA 1096.00 549.61 834.52 5.288 6.147 8 7
Charleston, SC 1125.00 759.97 889.41 7.312 6.552 6 6
Wilmington, NC 1220.00 41.93 61.66 0.403 0.454 24 23
Continued on next page
149
Table 14 – Continued
(1) (2) (3) (4) (5) (6) (7)
Ports Dist. to Pre Post Share Share Rank Rank
Epicenter ($ mil.) ($ mil.) Pre (%) Post (%) Pre Post
Newport News, VA 1458.00 764.62 1,025.70 7.357 7.556 5 5
Baltimore, MD 1584.00 210.27 276.79 2.023 2.039 15 13
Chester, PA 1624.00 58.17 105.29 0.560 0.776 22 19
Philadelphia, PA 1639.00 75.71 110.22 0.728 0.812 20 18
Perth Amboy, NJ 1642.00 1.32 3.96 0.013 0.029 39 34
Newark, NJ 1650.00 241.53 384.27 2.324 2.831 13 11
New York, NY 1662.00 1,135.75 1,596.33 10.928 11.759 2 2
Boston, MA 1900.00 50.27 73.22 0.484 0.539 23 21
Portland, ME 1940.00 3.60 2.86 0.035 0.021 34 38
Detroit, MI 3645.00 202.64 216.74 1.950 1.597 16 16
Port Huron, MI 3707.00 16.44 13.52 0.158 0.100 26 30
Chicago, IL 4278.00 25.68 34.13 0.247 0.251 25 25
San Diego, CA 4309.00 9.01 3.08 0.087 0.023 29 37
Long Beach, CA 4381.00 1,038.09 1,419.06 9.989 10.453 3 3
Los Angeles, CA 4382.00 1,280.06 1,644.61 12.317 12.115 1 1
Port Hueneme, CA 4456.00 2.12 3.31 0.020 0.024 35 36
San Francisco, CA 4712.00 4.06 5.37 0.039 0.040 32 33
Richmond, CA 4723.00 1.43 1.46 0.014 0.011 38 41
Oakland, CA 4767.00 575.55 668.37 5.538 4.923 7 8
Portland, OR 5330.00 84.62 65.60 0.814 0.483 19 22
Seattle, WA 5486.00 451.02 505.86 4.340 3.726 10 9
Tacoma, WA 5511.00 255.11 273.14 2.455 2.012 11 14
Source: U.S. Census Bureau USA Trade Online dataset
TABLE 15. U.S. Port of Entry - Import Summary Pre & Post Hurricane Katrina
(1) (2) (3) (4) (5) (6) (7)
Ports Dist. to Pre Post Share Share Rank Rank
Epicenter ($ mil.) ($ mil.) Pre (%) Post (%) Pre Post
Gulfport, MS 15.66 190.94 116.85 0.550 0.271 17 22
New Orleans, LA 40.44 259.18 223.57 0.747 0.518 16 17
Mobile, AL 113.67 37.12 69.44 0.107 0.161 27 26
Panama City, FL 237.67 4.17 86.71 0.012 0.201 37 25
Tampa, FL 428.67 10.72 18.87 0.031 0.044 34 33
Port Arthur, TX 481.44 8.06 7.20 0.023 0.017 35 37
Galveston, TX 486.44 20.67 27.72 0.060 0.064 30 31
Freeport, TX 529.44 26.84 11.38 0.077 0.026 29 35
Houston, TX 530.44 999.29 1,433.85 2.881 3.322 10 10
Continued on next page
150
Table 15 – Continued
(1) (2) (3) (4) (5) (6) (7)
Ports Dist. to Pre Post Share Share Rank Rank
Epicenter ($ mil.) ($ mil.) Pre (%) Post (%) Pre Post
Corpus Christi, TX 595.44 0.99 3.00 0.003 0.007 40 40
Miami, FL 703.00 821.43 850.75 2.368 1.971 11 12
Port Everglades, FL 718 428.57 531.95 1.236 1.232 14 14
West Palm Beach, FL 759.00 58.28 47.90 0.168 0.111 26 28
Jacksonville, FL 1015.00 85.63 119.77 0.247 0.277 22 21
Brunswick, GA 1041.00 131.40 88.36 0.379 0.205 21 24
Savannah, GA 1096.00 1,110.64 1,705.11 3.202 3.950 9 9
Charleston, SC 1125.00 2,001.17 2,463.87 5.769 5.708 4 4
Wilmington, NC 1220.00 71.40 113.37 0.206 0.263 25 23
Newport News, VA 1458.00 1,660.56 2,158.80 4.787 5.001 6 6
Baltimore, MD 1584.00 785.69 1,043.14 2.265 2.417 12 11
Chester, PA 1624.00 142.21 185.40 0.410 0.430 19 19
Philadelphia, PA 1639.00 266.94 389.88 0.770 0.903 15 15
Perth Amboy, NJ 1642.00 1.91 0.86 0.005 0.002 39 41
Newark, NJ 1650.00 4,474.18 5,512.94 12.899 12.772 2 2
New York, NY 1662.00 610.09 733.90 1.759 1.700 13 13
Boston, MA 1900.00 158.19 222.42 0.456 0.515 18 18
Portland, ME 1940.00 3.32 3.07 0.010 0.007 38 39
Detroit, MI 3645.00 13.59 19.43 0.039 0.045 33 32
Port Huron, MI 3707.00 5.34 6.01 0.015 0.014 36 38
Chicago, IL 4278.00 16.03 17.28 0.046 0.040 31 34
San Diego, CA 4309.00 74.63 165.40 0.215 0.383 23 20
Long Beach, CA 4381.00 3,491.18 4,236.57 10.065 9.815 3 3
Los Angeles, CA 4382.00 10,929.78 13,532.80 31.510 31.352 1 1
Port Hueneme, CA 4456.00 73.77 27.89 0.213 0.065 24 30
San Francisco, CA 4712.00 28.13 28.58 0.081 0.066 28 29
Richmond, CA 4723.00 14.89 8.86 0.043 0.021 32 36
Oakland, CA 4767.00 1,488.80 1,832.66 4.292 4.246 8 8
Portland, OR 5330.00 138.78 253.46 0.400 0.587 20 16
Seattle, WA 5486.00 1,917.51 2,231.22 5.528 5.169 5 5
Tacoma, WA 5511.00 1,537.71 1,899.13 4.433 4.400 7 7
Source: U.S. Census Bureau USA Trade Online dataset
Columns (4) and (5) of these tables present the shares of each port’s throughput
relative to average total trade prior to and in the aftermath of Hurricane Katrina. Based
on these data, each port has been ranked before and after Katrina’s landfall, with lower
numbers representing larger market shares in U.S. trade. These rankings for exports
151
and imports are given in columns (6) and (7) of Tables 14 and 15, respectively and
point to spatial variation in trade disruptions. While the majority of ports experience
minor changes in rank, ports closest to the epicenter demonstrate rather large relative
adjustments. Gulfport, for example, exhibits a 6 and 5 point drop in export and import
ranking, respectively, whereas the port of Panama City, for example, experiences a 14
and 12 point increase concerning these rankings. In contrast, the port of New Orleans,
LA, displays rather small losses in export and import trade shares that results in a
common one point drop in the respective rankings. While the former findings suggests
that ports located closest to a disaster’s epicenter tend to encounter significant negative
or positive trade disruptions relative to other ports located at greater distances, the latter
points to very idiosyncratic effects. Overall, the data presented in these tables provide
supporting evidence of the local variation of trade disruptions across ports and point to
the importance of modeling the disaggregated trade effects induced by natural disasters.
Building on this initial summary, Figures 10 and 11 provide insights into the
short-run spatial distribution of the cross-sectional observations. In both figures, the
geo-referenced ports are scaled by their one month pre and post Hurricane Katrina
trade values. Overlaying the two trade values, a negative change in containerized trade
is indicated by a larger red circle (i.e. Gulfport), while a positive change in trade is
represented by a larger green circle (i.e. Panama City). Matching the previous medium-
run observations, ports within or just outside of the hurricane warning zone experience
the largest changes in trade after the landfall of Hurricane Katrina. Regardless of whether
exports or imports are considered, the port of Panama City clearly indicates increases
in trade post Hurricane Katrina, while the ports of Gulfport and New Orleans tend
to exhibit the largest losses in trade. In contrast, the geo-referenced ports at greater
distances appear to experience relatively small or no visible changes in trade.
152
TAMPA
MIAMI
MOBILE
HOUSTON
SAVANNAH
GULFPORT
FREEPORTGALVESTON
BRUNSWICK
PALM BEACH
CHARLESTON
PORT ARTHUR PANAMA CITYNEW ORLEANS
JACKSONVILLE
WILMINGTON_NC
PORT EVERGLADES
Source: US National Park Service
Legend
Log(Trade) 07/05
5.774 - 6.000
6.001 - 6.500
6.501 - 7.000
7.001 - 7.500
7.501 - 8.000
8.001 - 8.500
8.501 - 9.000
9.001 - 9.500
Log(Trade) 09/05
5.729 - 6.000
6.001 - 6.500
6.501 - 7.000
7.001 - 7.500
7.501 - 8.000
8.001 - 8.500
8.501 - 9.000
9.001 - 9.500
Threat Level
Hurricane Warning
Hurricane Watch
FIGURE 10. Geo-referenced Port Exports Pre & Post Hurricane Katrina
To investigate the spatial variation and duration of local trade effects, the trends of
U.S. trade at the national, regional and local level are considered next. Figures 12a and
12b illustrate that aggregate U.S. exports and imports exhibit positive overall growth,
albeit large seasonal variations. The vertical red line indicates August, 2005, the month
during which Hurricane Katrina occurred. Both figures demonstrate that compared
to common seasonal variation, the aggregate trade effects of Hurricane Katrina appear
negligible and without any long-term impact. To explore the apparent disconnect between
these aggregate observations and significant trade disruptions indicated in Tables 14 and
15 as well as Figures 10 and 11, the analysis continues at the disaggregated regional trade
level. Figures 12c and 12d present regional trade shares and reveal slight variations from
153
TAMPA
MIAMI
MOBILE
HOUSTON
SAVANNAH
GULFPORT
FREEPORTGALVESTON
BRUNSWICK
PALM BEACH
CHARLESTON
PORT ARTHUR PANAMA CITYNEW ORLEANS
JACKSONVILLE
WILMINGTON_NC
PORT EVERGLADES
Source: US National Park Service
Legend
Log(Trade) 07/05
5.379 - 6.188
6.189 - 7.117
7.118 - 7.651
7.652 - 8.073
8.074 - 8.513
8.514 - 8.951
8.952 - 9.393
9.394 - 10.09
Log(Trade) 09/05
4.869
4.870 - 6.852
6.853 - 7.435
7.436 - 7.991
7.992 - 8.470
8.471 - 9.096
9.097 - 9.421
9.422 - 10.11
Threat Level
Hurricane Warning
Hurricane Watch
FIGURE 11. Geo-referenced Port Imports Pre & Post Hurricane Katrina
the aggregate conclusions. That is, Figure 12c shows that U.S. exports facilitated through
ports located in the U.S. Gulf Coast (including Alabama, Louisiana, Mississippi and
Texas) faced a sharp but temporary decline, while the adjacent Lower Atlantic region
(including Florida, Georgia, North Carolina, South Carolina and Virginia) responded with
an apparent increase in relative exports.10 In addition to that, Figure 12c reveals that
the share of exports held by the remaining U.S. regions remains rather stable during this
period. In contrast, to this preliminary evidence of small short-run trade disruptions at
the regional export level, Figure 12d presents a much less noticeable impact of Hurricane
10These regional definitions follow the categorization by the U.S. Energy Information Administration.
154
Katrina on regional U.S. imports, where we observe very slight reductions in the U.S. Gulf
Coast and no visible effects on the Lower Atlantic or other regions.
22
.8
23
23
.2
23
.4
23
.6
Lo
g 
Tr
ad
e
2003m7 2004m7 2005m7 2006m7 2007m7
Time
(a) National Exports
24
24
.2
24
.4
24
.6
Lo
g 
Tr
ad
e
2003m7 2004m7 2005m7 2006m7 2007m7
Time
(b) National Imports
10
20
30
40
50
60
R
eg
io
na
l T
ra
de
 S
ha
re
 (%
)
2003m7 2004m7 2005m7 2006m7 2007m7
Time
region = Gulf Coast region = Lower Atlantic
region = Other
(c) Regional Export
0
20
40
60
80
R
eg
io
na
l T
ra
de
 S
ha
re
 (%
)
2003m7 2004m7 2005m7 2006m7 2007m7
Time
region = Gulf Coast region = Lower Atlantic
region = Other
(d) Regional Imports
0
5
10
15
20
Lo
ca
l T
ra
de
 S
ha
re
 (%
)
2003m7 2004m7 2005m7 2006m7 2007m7
Time
Contig~y = First Order Contig~y = Second Order
Contig~y = Third Order
(e) Local Exports
0
5
10
15
20
Lo
ca
l T
ra
de
 S
ha
re
 (%
)
2003m7 2004m7 2005m7 2006m7 2007m7
Time
Contig~y = First Order Contig~y = Second Order
Contig~y = Third Order
(f) Local Imports
FIGURE 12. Trends in Aggregate, Regional and Local U.S. Exports and Imports
155
Given these mixed findings at the regional level, a visual representation of the
disaster induced trade disruptions at the local level is offered by Figures 12e and 12f.
Again, the timing of Hurricane Katrina is given by the vertical red line, but now marks
a point of significant disruptions regarding local export and import trade shares across
the more narrowly defined first, second and third order contiguous ports located in the
U.S. Gulf Coast and Florida. Indeed, Figures 12e and 12f provide supporting evidence
that natural disasters cause negative trade disruptions at the immediately affected ports,
whereas positive trade effects are encountered by ports with close proximity. Interestingly,
the depicted trade time paths post Hurricane Katrina further suggest that the recovery
of local trade is much slower than indicated by regional or national comparisons. While
exports appear to recover to pre-disaster levels within the first two years, import trade
shares exhibit long-lasting positive and negative trade disruptions.
Uncovering this initial evidence concerning the spatial and time heterogeneity of
disaster induced trade effects, raises the question of the specific mechanisms driving
the prolonged recovery of local trade. To this end, I provide Figures 13a and 13b which
depict the number of two-digit HS traded product groups pre and post Hurricane Katrina.
Both figures reveal dramatic and long-lasting reductions in the number of imported and
exported products at the first order contiguous ports which coincide with a significant
and permanent increase in the number of exported and imported products at the third
order contiguous ports. Complementing the spatial heterogeneity in trade effects, second
order contiguous ports, however, reveal no visual treatment effects concerning the
number of traded products. In line with the negligible aggregate trade effects, higher
order contiguous ports exhibit no change concerning their average trade composition
in response to this natural disaster. These observations offer an first insights into port-
specific the short-run and long-run recovery of local trade and suggest that both the local
156
0
20
40
60
80
# 
of
 T
ra
de
d 
Pr
od
uc
t G
ro
up
s
2003m7 2004m7 2005m7 2006m7 2007m7
date_index
Contig~y = First Order Contig~y = Second Order
Contig~y = Third Order Contig~y = other
(a) Exports
0
20
40
60
# 
of
 T
ra
de
d 
Pr
od
uc
t G
ro
up
s
2003m7 2004m7 2005m7 2006m7 2007m7
date_index
Contig~y = First Order Contig~y = Second Order
Contig~y = Third Order Contig~y = other
(b) Imports
FIGURE 13. Trends in the Number of Traded Products
patterns and composition of trade are subject to caused by natural disasters. Overall, this
summary of the data provides strong preliminary evidence of the spatial and dynamic
heterogeneity of disaster induced trade effects and resilience of trade.
Results
In this section, I present the empirical findings obtained from a variety of analyses
culminating in the estimation of the empirical model specification given by equation 4.4.
Starting with the aggregate effects on U.S. trade, I show that Hurricane Katrina has had
negligible average export and import treatment effects across all ports. Dissecting these
insignificant treatment effects, shown in Table 16, at a more disaggregate level, I then
turn towards an estimation differentiating regional effects across ports located in the U.S.
Gulf Coast and Lower Atlantic from other U.S. regions. Still, the results, given in Table
17, provide insignificant trade effects at the regional level and point to the intraregional
resilience of trade. Further investigating this resilience, I focus on the estimation of local
trade disruptions and turn towards a more flexible model that allows for port-specific
short-run, medium-run or long-run treatment effects provided in Table 18. This analysis
157
provides strong supporting evidence for localized trade disruptions and their systematic
spatial heterogeneity.
Having obtained these port-specific average treatment effects, I provide potential
explanations for their specific spatial distribution. To this end, I estimate the effects of
port characteristics on these local disruptions. While both import and export disruptions
depend on the distance to Hurricane Katrina, only the rerouting of exports exhibits
dependence on port characteristics, such as harbor type or entry restrictions. Given
this predominant role of distance in explaining the static resilience of trade, I further
develop this analysis by estimating the variation in trade disruptions over nautical
distance to Katrina’s epicenter. The marginal effects obtained from these regressions
are presented in Figures 14a-14d. In addition to these static estimations of spatial trade
effects, I also present the dynamic results providing evidence of both long-run recovery and
permanent treatment effects for some of the first to sixth order contiguous ports relative
to Hurricane Katrina’s epicenter. Concluding this section, I explore the main drivers of
the unexpected permanent long-run effects and estimate trade disruptions for individual
product categories. The results show that the port-specific change in the number of
handled products exhibits industry-specific path dependencies and is mainly driven by
trade in textiles.
Aggregate Analysis
Most of these analyses are based on the respective export and import estimations
using the Poisson Pseudo-Maximum Likelihood (PPML) estimator, standard in the
trade literature (Santos Silva and Tenreyro, 2006), and a variety of spatial econometric
specifications, including the SAR, SEM and SAC models. The primary spatial results are
based on a third order contiguity weight matrix, as discussed by Ord (1975), LeSage and
158
TABLE 16. Aggregate Trade Disruptions
(1) (2) (3) (4) (5) (6) (7) (8)
PPML SAR SEM SAC
VARIABLES Export Import Export Import Export Import Export Import
Hurricane -0.012 -0.040 0.307 -0.076 0.287 -0.074 0.297 -0.075
Katrina (0.019) (0.029) (0.261) (0.133) (0.238) (0.129) (0.249) (0.131)
Observations 1,960 1,960 1,960 1,960 1,960 1,960 1,960 1,960
R-squared 0.987 0.995 0.006 0.002 0.006 0.002 0.006 0.002
Trend Yes Yes Yes Yes Yes Yes Yes Yes
Port FE Yes Yes Yes Yes Yes Yes Yes Yes
Month FE Yes Yes Yes Yes Yes Yes Yes Yes
Time FE No No No No No No No No
Spatial- None None 3rd 3rd 3rd 3rd 3rd 3rd
Weighting Order Order Order Order Order Order
Convergence yes yes yes yes yes yes yes yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Pace (2009) or Getis (2009), for example, while a host of robustness analyses point to the
consistency of the results across estimations involving alternative weight matrices, such as
1st order contiguity or inverse distance, for example. The aggregate effects are presented
in Table 16 and match the findings of the previous literature (Parsons, 2014). Controlling
for a time trend, port specific fixed effects and seasonal variation, I find that Hurricane
Katrina has a statistically insignificant average impact across the major U.S. ports of
entry and exit included in this sample. This negligible effect is consistent across the PPML
and spatial estimators as well as exports and imports and is robust to alternative weight
matrix specifications.11
Turning towards the regional analysis, I find similarly insignificant results for the
U.S. coastal regions closest to Hurricane Katrina’s epicenter. Table 17 provides treatment
11See Table 31 in Appendix C. While columns (1)-(6) of Table 31 present the export and import trade
effects based on spatial weighting involving contiguity matrices of orders 1, 2 and 4, columns (7) and
(8) depict the results obtained with an inverse distance weighting matrix. The port-to-port distances
underlying this inverse distance weight matrix are presented in Table 30 of Appendix C.
159
TABLE 17. Regional Trade Disruptions
(1) (2) (3) (4) (5) (6) (7) (8)
PPML SAR SEM SAC
VARIABLES Export Import Export Import Export Import Export Import
Gulf Coast 0.056 -0.008 0.270 0.145 0.201 0.147 -0.213 0.145
(0.088) (0.136) (0.237) (0.410) (0.241) (0.388) (1.223) (0.418)
Lower Atlantic 0.023 0.035 1.202 0.645 0.946 0.592 1.450 0.646
(0.059) (0.047) (0.936) (0.492) (0.808) (0.447) (1.819) (0.526)
Observations 1,960 1,960 1,960 1,960 1,960 1,960 1,960 1,960
R-squared 0.988 0.996 0.011 0.007 0.011 0.007 0.016 0.007
Port FE Yes Yes Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes Yes Yes
Spatial- None None 3rd 3rd 3rd 3rd 3rd 3rd
Weighting Order Order Order Order Order Order
Convergence yes yes yes yes yes yes yes yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
effect estimates that differentiate the average trade disruptions experienced by ports
located in the U.S. Gulf Coast and Lower Atlantic regions relative to ports located in all
other U.S. coastal regions. In line with previous findings, columns (1) through (8) display
statistically insignificant results for both regions across all export and import estimations.
That is, controlling for port and time specific fixed effects, Hurricane Katrina appears to
have had no discernibly different impact on trade across the U.S. Gulf Coast and Lower
Atlantic regions relative to all others. Again, this finding is robust to variations in weight
matrix specifications across all spatial econometric models.12
12See Table 32 in Appendix C.
160
Static Local Analysis
Based on the previous summary of the data, the insignificant regional findings come
at no surprise. Instead, these results speak to the intraregional resilience of international
trade offsetting local port trade reductions through immediate rerouting of traded goods
to nearby and within region facilities with available capacity. To better understand the
dependence of the disaster induced trade effects on port proximity to the epicenter of a
natural disaster, I reestimate the model at a more disaggregated level. For this analysis
I adapt the spatial autoregression model to inlcude port-specific export and import
treatment effects across various time horizons. While columns (1) and (2) of Table 18
display the estimated trade disruptions averaged over three months post Hurricane
Katrina’s landfall, columns (3) and (4) as well as (5) and (6) present average treatment
effects over two and eight years post disaster, respectively.
TABLE 18. Port-Specific Trade Disruptions
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
Seattle, WA 0.070*** 0.026*** 0.042*** -0.055*** -0.169*** -0.048***
(0.004) (0.000) (0.002) (0.002) (0.015) (0.001)
Portland, OR -0.375*** -0.004*** -0.247*** 0.378*** -0.406*** 0.240***
(0.002) (0.001) (0.004) (0.002) (0.019) (0.001)
Oakland, CA 0.025*** 0.011*** 0.073*** -0.003 0.121*** 0.044***
(0.000) (0.002) (0.007) (0.003) (0.030) (0.002)
Richmond, CA -0.134*** -0.450*** 0.287*** -0.663*** 0.747*** -0.154***
(0.000) (0.001) (0.006) (0.002) (0.022) (0.001)
San Francisco, CA 1.405*** -1.129*** -0.180*** -0.174*** -0.674*** 0.229***
(0.007) (0.002) (0.007) (0.003) (0.031) (0.002)
Port Hueneme, CA 1.050*** -1.200*** 0.936*** -1.043*** 1.308*** -0.672***
(0.004) (0.003) (0.003) (0.001) (0.019) (0.001)
Los Angeles, CA -0.016 -0.143*** 0.167*** -0.005*** 0.103* 0.084***
(0.027) (0.015) (0.006) (0.002) (0.055) (0.000)
Long Beach, CA 0.184*** -0.095*** 0.227*** -0.020*** 0.179*** -0.043***
(0.026) (0.014) (0.005) (0.001) (0.049) (0.000)
San Diego, CA -0.172*** 0.501*** -0.452*** 0.605*** -0.743*** 0.328***
(0.021) (0.012) (0.007) (0.001) (0.058) (0.000)
Chicago, IL 0.148*** -0.204*** 0.209*** 0.017*** -2.435*** 0.472***
Continued on next page
161
Table 18 – Continued from previous page
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
(0.007) (0.008) (0.002) (0.006) (0.027) (0.012)
Port Huron, MI 0.195*** -0.316*** -0.294*** -0.076*** -0.000 0.048***
(0.006) (0.007) (0.004) (0.004) (0.043) (0.008)
Detroit, MI -0.004 -0.169*** -0.007** 0.138*** -0.737*** 0.297***
(0.005) (0.006) (0.003) (0.004) (0.021) (0.007)
Portland, ME 0.084*** 0.690*** -0.327*** -1.461*** -1.954*** -4.020***
(0.006) (0.003) (0.006) (0.000) (0.003) (0.001)
Boston, MA 0.260*** -0.186*** 0.285*** 0.113*** 0.060*** 0.280***
(0.006) (0.000) (0.005) (0.006) (0.007) (0.005)
New York, NY 0.113*** -0.161*** 0.252*** -0.033*** 0.303*** 0.289***
(0.006) (0.000) (0.007) (0.005) (0.007) (0.004)
Newark, NJ 0.198*** -0.130*** 0.368*** -0.003 0.351*** 0.151***
(0.005) (0.000) (0.008) (0.005) (0.017) (0.004)
Perth Amboy, NJ 0.120*** -2.211*** 0.490*** -1.469*** 0.730*** 0.998***
(0.005) (0.004) (0.010) (0.000) (0.030) (0.001)
Philadelphia, PA 0.198*** -0.144*** 0.342*** 0.173*** 0.177*** 0.193***
(0.005) (0.002) (0.010) (0.003) (0.035) (0.002)
Chester, PA 0.230*** -0.190*** 0.512*** 0.050*** 0.371*** 0.159***
(0.003) (0.001) (0.010) (0.002) (0.038) (0.003)
Baltimore 0.085*** -0.103*** 0.192*** 0.073*** 0.149*** 0.197***
(0.003) (0.001) (0.010) (0.002) (0.036) (0.002)
Newport News, VA 0.176*** -0.033*** 0.214*** 0.048*** 0.140*** 0.091***
(0.004) (0.005) (0.009) (0.001) (0.035) (0.002)
Wilmington, NC 0.654*** 0.015*** 0.307*** 0.276*** 0.790*** 0.893***
(0.012) (0.004) (0.007) (0.001) (0.030) (0.001)
Charleston, SC 0.044*** -0.087*** 0.085*** -0.003*** -0.140*** 0.026***
(0.009) (0.004) (0.007) (0.001) (0.037) (0.001)
Savannah, GA 0.129*** 0.062*** 0.333*** 0.208*** 0.522*** 0.522***
(0.009) (0.004) (0.006) (0.002) (0.030) (0.002)
Brunswick, GA -1.252*** -0.381*** -0.291*** -0.749*** 0.225*** -1.051***
(0.003) (0.005) (0.009) (0.000) (0.035) (0.000)
Jacksonville, FL 0.233*** -0.258*** 0.270*** 0.124*** 0.300*** 0.538***
(0.012) (0.004) (0.006) (0.003) (0.027) (0.004)
Palm Beach, FL 0.236*** -0.229*** 0.294*** -0.419*** 0.001 -0.911***
(0.013) (0.004) (0.007) (0.001) (0.033) (0.000)
Port Everglades, FL 0.232*** 0.039** 0.324*** -0.001 0.224** -0.094***
(0.032) (0.019) (0.043) (0.007) (0.104) (0.005)
Miami, FL -0.001 -0.200*** -0.037 -0.184*** -0.340*** -0.191***
(0.038) (0.022) (0.048) (0.010) (0.115) (0.008)
Corpus Christi, TX -4.999*** 3.984*** -0.269*** 2.332*** 2.894*** 1.387***
(0.004) (0.005) (0.004) (0.000) (0.024) (0.001)
Houston, TX 0.106*** 0.058*** 0.310*** 0.148*** 0.320*** 0.239***
(0.032) (0.004) (0.006) (0.002) (0.040) (0.005)
Freeport, TX 0.218*** -0.647*** 0.131*** -0.796*** -0.101** -0.654***
(0.021) (0.019) (0.008) (0.003) (0.046) (0.002)
Galveston, TX 0.942*** -0.192*** -0.088*** 0.250*** 0.479*** 0.475***
(0.030) (0.011) (0.008) (0.000) (0.040) (0.004)
Continued on next page
162
Table 18 – Continued from previous page
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
Port Arthur, TX 1.232*** -0.175*** 1.728*** -1.515*** 0.145*** -2.980***
(0.015) (0.007) (0.003) (0.003) (0.029) (0.001)
Tampa, FL -0.149*** -0.118*** 0.373*** 0.389*** 0.177* 0.867***
(0.032) (0.017) (0.043) (0.007) (0.102) (0.005)
Panama City, FL 10.512*** 5.369*** 10.566*** 5.236*** 10.133*** 5.034***
(0.019) (0.005) (0.005) (0.002) (0.027) (0.000)
Mobile, AL 0.045 0.341*** 0.480*** 0.406*** 1.070*** 1.060***
(0.029) (0.009) (0.046) (0.004) (0.094) (0.002)
New Orleans, LA -1.273*** -2.043*** -0.078* -0.487*** -0.149 0.003
(0.039) (0.017) (0.048) (0.006) (0.108) (0.003)
Gulfport, MS -1.387*** -1.908*** -0.695*** -0.787*** -1.104*** -0.728***
(0.035) (0.017) (0.051) (0.007) (0.114) (0.005)
Observations 1,120 1,120 1,960 1,960 4,840 4,840
R-squared 0.907 0.863 0.900 0.855 0.825 0.829
Random Effects Yes Yes Yes Yes Yes Yes
Spatial- 3rd 3rd 3rd 3rd 3rd 3rd
Weighting Order Order Order Order Order Order
Convergence yes yes yes yes yes yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
With respect to the excluded port of Tacoma, WA, nearly all treatment effects are
statistically significant at the 1% level regardless of the time horizon under consideration
and speak to the considerable month-to-month seasonal volatility and port-to-port
variation observed in the trade data. While their statistical significance is largely invariant
across ports, point estimates of the short to long-run average treatment effects exhibit
large systematic variation. Albeit some natural outliers among the forty sample ports,
such as Port Hueneme, CA or Portland, ME, Table 18 shows that those ports closer to
Hurricane Katrina’s epicenter tend to experience the largest short-run trade disruptions.13
Specifically, both the port of Gulfport and New Orleans, the ports closest to Katrina’s
epicenter, reflect considerable short-run reductions in both exports and imports ranging
from 72% to 87%. While these short-run effects are consistent for both ports, their
13For the ease of comparison, the parameters of interest are bold-faced and highlighted.
163
persistence varies. The port of New Orleans, for example, experiences a more rapid
recovery relative to Gulfport. This recovery leads to significant reductions in absolute
treatment effects over the medium to long-run and ultimately insignificant reductions in
trade eight years post Hurricane Katrina. In contrast, Gulfport experiences incomplete
recovery leading to statistically significant long-run trade reductions of 67% and 52% for
exports and imports, respectively, relative to pre-treatment levels.
Extending the scope just beyond the hurricane warning zone, as depicted in Figures
10 and 11, the port of Panama City experiences the largest positive short-run trade
disruptions out of any sample port. Point estimates range from 5.369 to 10.512 for imports
and exports, respectively. Equally important is the fact that these estimated treatment
effects stay relatively constant and statistically significant when extending the sample
to the medium and long-run. The suggested permanence of these disaster-induced trade
effects could speak towards the substantial switching costs for carriers choosing their ports
of entry and exit and the resulting path dependence. Interestingly, the port of Mobile
exhibits an insignificant export treatment effect and statistically significant, but relatively
small increase in imports in the short-run. However, when considering the medium to
long-run, these estimated treatment effects increase and become statistically significant for
both exports and imports. An intuitive explanation for this initially surprising finding may
be the fact that this port was located within the hurricane warning zone. Albeit being
sparred from considerable damage, carriers may have avoided the entire hazard zone,
including the port of Mobile, in the short-run. In the medium to long-run, however, the
port’s proximity to the severely damaged infrastructure of New Orleans and Gulfport,
in particular, may have swayed carriers to consider the port of Mobile as a low-cost
alternative.14
14In general, the port-specific results are qualitatively and quantitatively very consistent when estimated
via the SAC model. The results of this robustness check are presented in Table 33 in Appendix C.
164
To address the evidenced short-run to long-run spatial heterogeneity in trade effects
and gain a better understanding of the underlying factors driving the indicated resilience
of international trade and persistence in trade effects, I conduct an additional analysis
regressing the previously estimated port-specific treatment effects on time-invariant
port characteristics and inverse distance. To gain insights into the selection of ports for
rerouting carriers, I restrict the analysis to include only those treatment effects that are
obtained for ports outside of the hurricane warning zone, excluding Gulfport, New Orleans
and Mobile from the sample. The available characteristics include harbor type as well
as tidal and other entry restrictions. The legend key to these characteristics is given
by Table 34 in Appendix C and shows that the respective reference groups are coastal
breakwater ports for the available harbor types and ports without any tidal or other
entry restrictions. Given these reference groups, negative point estimates on any port
characteristics imply that the presence of entry restrictions and specific harbor types other
than coastal breakwater reduce the positive disaster induced treatment effects. Inverse
distance is calculated as the 1/(nautical distance to Katrina’s epicenter), as presented
in Tables 14 and 15, so that a positive coefficient implies that closer ports experience
exponentially larger treatment effects.
The results presented in Table 19 are quite interesting. That is, regardless of the
time horizon under consideration, export treatment effects tend to be much more sensitive
to the harbor type and entry restrictions than import treatment effects. In fact, the
statistically significant negative point estimates suggest that almost all harbor types
are less preferred relative to coastal breakwater ports when it comes to redirecting U.S.
containerized exports. Carriers rerouting U.S. containerized imports, on the other hand,
only seem to avoid lake and canal ports with a statistical significant impact only for the
medium to long-run treatment effects. Similarly, carriers facilitating rerouted exports
165
tend to avoid ports with tidal and other entry restrictions, particularly in the short-run,
while these entry restrictions seem to have little influence on a carrier’s port choice when
rerouting imports.
TABLE 19. Influence of Port Characteristics on Treatment Effects
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
HT (=1) -1.648* 0.094 -1.589** -0.111 -1.638** -0.430
(0.889) (0.527) (0.672) (0.528) (0.692) (0.569)
HT (=2) -0.531 0.266 -0.463 0.266 -3.231*** 0.394
(0.375) (0.370) (0.281) (0.403) (0.385) (0.423)
HT (=3) -1.803 -1.342 -1.743 -2.866*** -3.750** -4.706***
(1.679) (0.827) (1.445) (0.855) (1.359) (0.920)
HT (=4) -0.957* 0.009 -0.870* -0.077 -1.071** -0.259
(0.546) (0.409) (0.448) (0.446) (0.503) (0.477)
HT (=5) -1.063* -0.326 -1.101** -0.188 -1.208** 0.041
(0.571) (0.447) (0.475) (0.478) (0.514) (0.497)
HT (=6) -2.685* -0.939 -2.869** -1.036 -3.252** -1.315
(1.522) (0.762) (1.309) (0.791) (1.238) (0.852)
ER - Tide -0.793* -0.009 -0.657 -0.303 -0.764* -0.846*
(0.458) (0.323) (0.397) (0.343) (0.406) (0.452)
ER - Other -1.790* -0.404 -1.412* -0.172 -0.792 0.137
(1.048) (0.550) (0.792) (0.459) (0.742) (0.645)
Inv. Distance 1,278.532 888.297** 1,518.094** 868.057** 1,681.201** 893.954**
(836.505) (381.456) (720.448) (378.613) (659.917) (396.400)
Observations 37 37 37 37 37 37
R-squared 0.415 0.451 0.613 0.503 0.637 0.536
Std. Errors Robust Robust Robust Robust Robust Robust
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
In contrast to the varying responses of export and import treatment effects with
respect to port characteristics, both types of treatment effects are dependent upon
the proximity of a given port. With the exception of the estimated short-run export
disruptions, the inverse of nautical distance to the epicenter has a statistically significant
positive impact on all export and import treatment effects. This, of course, translates into
an exponential decline of the port-specific treatment effects in absolute distance; a finding
that is increasingly pronounced in the medium and long-run. Both of these findings are
166
quite intuitive. In the short-run, distance may not be as important of a factor as available
capacity to handle rerouted trade immediately. Carriers may travel considerable distances
to avoid being delayed and incur penalties. In the long-run, however, capacity could
potentially be available at any port, so that distance becomes a more prominent factor
in the determination of port choice, particularly when transport costs are an exponential
function of this factor.
To further explore this dependence on distance, offer insights for both positive and
negative treatment effects alike and gain a better understanding of potential differences
in this distance dependence across export and import treatment effects, I develop and
estimate a model interacting intraregional port groups with the inverse of nautical
distance to Hurricane Katrina’s epicenter averaging treatment effects over two years
post Hurricane Katrina. For the purposes of this estimation, I differentiate between core
and peripheral ports. The core is defined as those ports with first order contiguity to
Hurricane Katrina’ epicenter, while the periphery is defined as those ports located at
greater orders of contiguity but still within the Gulf Coast and Lower Atlantic regions.15
The estimation includes the traditional PPML estimator and spatial SAR, SEM and
SAC models. The point estimates obtained from these regression analyses are presented
in Table 36 of Appendix C and illustrate the statistically significant dependence of core
and periphery treatment effects on distance. Based on these point estimates, Figures 14a
through 14d illustrate the converted marginal export and import treatment effects for core
and peripheral ports across distance for each of the respective models.16
15Since the port of Mobile, AL is located within the hurricane warning zone, but is yet a 2nd order
contiguous port, it is unclear whether to consider this port as part of the core or periphery and it is, thus,
excluded from this analysis. While their statistical significance suffers when including this port, point
estimates vary only slightly.
16The conversion of the core and periphery point estimates into percentage changes in the value of trade
is based on the following calculations: ∆%Core = (exp(βcore) ∗ exp(βcore−dist. ∗ (1/(Distance))− 1) ∗ 100
and ∆%Periphery = (exp(βperiphery) ∗ exp(βperiphery−dist. ∗ (1/(Distance))− 1) ∗ 100, respectively.
167
Several key features of these graphs are important to point out. While the estimated
percentage changes for core ports are similar near the epicenter across all models,
correction for spatial correlations leads to diverging estimates as distance increases.
Concerning peripheral ports, the adjustment for potential spatial correlation leads to
significantly larger treatment effects in close proximity to Hurricane Katrina’s epicenter
and a much steeper decline of these effects as distance increases relative to the PPML
estimates. Although the estimated peripheral changes appear very large, the underlying
differences in average pre and post port throughput listed in Tables 14 and 15 are also
quite substantial. Rerouting of traded products away from a fairly large port, like New
Orleans, can result in dramatic throughput changes in smaller nearby ports, like Panama
City, which exhibits a 198 fold, or in other words roughly 19000%, increase in average
exports post Hurricane Katrina.17
Overall the graphs show that the core ports experience economically significant
export and import declines, although the reduction in exports is estimated to decay
much faster over distance than the reduction in imports. That is, while trade handled
by Gulfport, 15.7 nautical miles east of Katrina’s epicenter, for example, is estimated to
experience a 45.5% average reduction in exports and 45.0% average reduction in imports,
the average trade effects at the port of New Orleans, 40.4 nautical miles west of Katrina’s
epicenter, for example, are estimated to yield only a 1.0% reduction in exports, but a
27.0% reduction in imports.18 Treatment effects of the peripheral ports also reflect the
expected patterns. That is, ports of close proximity to the epicenter experience larger
17A natural concern is the feasibility of such staggering increases and whether a port, like Panama City,
can handle such shipments. The answer to this pressing question, lies within the presence of large sporadic
pre-treatment shipments as evidenced by Panama City, FL prior to Hurricane Katrina. Although these
shipments are large and require sufficient capacity, their sporadic nature implies little influence on pre-
treatment average port throughput. Thus, Panama City appears as a small port, but with considerable
excess capacity prior to Hurricane Katrina, capable of handling tremendous short-run increases.
18Calculations are based on the point estimates obtained from the SAR model.
168
−
10
0
−
80
−
60
−
40
−
20
0
%
 C
ha
ng
e 
in
 T
ra
de
0 20 40 60 80 100
Distance to Epicenter
PPML SAR
SEM SAC
(a) Exports - Core
−
10
0
−
80
−
60
−
40
−
20
%
 C
ha
ng
e 
in
 T
ra
de
0 20 40 60 80 100
Distance to Epicenter
PPML SAR
SEM SAC
(b) Imports - Core
0
10
00
00
20
00
00
30
00
00
%
 C
ha
ng
e 
in
 T
ra
de
200 300 400 500 600
Distance to Epicenter
PPML SAR
SEM SAC
(c) Exports - Periphery
0
10
00
20
00
30
00
40
00
50
00
%
 C
ha
ng
e 
in
 T
ra
de
200 300 400 500 600
Distance to Epicenter
PPML SAR
SEM SAC
(d) Exports - Periphery
FIGURE 14. Treatment Effects over Distance
gains in trade than those ports at greater distances. Similarly to the marginal effects
on trade handled by core ports, increases in exports decay at a faster rate over distance
than increases in imports. That is, a peripheral port at a distance of roughly 400 nautical
miles, such as Tampa, FL, experiences a 1957.0% increase in exports and 375.0% increase
imports, while a port at a distance of roughly 700 nautical miles, such as Miami, FL,
experiences only a 7.2% increase in exports, but 10.9% increase in imports. The fact that
export trade disruptions are estimated to be more sensitive to distance than imports is
intuitive. Imported containers are facilitated on container vessels that incur lower costs
169
per mile traveled than exported containers that have to be rerouted via rail or trucking,
both commanding much higher rates per mile than container vessels.
Dynamic Local Analysis
Having provided considerable evidence in support of the static local resilience of
international trade and some preliminary evidence supporting its dynamic resilience, I now
turn towards the primary estimation of the dynamic and spatially heterogeneous trade
effects induced by Hurricane Katrina. To this end, I estimate the empirical specification
given by equation (4.4) over the entire sample period, R = 24 and S = 96. Given the
preliminary port-specific short-run to long-run treatment effects presented in Table 18, I
focus this discussion around the ports closest to Hurricane Katrina, including Gulfport,
New Orleans, Mobile and Panama City.19 Since a tabular representation of these dynamic
trade disruptions is quite convoluted, I use graphical representations of the estimated
port-specific treatment effects instead. Figures 15a through 17b display port-specific
time fixed effects in relation to August, 2005, the landfall of Hurricane Katrina. The
vertical red line at 0 indicates this landfall, while the green vertical line at a value of 34
month post treatment indicates June, 2008, the midst of the Great Trade Collapse (GTC).
Both of these vertical lines are included to aid in an additional comparison between the
local trade disruptions of Hurricane Katrina and the GTC between 2008 and 2009 and
help put the magnitude of these localized disaster effects into perspective. Given the
empirical specification, each point on the graph can be thought of as a difference-in-
differences estimator. That is, each point estimate post treatment reflects a month-port-
specific treatment effect relative to that port’s trade during the month of August, 2005,
19As part of the robust analysis, I also provide the dynamic treatment effects for ports at greater
distances. The results are provided via Figures 18a through 25b and show no systematical treatment
effects resulting from Hurricane Katrina.
170
−
2
−
1
0
1
2
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
1st Order Neighbor − West (New Orleans (LA))
(a) Exports - New Orleans, LA
−
4
−
2
0
2
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
1st Order Neighbor − West (New Orleans (LA))
(b) Imports - New Orleans, LA
−
3
−
2
−
1
0
1
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
1st Order Neighbor − East (Gulfport (MS))
(c) Exports - Gulfport, MS
−
4
−
3
−
2
−
1
0
1
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
1st Order Neighbor − East (Gulfport (MS))
(d) Imports - Gulfport, MS
FIGURE 15. Dynamic Variation in Treatment Effects - New Orleans, LA & Gulfport, MS
the time of Hurricane Katrina’s landfall, and relative to the average change in trade
across the excluded ports from August, 2005 to the given month under consideration.
The flexibility of this specification allows for a clearer evaluation of the persistence of the
disaster induced trade effects and facilitates dynamic cross port comparisons.20
Figures 15a through 15d, for example, display tremendous short-run trade
disruptions at the ports of New Orleans and Gulfport, but also reveal that the duration
and magnitude of these disruptions vary across imports and exports as well as across the
20The presented trade effects are based on the SAR model and are robust to the use of the SEM or SAC
models.
171
individual ports. For both ports, imports, for example, suffer a larger short-run reduction
than exports. This finding is quite intuitive, given the fact that imports facilitated by
container vessels are more easily rerouted than exports transported by train, truck or
inland waterways. Relative to importing container vessels, these types of transportation
modes are subject to significant rerouting barriers. This is particularly true for the port
of New Orleans which is strategically located at the mouth of the Mississippi River and
one of the main facilitator of bulk exports that have few transportation substitutes to the
inland waterways. This route dependence of exports causes rather short-lived delays in
export shipments and leads to smaller short-run export reductions.
While these short-run export/import comparisons are similar for both ports, a long-
run cross port comparison reveals that both exports and imports for the port of Gulfport
experience rather lasting reductions relative to the port of New Orleans. In fact, while
exports and imports handled by the port of New Orleans experience a rapid recovery
over the first six month post treatment, depicted in Figures 15a and 15b, Gulfport’s
merely partial recovery is much more prolonged for both exports and imports, as shown by
Figures 15c and 15d. These negative long-run effects can be traced back to the substantial
damages sustained by the infrastructure at the port of Gulfport and documented by
Grenzeback et al. (2008). Interestingly, these figures also illustrate that despite smaller
short-run reductions, exports facilitated through Gulfport exhibit larger long-run effects
than imports. A potential explanation for this amplified long-run response in exports
may be the observable negative pre-treatment trend that was enhanced through Hurricane
Katrina. Lastly, in the relation to the GTC, Figures 15a through 15d show that Hurricane
Katrina had a much more severe localized short-run and even long-run impact on exports
and imports than the GTC, raising the importance of a local trade analysis that can bring
to light significant disruptions missed by aggregate estimations.
172
−
1
0
1
2
3
4
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
2nd Order Neighbor − East (Mobile (AZ))
(a) Exports - Mobile, AL
−
1
0
1
2
3
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
2nd Order Neighbor − East (Mobile (AZ))
(b) Imports - Mobile, AL
FIGURE 16. Dynamic Variation in Treatment Effects - Mobile, AL
−
5
0
5
10
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
3rd Order Neighbor − East (Panama City (FL))
(a) Exports - Panama City, FL
−
15
−
10
−
5
0
5
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
3rd Order Neighbor − East (Panama City (FL))
(b) Imports - Panama City, FL
FIGURE 17. Dynamic Variation in Treatment Effects - Panama City, FL
Expanding the spatial scope of the analysis, I now consider the dynamic trade effects
of Hurricane Katrina on the ports of Mobile, AL and Panama City, FL. Both of these
ports are just outside of the disaster stricken region and, as the previous static port-
specific analysis suggests, are the primary candidates for the evaluation of the static
and dynamic resilience of international trade. The results for both of these ports are
presented in Figures 16a through 17b and illustrate significant short-run as well as long-
run disruptions across both exports and imports. However, the depicted responses are very
173
idiosyncratic across the two ports. While the short-run effects on Mobile exports suggest
a one month significant reduction followed by a five month period of no change relative to
pre-treatment export levels, the import trade effects at the port of Mobile are estimated
to be positive and economically as well as statistically significant in the first two month
following treatment before tapering off to pre-treatment levels over the next four month.
Contrary to these opposing short-run responses, Figures 16a and 16b also reveal that both
exports and imports facilitated through Mobile start experiencing positive growth roughly
six to nine month post treatment. Again, these long-run effects vary across imports and
exports. In the case of imports this growth merely matches the pre-treatment trend, while
the growth in exports triggered by Hurricane Katrina is a reversal of a slightly negative
pre-treatment trend. A potential explanation for this considerable change in the long-
run growth rate of exports at the port of Mobile and reversal of the short-run reductions,
is the pronounced long-run decline in exports experienced by the neighboring port of
Gulfport.
In contrast to these varying and mild short-run responses as well as persistent
growth rate changes of trade handled by the port of Mobile, exports and imports
facilitated through the port of Panama City, depicted by Figures 17a and 17b, experience
immediate and large increases in response to Hurricane Katrina that are very persistent
in the long-run. While imports reveal a statically and economically significant increase
post Hurricane Katrina, Panama City’s exports reflect an unparalleled economically
and statistically significant upward jump that persists through the entire sample period.
Even during the GTC, these long-run responses of exports and imports remain effectively
undisturbed.
Based on these estimated treatment effects across the ports of Gulfport, New
Orleans, Mobile and Panama City, one can conclude that the static resilience of
174
international trade and resulting negligible aggregate trade effects are mainly driven by
rerouted exports and imports through the port of Panama City. Moreover, the estimates
show that the dynamic resilience of international trade in response to Hurricane Katrina
is driven by the long-run recovery by the port of New Orleans and the persistent long-run
increases in exports and imports at the ports of Mobile and Panama City that offset the
permanent reductions experienced by port of Gulfport.
In contrast to these consistent and considerable short-run and long-run trade effects
displayed by the ports of Gulfport, New Orleans, Mobile and Panama City, ports located
at greater distances show no identifiable response to Hurricane Katrina. In fact, Figures
18a through 25b in Appendix C provide no evidence of any short-term or long-term
responses in exports or imports that are distinguishable from pre-hurricane variation or
can be directly linked to Hurricane Karina. In summary, these findings suggest that albeit
significant disaster induced reductions in trade for those ports closest to the epicenter, a
transport network, similar to that of the U.S., can offset the majority of trade reductions
caused by a natural disaster, similar to Hurricane Katrina, within a small geographic
region. The indicated mechanisms of the resilience of trade include rerouting and delaying
of traded products within a narrow band of nearby ports leaving short-run and long-run
trade effects unaffected at the aggregate level.
Disaggregated Analysis
To gain a better understanding of the underlying forces driving these dynamic
disruptions in the value of trade, I redirect the focus of the analysis towards the effects
at a more disaggregated product level. Specifically, I consider the effects of Hurricane
Katrina on the number of products facilitated by a given port differentiating across several
different product categories. Reestimating the port-specific treatment effects over the
175
short, medium and long-run on the number of traded products, I provide insights into the
specific industries driving the static and dynamic resilience of international trade. Similar
to the dynamic analysis, the results reported here focus on the primary ports of interest
including Gulfport, New Orleans, Mobile and Panama City and are summarized in Tables
20 and 21 for exports and imports, respectively.21
The number of traded products for a given port-month pair is based on the 2 digit
HS classification that consolidates individually traded products into 98 product groups.
While column (1) of Tables 20 and 21 provides the treatment effects of Hurricane Katrina
on the total number of traded products, columns (2) through (8) depict these trade
effects for seven slightly more disaggregated product categories. The consolidation of
the original 98 product groups into these seven product categories is based on sections
and individual product group definitions obtained from Schedule B reported by the U.S.
Census Bureau.22 A legend providing the necessary details for each of these product
categories is presented in Table 22. The estimation results at this disaggregated level are
intended to disentangle the aggregate effects observed in column (1) and help identify
whether the disaster induced trade disruptions are experienced across all industries or
rather idiosyncratic.
When considering the aggregate effects on the number of traded products presented
in column (1), it becomes clear that these results reflect the expected patterns. In the
short-run, the ports of New Orleans and Gulfport experience economically and statistically
significant reductions in the number of exported and imported products ranging from
21The remaining short-run to long-run port-specific treatment effects of ports located at greater
distances have been estimated and the results are available upon request. As expected, the majority of
these treatment effects for more remote ports are again statistically significant due to the large port-to-
port and month-to-month volatility in trade, but without any significant outliers or discernible patterns
relative to the primary ports under consideration.
22Schedule B is the official schedule of commodity classifications to be used by shippers in reporting
export shipments from the United States.
176
TABLE 20. Port-specific Effects on the Number of Exported Products
(1) (2) (3) (4) (5) (6) (7) (8)
Ports Total Category 1 Category 2 Category 3 Category 4 Category 5 Category 6 Category 7
Short-Run
Panama City, FL 47.781*** -0.122*** 3.335*** 9.587*** 5.335*** 12.768*** 6.843*** 8.326***
(0.316) (0.032) (0.038) (0.003) (0.011) (0.040) (0.061) (0.005)
Mobile, AL 7.914*** -1.803*** 0.427*** 1.109*** 1.038*** 1.535*** 2.556*** 3.486***
(0.101) (0.029) (0.015) (0.058) (0.014) (0.044) (0.042) (0.042)
New Orleans, LA -15.534*** -4.985*** -0.280*** -1.175*** -0.577*** -4.548*** -1.720*** -1.677***
(0.456) (0.006) (0.007) (0.082) (0.025) (0.110) (0.013) (0.110)
Gulfport, MS -15.661*** -2.239*** -3.148*** -2.782*** -1.500*** 1.440*** -3.744*** -2.770***
(0.591) (0.040) (0.045) (0.119) (0.039) (0.061) (0.016) (0.147)
Medium-Run
Panama City, FL 57.177*** 2.071*** 5.583*** 11.947*** 4.780*** 14.939*** 7.818*** 9.759***
(0.097) (0.002) (0.003) (0.030) (0.005) (0.004) (0.006) (0.014)
Mobile, AL 12.343*** -0.750*** 0.962*** 2.913*** 0.597*** 5.239*** 2.892*** 2.843***
(0.509) (0.013) (0.031) (0.081) (0.029) (0.050) (0.044) (0.054)
New Orleans, LA -8.206*** -2.480*** 0.086** -0.468*** -1.012*** -0.491*** -0.249*** -0.733***
(0.838) (0.023) (0.043) (0.117) (0.045) (0.092) (0.089) (0.095)
Gulfport, MS -7.156*** -1.566*** -0.450*** -0.497*** -1.149*** 3.297*** -1.628*** -1.841***
(0.960) (0.014) (0.054) (0.140) (0.048) (0.084) (0.125) (0.125)
Long-Run
Panama City, FL 59.267*** 3.254*** 5.452*** 12.081*** 6.001*** 14.511*** 8.574*** 9.228***
(0.319) (0.010) (0.039) (0.041) (0.017) (0.042) (0.024) (0.022)
Mobile, AL 15.751*** 0.835*** 2.467*** 2.642*** 0.899*** 4.779*** 2.903*** 2.245***
(0.850) (0.021) (0.066) (0.130) (0.048) (0.115) (0.071) (0.058)
New Orleans, LA -4.244*** -2.274*** 0.267*** -0.319 -0.283*** -0.265 -0.070 -0.402***
(1.305) (0.043) (0.098) (0.194) (0.062) (0.175) (0.120) (0.091)
Gulfport, MS -7.326*** -1.750*** 0.428*** -1.032*** -0.801*** 1.429*** -1.996*** -2.368***
(1.620) (0.044) (0.112) (0.244) (0.074) (0.187) (0.189) (0.127)
Random Effects Yes Yes Yes Yes Yes Yes Yes Yes
Spatial- 3rd 3rd 3rd 3rd 3rd 3rd 3rd 3rd
Weighting Order Order Order Order Order Order Order Order
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
177
15 to 31 lost product groups. Over time, however, these large initial reductions in the
portfolio of traded products revert back towards pre-treatment levels. As expected, New
Orleans experiences a more rapid recovery over the medium to long-run relative to the
port of Gulfport. In the long-run the estimated treatment effects for New Orleans, for
example, result in statistically significant reductions of the number of traded products by
4 and 6 for exports and imports, respectively, while Gulfport continues to suffer reductions
of 7 to 10 exported and imported product groups.
Considering the adjacent ports, the estimated treatment effects on the aggregate
number of traded products, once again, match the previous findings and provide
supporting evidence of the static and dynamic resilience of trade. In the short-run the
port of Panama City experiences substantial increases in the number of imported and
exported products ranging from 23 to 47, respectively, while the port of Mobile only adds
1 additional product on the import side and 7 on the export side. Expanding the time
horizon under consideration emphasizes the expected persistence of the positive gains
in the portfolio of products handled by the port of Panama City and the considerable
growth in the number of exported and imported products for the port of Mobile. That is,
the estimated average treatment effects over the first eight years after Hurricane Katrina
suggest a statistically significant and persistent increase in the number of exported and
imported products by 59 and 27 for the port of Panama City and 16 and 18 for the port of
Mobile, respectively.
At the disaggregated level, the port of New Orleans experiences a relatively even
reduction in all of the seven imported product categories with point estimates ranging
from 27% and 44% in the short-run. On the export side, however, we observe a very
heterogeneous response concerning the number of traded products that is driven by
decreases in Categories 1 (-33%) and 5 (-25%). This heterogeneity in treatment effects
178
continues to persist for the number of exported products and also manifests itself for
the number of imported products in the medium to long-run. While the number of
traded products recovers over time and reverts to pre-treatment levels for the majority of
product categories handled by the port of New Orleans, the remaining negative aggregate
treatment effect is almost exclusively driven by permanent reductions in trade of Animal
and Vegetable Products, shown in column (2) of Tables 20 and 21.
In contrast to New Orleans, the short-run effects for the port of Gulfport are fairly
evenly spread across all imported and exported categories. The exception to this rule
are textiles which experience a surprising increase in the number of exported products.23
While this relatively even distribution of treatment effects persists in the medium to long-
run for the number of exported products facilitated through Gulfport, the distribution of
the trade disruptions becomes rather bimodal for the number of imported products. In
particular, the estimates suggest that the sustained reduction in the number of imported
products and incomplete overall long-term recovery appears to be mainly driven by the
lack of imported Animal and Vegetable products and Textiles.
Turning the attention towards the positively affected ports of Panama City and
Mobile, the estimated category-specific treatment effects suggest rather heterogeneous
responses across industries. For Panama City, for example, all product categories but
Animal and Vegetable Products reveal economically and statistically significant increases
ranging from 37% to 71% in the short-run. In the medium to long-run, aggregate
increases in the total number of exported products are predominantly driven by changes
in categories 3 (Mineral products, etc.) and 5 (Textiles), which increase by 75% to 82%,
respectively, while Animal and Vegetable Products continue to be a disproportionately
small contributor to the increase in exported products. A similar pattern is observed on
23This surprising increase in the number of exported Textiles continues to be estimated even in the
long-run.
179
TABLE 21. Port-specific Effects on the Number of Imported Products
(1) (2) (3) (4) (5) (6) (7) (8)
Ports Total Category 1 Category 2 Category 3 Category 4 Category 5 Category 6 Category 7
Short-Run
Panama City, FL 23.927*** 2.757*** -0.240*** 0.282*** 3.074*** 9.135*** 3.429*** 3.785***
(0.266) (0.066) (0.001) (0.035) (0.031) (0.059) (0.051) (0.059)
Mobile, AL 0.609*** -0.405*** 0.139*** -0.663*** 0.384*** 0.958*** 0.181*** 0.061
(0.157) (0.044) (0.015) (0.036) (0.020) (0.001) (0.031) (0.037)
New Orleans, LA -31.928*** -4.317*** -2.567*** -5.978*** -3.182*** -7.855*** -3.999*** -3.788***
(0.081) (0.001) (0.013) (0.004) (0.003) (0.098) (0.001) (0.006)
Gulfport, MS -15.751*** -1.886*** -0.108*** -2.888*** -1.881*** -4.057*** -3.226*** -1.812***
(0.049) (0.044) (0.030) (0.031) (0.001) (0.061) (0.000) (0.021)
Medium-Run
Panama City, FL 24.154*** 3.432*** 1.178*** 1.320*** 2.900*** 7.796*** 2.914*** 3.501***
(0.058) (0.035) (0.003) (0.002) (0.009) (0.003) (0.000) (0.000)
Mobile, AL 6.959*** -0.391*** 1.148*** 0.198*** 0.497*** 2.801*** 1.695*** 1.481***
(0.113) (0.015) (0.005) (0.007) (0.000) (0.042) (0.014) (0.021)
New Orleans, LA -12.890*** -3.399*** -1.689*** -0.807*** -1.047*** -2.850*** -1.186*** -0.950***
(0.383) (0.010) (0.026) (0.014) (0.011) (0.106) (0.043) (0.049)
Gulfport, MS -10.621*** -1.991*** -0.298*** -1.568*** -0.909*** -2.183*** -1.960*** -0.952***
(0.294) (0.017) (0.011) (0.012) (0.010) (0.090) (0.045) (0.044)
Long-Run
Panama City, FL 27.493*** 4.358*** 2.290*** 1.301*** 3.592*** 8.187*** 2.400*** 4.628***
(0.217) (0.025) (0.022) (0.014) (0.002) (0.108) (0.013) (0.040)
Mobile, AL 17.805*** 0.455*** 2.259*** 0.814*** 1.934*** 6.569*** 3.256*** 2.931***
(0.384) (0.005) (0.030) (0.018) (0.006) (0.151) (0.013) (0.064)
New Orleans, LA -6.392*** -2.548*** -0.958*** -0.479*** 0.316*** -0.797** -0.093*** -0.747***
(0.865) (0.023) (0.067) (0.031) (0.023) (0.310) (0.036) (0.116)
Gulfport, MS -9.610*** -3.014*** 1.379*** -1.011*** -0.445*** -1.684*** -3.166*** -0.648***
(0.778) (0.009) (0.037) (0.021) (0.026) (0.289) (0.048) (0.107)
Random Effects Yes Yes Yes Yes Yes Yes Yes Yes
Spatial- 3rd 3rd 3rd 3rd 3rd 3rd 3rd 3rd
Weighting Order Order Order Order Order Order Order Order
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
180
the import side, where increases in the number of imported Textiles dominate the average
changes of the product portfolio across all time horizons under consideration (43%-51%).
The same holds true for the port of Mobile in the medium to long-run. Textiles are shown
to be the main factor underlying Mobile’s disaster induced export and import growth and
the most prominent contributor to the changes in the local trade composition increasing
by 27% and 36% in the long-run, respectively.
Overall, these results provide considerable insights into the previously described
trade disruptions caused by Hurricane Katrina. While the static and dynamic resilience of
trade is evidenced for both the value of trade and number of traded products, the product-
specific contributions to these offsetting trade disruptions in the short-run and overall
recovery in the long-run are very heterogeneous across industries. High value containers
of Textiles, for example, tend to be very resilient types of trade, whereas containerized
trade of Live Animals and Vegetable Products appears to be less resilient and suffers
disproportionately large disruptions from natural disaster. Given the fact that origin
and/or destination changes in containerized shipments of Live Animals and Vegetable
Products are rather costly, due to the required special handling and equipment, relative
to shipments of Textiles, this finding is quite intuitive.
Conclusion
The increasing presence and reliance of global economic output on international
transactions has lead to the significant growth of international trade and its exposure
to the omnipresent devastation caused by natural disasters. Frequent, yet uncertain
calamities continue to cause tremendous human and economic hardship, but have been
largely ignored in the trade literature. In the present study, I evaluate the impacts of
Hurricane Katrina on U.S. trade and find that natural disasters represent significant
181
TABLE 22. Product Category Legend Key
Product # of Sections Type of Products
Category Products included
1 15 1-3 Animal and Vegetable Products
2 9 4 Prepared Food Stuffs, Beverages, Spirits, Tobacco, etc.
3 16 5-7 Mineral, Chemical, Rubber and Plastic Products
4 9 8-10 Wood and Paper and products thereof, Leather, etc.
5 18 11, 12 Textiles, Footwear, Umbrellas, etc.
6 15 13-15 Bulk products including Stone, Plaster and Base Metals
7 14 16-19, 21 Work of Art, Manufactured products (i.e. Vehicles, etc.)
Source: Schedule B published by the U.S. Census Bureau
Notes: Excluded products groups include special classification provisions
barriers to international trade at the local port level. In conjunction with negligible
aggregate treatment effects, the estimated dynamic and spatially heterogeneous trade
disruptions point to the static and dynamic resilience of international trade.
The mechanisms underlying this trade resiliency include considerable rerouting
of both exports and imports and their rapid recovery due to product-specific path
dependence that lead to statistically and economically significant gains in trade for
those ports closest to the disaster stricken region and negatively disrupted ports. The
local disruptions are shown to be port-specific, of temporary nature for some ports
and permanent for others, heterogeneous across industries and offsetting in aggregate.
While the rerouting of exports and imports is shown to strongly depend on the distance
between the affected and non-affected ports, only rerouting of exports depends on port-
specific characteristics, such as harbor type or entry restrictions. In conjunction, these
empirical results illustrate the importance of a closely knit infrastructure network to
mitigate aggregate repercussions resulting from natural disasters, even the ones as
monumental as Hurricane Katrina. As developing countries continue to experience
significant aggregate disruptions from natural disasters, the empirical evidence pertaining
182
to the negating transport network effects of rerouting of internationally traded products
are of considerable interest to global policy makers.
Interestingly, the empirical findings point to an east/west dichotomy concerning
the significance and magnitude of the trade effects resulting from Hurricane Katrina.
Further inquiry may consider the role of hinterland transportation networks and other
infrastructure characteristics that drive this spatially heterogeneous response and resulting
resilience of international trade.
183
CHAPTER V
CONCLUSION
The overarching theme of the research presented in the substantive chapters of
this dissertation are the interconnections between trade and transportation. Exploring
these interconnections, I have shown that transportation costs are not only an integral
factor in the determination of trade, but also an endogenously determined component
of these international transactions. Building on this simultaneity between trade and
transportation, my work highlights the relevance of international transportation industries
in the determination of trade and transport policy outcomes and the spatial distribution
of trade disruptions stemming from natural disasters. Given the current political context
as well as the considerable number of recent natural disasters, my findings are pertinent
to numerous policy considerations and contribute to the economic literature across various
subfields.
Chapter II contributes to the international and transportation economics
literatures by developing a model that accounts for the simultaneity between trade and
transportation and by providing novel empirical evidence of the co-integration underlying
this endogeneity. Using advanced time-series techniques, I estimate the structural
equations governing the long-run equilibrium between trade and transport costs across
three major international markets. Adding to a growing strand of the international
economics literature that considers various financial and supply chain spillovers in
international markets, I use these estimates to offer insights into the transportation
induced spillover effects arising from the joint determination of fronthaul and backhaul
trade and transport costs.
While building on this research, Chapter III offers several additional contributions
to the economic literature. Extending the theoretical framework, I illustrate that the
184
previously highlighted integration of bilateral fronthaul and backhaul transport markets,
in fact, depends on the balance of bilateral trade. Deriving several comparative statics,
I illustrate the importance of this dependence in terms of policy outcomes. Empirical
tests of the theoretical predictions provide novel evidence supporting the hypotheses that
trade and transport policy outcomes systematically vary across fronthaul and backhaul
transport markets, various levels of bilateral trade imbalances and differentiated product
groups. Not only do these novel findings help explain otherwise unanticipated variation in
policy outcomes, but they also have significant implications for current and future policy
considerations, such as the Trade Facilitation Agreement by the WTO. Furthermore,
linking the results to global trade patterns points to considerable implications concerning
the potential trade effects across developing and developed countries, offers new insights
into the source for potential policy outcome inequality across these types of countries, and
raises a host of future research questions to be explored.
In contrast to the evaluation of trade and transport policy outcomes under an
endogenously responding transport sector, my research presented in Chapter IV considers
the influence of this sector in the determination of the short-run and long-run trade
disruptions caused by natural disasters. More specifically, I evaluate the dynamic impact
of Hurricane Katrina on trade at a disaggregated U.S. port level. The empirical results
pertaining to the spatial distribution of the disaster induced trade effects point to
offsetting short-run disruptions, dynamic recovery and some permanent trade route
alterations. In doing so, these findings are the first to evidence the static and dynamic
resilience of international trade. Exploring the underlying factors determining this
resilience, I identify distance as the key variable of interest. The results show that the
availability of excess port capacity in close proximity to a natural disasters is essential
to overcome the incurred reductions in trade and highlight the importance of a sufficient
infrastructure network. Overall, the findings suggest that aggregate analyses considering
185
national trade effects are misleading and conceal the significant local disruptions caused by
natural disasters.
Lastly, I would like to highlight a few of the related research questions arising
from this work and acknowledge the significant influence it will have on shaping my
future research agenda. Building on the results I have derived in Chapters II and III,
I intend to further explore the connections between environmental policies and trade.
More specifically, I plan to analyze the trade effects of the recently established Emission
Control Areas in North America and Europe, while evaluating the potentially varying
welfare implications across developed and developing countries. Secondly, I plan to
empirically address the question of whether trade imbalances and the resulting freight
rate differentials act as a dispersion force to the Home Market Effect and deter foreign
direct investment. Finally, following the analysis provided in Chapter IV, I will further
investigate the consequences arising from the local resilience of international trade. The
underlying question considers whether the resilience of trade triggers a creative destruction
effect in local communities proximate to but sparred by a given natural disaster.
In conclusion, it is my hope that the theoretical and empirical findings provided in
this dissertation and my future work will be valued by and useful to researchers, policy-
makers and practitioners alike and make a difference in the realm of international trade
and transportation.
186
APPENDIX A
PROOFS
Proof of Proposition 1
Consider three countries, i, j, k, with imbalanced bilateral trade, such that transport
market ik facilitating trade from country i to country k is considered a fronthaul and
transport market ij facilitating trade from country i to country j is considered a backhaul,
Qik > Qki and Qij < Qji. Suppose that for these three countries, i, j, k, we have aij = aik
and τij = τik. Given these assumptions and the fact that JC
′ > 0, equation (3.13) shows
that ∣∣∣∣∂qij∂aij aijqij
∣∣∣∣ = σ aijpiτij + aij > σ aikpiτik + aik + JC ′ =
∣∣∣∣∂qik∂aik aikqik
∣∣∣∣ . (A.1)
This provides the proof that when equilibrium trade is imbalanced, trade facilitated in
fronthaul transport markets is more inelastic than otherwise identical trade facilitated in
backhaul transport markets.
Proof of Proposition 2
Recall that, in the imbalanced trade case, the elasticity of trade facilitated
in backhaul transport markets with respect to marginal access costs is negative,(
∂qij
∂aij
aij
qij
< 0
)
. In order for imbalanced trade in higher valued products to be less
responsive to a shock in marginal access costs than trade of lower valued products, the
elasticity of trade facilitated in backhaul transport markets given by equation (3.13) must
be increasing in the domestic sales price, pi:
∂
(
∂qij
∂aij
aij
qij
)
∂pi
= σ
aijτij
(piτij + aij)
2 > 0. (A.2)
187
APPENDIX B
DERIVATION OF TRADE ELASTICITIES IN THE VALUE CASE
The total value of county i’s exports to country j, denoted by xij, is defined as:
xij = pijqij = (τijpi + fij)
[
σ
σ − 1 (τijpi + fij)
]−σ
=
(
σ
σ − 1
)−σ
[(τijpi + fij)]
1−σ . (B.1)
Following the theoretical derivations presented in subsection 4.3, I initially consider a
shock to marginal access cost. In the balanced bilateral trade case, the elasticity of the
value of trade with respect to a change in marginal access costs is given by
∂xij
∂aij
aij
xij
= (1− σ) aij
piτij + pjτji + aij + aji + JC ′
< 0 if Qij = Qji, (B.2)
while in the imbalanced trade case, this elasticity can be represented as follows:
∂xij
∂aij
aij
xij
= (1− σ) aij
(piτij + fij)
∂fij
∂aij
=
 (1− σ)
aij
piτij+aij+JC′
< 0 if Qij > Qji (fronthaul)
(1− σ) aij
piτij+aij
< 0 if Qij < Qji (backhaul).
(B.3)
Next, I consider the effects of a change in marginal joint costs. Again, I differentiate
between the balanced and imbalanced trade cases but now consider the effects on
the value of trade instead of the volume. In the balanced case, I obtain the following
expression for the elasticity of the value of trade with respect to marginal joint costs:
∂xij
∂JC ′
JC ′
xij
= (1− σ) JC
′
piτij + pjτji + aij + aji + JC ′
if Qij = Qji, (B.4)
188
whereas in the imbalanced trade case this elasticity of trade can be represented as:
∂xij
∂JC ′
JC ′
xij
= (1− σ) JC
′
(piτij + fij)
∂fij
∂JC ′
=
 (1− σ)
JC′
piτij+aij+JC′
if Qij > Qji (fronthaul)
0 if Qij < Qji (backhaul).
(B.5)
A comparison between equation (3.12) and (B.2), (3.13) and (B.3), (3.14) and (B.4),
as well as (3.15) and (B.5) reveals that the trade elasticities with respect to marginal
access as well as joint cost in the volume and value cases are solely distinguished by the
scaling factors of −σ and (1 − σ), respectively. It is trivial to show that Proposition 1
and Proposition 2 continue to hold when the value rather than the volume of trade is
considered. Interestingly, these theoretical derivations also suggest that the response in
the value of trade is smaller than the response in the volume of trade, |−σ| > |1− σ|.
189
APPENDIX C
ADDITIONAL TABLES AND FIGURES
Chapter III Robustness Analyses
TABLE 23. Robustness Analysis - Time Trend Inclusive Specification
(1) (2) (3) (4) (5)
VARIABLES Full Sample Imports Exports Fronthaul Backhaul
Panel 1 - Primary Results -0.030 -0.010 -0.083** -0.029 -0.104**
ATE, (δ) (0.033) (0.041) (0.040) (0.040) (0.045)
Panel 2 - Treatment/Control -0.045 -0.027 -0.085** -0.044 -0.109**
Group Time Trends (0.037) (0.046) (0.040) (0.045) (0.047)
Panel 3 - Country-Specific -0.037 0.005 -0.078* -0.038 -0.111**
Time Trends (0.039) (0.043) (0.044) (0.045) (0.052)
Observations 410,134 205,067 205,067 205,067 205,067
Port FE Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
190
TABLE 24. Robustness Analysis - Varying Levels of Clustered Standard Errors
(1) (2) (3) (4) (5) (6) (7) (8)
VARIABLES Fronthaul Backhaul Fronthaul Backhaul Fronthaul Backhaul Fronthaul Backhaul
ATE, (δ) -0.029 -0.104** -0.029 -0.104* -0.029 -0.104*** -0.029 -0.104***
(0.040) (0.045) (0.039) (0.055) (0.030) (0.040) (0.025) (0.036)
Level of Clustering Port- Port- State- State- Ports Ports State State
Country Country Country Country
Number of Clusters 1419 1419 726 726 43 43 22 22
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
TABLE 25. Robustness Analysis - Varying Fixed Effects Specifications
(1) (2) (3) (4) (5) (6) (7) (8)
VARIABLES Fronthaul Backhaul Fronthaul Backhaul Fronthaul Backhaul Fronthaul Backhaul
ATE, (δ) -0.029 -0.104** 0.006 -0.080* -0.021 -0.091** -0.018 -0.091*
(0.040) (0.045) (0.037) (0.042) (0.035) (0.046) (0.033) (0.047)
Port FE Yes Yes Yes Yes Yes Yes No No
Time FE Yes Yes No No No No No No
Imp-Exp FE Yes Yes Yes Yes Yes Yes Yes Yes
Region-Time FE No No Yes Yes No No No No
State-Time FE No No No No Yes Yes No No
Port-Time FE No No No No No No Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
191
TABLE 26. Robustness Analysis - Various Sample Restrictions
(1) (2) (3) (4) (5)
Full Sample Imports Exports Fronthaul Backhaul
Panel 1 - Primary Results -0.030 -0.010 -0.083** -0.029 -0.104**
ATE, (δ) (0.033) (0.041) (0.040) (0.040) (0.045)
Panel 2 - Incl. Landlocked -0.020 0.006 -0.088** -0.018 -0.100**
European Countries (0.032) (0.039) (0.040) (0.038) (0.045)
Panel 3 - Incl. North -0.022 0.004 -0.092** -0.019 -0.101**
American Countries (0.032) (0.039) (0.039) (0.038) (0.045)
Panel 4 - Excl. China & -0.048 -0.031 -0.093* -0.062 -0.100*
Japan (0.042) (0.051) (0.048) (0.045) (0.057)
Port FE Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
192
TABLE 27. Robustness Analysis - Alternative Empirical Model Specifications
(1) (2) (3) (4) (5)
Full Sample Imports Exports Fronthaul Backhaul
Panel 1 - Primary Results -0.030 -0.010 -0.083** -0.029 -0.104**
ATE, (δ) (0.033) (0.041) (0.040) (0.040) (0.045)
Panel 2 - Excl. -0.025 -0.010 -0.083** -0.021 -0.092**
U.S. State Employment (0.033) (0.042) (0.040) (0.040) (0.047)
Panel 3 - Excl. -0.031 -0.012 -0.083** -0.029 -0.111**
U.S. FTAs (0.034) (0.041) (0.040) (0.040) (0.049)
Panel 4 - Incl. -0.041 -0.013 -0.086** -0.029 -0.107**
Lagged Control Variables (0.034) (0.043) (0.039) (0.043) (0.046)
Panel 5 - -0.028 -0.005 -0.091** -0.022 -0.117**
Aggregated Quarterly Data (0.034) (0.042) (0.040) (0.042) (0.049)
Port FE Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
TABLE 28. Robustness Analysis - Varying Backhaul Identifications
(1) (2) (3) (4) (5) (6)
FH - BH - FH - BH - FH- BH -
VARIABLES Primary Primary Exp. Lag Exp. Lag Imp. Lag Imp. Lag
ATE, (δ) -0.029 -0.104** -0.027 -0.114** -0.040 -0.103**
(0.040) (0.045) (0.041) (0.047) (0.039) (0.047)
Port FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
193
TABLE 29. Robustness Analysis - BH Identification at Varying Levels of Aggregation
(1) (2) (3) (4) (5)
Full Sample Imports Exports Fronthaul Backhaul
Panel 1 - Primary Results -0.030 -0.010 -0.083** -0.029 -0.104**
ATE, (δ) (0.033) (0.041) (0.040) (0.040) (0.045)
Panel 2 - Supranational -0.054 -0.050 -0.078 -0.053 -0.098*
ATE, (δ) (0.036) (0.031) (0.069) (0.039) (0.059)
Panel 3 - U.S. State - -0.054 -0.050 -0.078 -0.043 -0.105**
Supranational (0.033) (0.032) (0.048) (0.036) (0.045)
Panel 4 - U.S. Region - -0.045* -0.050 -0.078*** -0.068* -0.096**
Supranational (0.027) (0.033) (0.029) (0.036) (0.038)
Port FE Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes
Imp-Exp FE Yes Yes Yes Yes Yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Chapter IV Robustness Analyses
194
TABLE 30. Distances between U.S. Ports of Entry
T
a
co
m
a,
W
A
S
ea
tt
le
,
W
A
P
o
rt
la
n
d
,
O
R
S
an
F
ra
n
ci
sc
o
,
C
A
R
ic
h
m
o
n
d
,
C
A
O
a
k
la
n
d
,
C
A
P
o
rt
H
u
en
em
e,
C
A
L
os
A
n
ge
le
s,
C
A
L
on
g
B
ea
ch
,
C
A
S
a
n
D
ie
g
o
,
C
A
Tacoma 0
Seattle 25 0
Portland, OR 387 362 0
San Francisco 1039 1014 652 0
Richmond 1050 1025 663 11 0
Oakland 1064 1039 677 25 14 0
Port Hueneme 1379 1354 992 340 329 315 0
Los Angeles 1441 1416 1054 402 391 377 62 0
Long Beach 1444 1419 1057 405 394 380 65 3 0
San Diego 1538 1513 1151 499 488 474 159 97 94 0
Corpus Christi 6000 5975 5613 4961 4950 4936 4621 4559 4556 4462
Freeport 6160 6135 5773 5121 5110 5096 4781 4719 4716 4622
Galveston 6244 6219 5857 5205 5194 5180 4865 4803 4800 4706
Houston 6291 6266 5904 5252 5241 5227 4912 4850 4847 4753
Port Arthur 6423 6398 6036 5384 5373 5359 5044 4982 4979 4885
New Orleans 6864 6839 6477 5825 5814 5800 5485 5423 5420 5326
Gulfport 7135 7110 6748 6096 6085 6071 5756 5694 5691 5597
Mobile 7233 7208 6846 6194 6183 6169 5854 5792 5789 5695
Panama City 7426 7401 7039 6387 6376 6362 6047 5985 5982 5888
Tampa 7691 7666 7304 6652 6641 6627 6312 6250 6247 6153
Miami 8134 8109 7747 7095 7084 7070 6755 6693 6690 6596
Port Everglades 8161 8136 7774 7122 7111 7097 6782 6720 6717 6623
Palm Beach 8207 8182 7820 7168 7157 7143 6828 6766 6763 6669
Jacksonville 8463 8438 8076 7424 7413 7399 7084 7022 7019 6925
Brunswick 8545 8520 8158 7506 7495 7481 7166 7104 7101 7007
Savannah 8649 8624 8262 7610 7599 7585 7270 7208 7205 7111
Charleston 8751 8726 8364 7712 7701 7687 7372 7310 7307 7213
Wilmington 8902 8877 8515 7863 7852 7838 7523 7461 7458 7364
Newport News 9262 9237 8875 8223 8212 8198 7883 7821 7818 7724
Baltimore 9432 9407 9045 8393 8382 8368 8053 7991 7988 7894
Chester 9509 9484 9122 8470 8459 8445 8130 8068 8065 7971
Philadelphia 9524 9499 9137 8485 8474 8460 8145 8083 8080 7986
Perth Amboy 9744 9719 9357 8705 8694 8680 8365 8303 8300 8206
Newark 9759 9734 9372 8720 8709 8695 8380 8318 8315 8221
New York City 9771 9746 9384 8732 8721 8707 8392 8330 8327 8233
Boston 10157 10132 9770 9118 9107 9093 8778 8716 8713 8619
Portland, ME 10257 10232 9870 9218 9207 9193 8878 8816 8813 8719
Detroit 12146 12121 11759 11107 11096 11082 10767 10705 10702 10608
Port Huron 12208 12183 11821 11169 11158 11144 10829 10767 10764 10670
Chicago 12779 12754 12392 11740 11729 11715 11400 11338 11335 11241
Continued on next page
195
C
or
p
u
s
C
h
ri
st
i,
T
X
F
re
ep
o
rt
,
T
X
G
a
lv
es
to
n
,
T
X
H
o
u
st
on
,
T
X
P
or
t
A
rt
h
u
r,
T
X
N
ew
O
rl
ea
n
s,
L
A
G
u
lf
p
o
rt
,
M
S
M
ob
il
e,
A
L
P
an
a
m
a
C
it
y,
F
L
T
a
m
p
a,
F
L
M
ia
m
i,
F
L
Corpus Christi 0
Freeport 160 0
Galveston 244 84 0
Houston 291 131 47 0
Port Arthur 423 263 179 132 0
New Orleans 864 704 620 573 441 0
Gulfport 1135 975 891 844 712 271 0
Mobile 1233 1073 989 942 810 369 98 0
Panama City 1426 1266 1182 1135 1003 562 291 193 0
Tampa 1691 1531 1447 1400 1268 827 556 458 265 0
Miami 2134 1974 1890 1843 1711 1270 999 901 708 443 0
Port Everglades 2161 2001 1917 1870 1738 1297 1026 928 735 470 27
Palm Beach 2207 2047 1963 1916 1784 1343 1072 974 781 516 73
Jacksonville 2463 2303 2219 2172 2040 1599 1328 1230 1037 772 329
Brunswick 2545 2385 2301 2254 2122 1681 1410 1312 1119 854 411
Savannah 2649 2489 2405 2358 2226 1785 1514 1416 1223 958 515
Charleston 2751 2591 2507 2460 2328 1887 1616 1518 1325 1060 617
Wilmington 2902 2742 2658 2611 2479 2038 1767 1669 1476 1211 768
Newport News 3262 3102 3018 2971 2839 2398 2127 2029 1836 1571 1128
Baltimore 3432 3272 3188 3141 3009 2568 2297 2199 2006 1741 1298
Chester 3509 3349 3265 3218 3086 2645 2374 2276 2083 1818 1375
Philadelphia 3524 3364 3280 3233 3101 2660 2389 2291 2098 1833 1390
Perth Amboy 3744 3584 3500 3453 3321 2880 2609 2511 2318 2053 1610
Newark 3759 3599 3515 3468 3336 2895 2624 2526 2333 2068 1625
New York City 3771 3611 3527 3480 3348 2907 2636 2538 2345 2080 1637
Boston 4157 3997 3913 3866 3734 3293 3022 2924 2731 2466 2023
Portland, ME 4257 4097 4013 3966 3834 3393 3122 3024 2831 2566 2123
Detroit 6146 5986 5902 5855 5723 5282 5011 4913 4720 4455 4012
Port Huron 6208 6048 5964 5917 5785 5344 5073 4975 4782 4517 4074
Chicago 6779 6619 6535 6488 6356 5915 5644 5546 5353 5088 4645
Continued on next page
196
P
or
t
E
v
er
gl
ad
es
,
F
L
P
a
lm
B
ea
ch
,
F
L
J
ac
k
so
n
v
il
le
,
F
L
B
ru
n
sw
ic
k
,
G
A
S
av
an
n
ah
,
G
A
C
h
ar
le
st
o
n
,
S
C
W
il
m
in
gt
on
,
N
C
N
ew
p
o
rt
N
ew
s,
V
A
B
a
lt
im
o
re
,
M
D
C
h
es
te
r,
P
A
P
h
il
ad
el
p
h
ia
,
P
A
Port Everglades 0
Palm Beach 46 0
Jacksonville 302 256 0
Brunswick 384 338 82 0
Savannah 488 442 186 104 0
Charleston 590 544 288 206 102 0
Wilmington 741 695 439 357 253 151 0
Newport News 1101 1055 799 717 613 511 360 0
Baltimore 1271 1225 969 887 783 681 530 170 0
Chester 1348 1302 1046 964 860 758 607 247 77 0
Philadelphia 1363 1317 1061 979 875 773 622 262 92 15 0
Perth Amboy 1583 1537 1281 1199 1095 993 842 482 312 235 220
Newark 1598 1552 1296 1214 1110 1008 857 497 327 250 235
New York City 1610 1564 1308 1226 1122 1020 869 509 339 262 247
Boston 1996 1950 1694 1612 1508 1406 1255 895 725 648 633
Portland, ME 2096 2050 1794 1712 1608 1506 1355 995 825 748 733
Detroit 3985 3939 3683 3601 3497 3395 3244 2884 2714 2637 2622
Port Huron 4047 4001 3745 3663 3559 3457 3306 2946 2776 2699 2684
Chicago 4618 4572 4316 4234 4130 4028 3877 3517 3347 3270 3255
P
er
th
A
m
b
oy
,
N
J
N
ew
ar
k
,
N
J
N
ew
Y
o
rk
C
it
y,
N
Y
B
os
to
n
,
M
A
P
or
tl
a
n
d
,
M
E
D
et
ro
it
,
M
I
P
or
t
H
u
ro
n
,
M
I
C
h
ic
a
go
,
IL
Perth Amboy 0
Newark 15 0
New York City 27 12 0
Boston 413 398 386 0
Portland, ME 513 498 486 100 0
Detroit 2402 2387 2375 1989 1889 0
Port Huron 2464 2449 2437 2051 1951 62 0
Chicago 3035 3020 3008 2622 2522 633 571 0
Source: Department of Commerce, NOAA & National Ocean Service
197
TABLE 31. Aggregate Trade Disruptions - Various Spatial Weights
(1) (2) (3) (4) (5) (6) (7) (8)
VARIABLES Export Import Export Import Export Import Export Import
SAR - ATE 0.281 -0.074 0.300 -0.075 0.298 -0.087 0.287 -0.075
(0.257) (0.134) (0.260) (0.134) (0.258) (0.134) (0.256) (0.134)
SEM - ATE 0.286 -0.074 0.286 -0.074 0.286 -0.073 0.285 -0.075
(0.261) (0.134) (0.244) (0.132) (0.245) (0.112) (0.257) (0.133)
SAC - ATE 0.284 -0.074 0.293 -0.075 0.292 -0.080 0.287 -0.075
(0.259) (0.134) (0.252) (0.133) (0.251) (0.122) (0.258) (0.133)
Observations 1,960 1,960 1,960 1,960 1,960 1,960 1,960 1,960
Trend Yes Yes Yes Yes Yes Yes Yes Yes
Port FE Yes Yes Yes Yes Yes Yes Yes Yes
Month FE Yes Yes Yes Yes Yes Yes Yes Yes
Time FE No No No No No No No No
Spatial- 1st 1st 2nd 2nd 4th 4th Inv. Inv.
Weighting Order Order Order Order Order Order Dist. Dist.
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
198
TABLE 32. Regional Trade Disruptions - Various Spatial Weights
(1) (2) (3) (4) (5) (6) (7) (8)
VARIABLES Export Import Export Import Export Import Export Import
SAR
Gulf Coast 0.065 0.114 0.132 0.126 0.174 0.216 0.059 0.112
(0.253) (0.408) (0.239) (0.408) (0.240) (0.384) (0.249) (0.406)
Lower Atlantic 1.003 0.618 1.052 0.630 1.082 0.693 0.997 0.615
(0.920) (0.487) (0.926) (0.490) (0.928) (0.488) (0.913) (0.486)
SEM
Gulf Coast 0.047 0.113 0.147 0.126 0.203 0.220 0.061 0.111
(0.265) (0.405) (0.237) (0.395) (0.235) (0.326) (0.250) (0.406)
Lower Atlantic 1.008 0.612 0.958 0.601 0.959 0.541 1.001 0.614
(0.942) (0.484) (0.847) (0.466) (0.824) (0.374) (0.925) (0.486)
SAC
Gulf Coast -0.116 0.109 -0.345 0.122 0.719* 0.213 0.062 0.113
(0.772) (0.447) (1.487) (0.431) (0.397) (0.410) (0.250) (0.407)
Lower Atlantic 1.278 0.672 1.557 0.654 1.271 0.704 1.003 0.616
(1.777) (0.619) (2.118) (0.563) (1.105) (0.603) (0.926) (0.487)
Observations 1,960 1,960 1,960 1,960 1,960 1,960 1,960 1,960
Port FE Yes Yes Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes Yes Yes
Spatial- 1st 1st 2nd 2nd 4th 4th Inv. Inv.
Weighting Order Order Order Order Order Order Dist. Dist.
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
TABLE 33. Port-Specific Trade Disruptions - SAC
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
Seattle, WA -0.121*** 0.019*** 0.047*** -0.062*** -0.131*** -0.053***
(0.003) (0.006) (0.002) (0.003) (0.020) (0.005)
Portland, OR 0.126*** 0.120*** -0.237*** 0.368*** -0.357*** 0.230***
(0.011) (0.005) (0.003) (0.004) (0.026) (0.009)
Oakland, CA 0.052*** 0.010*** 0.091*** -0.015*** 0.197*** 0.028**
(0.017) (0.003) (0.006) (0.006) (0.040) (0.013)
Richmond, CA 0.016 0.287*** 0.302*** -0.671*** 0.803*** -0.168***
(0.016) (0.001) (0.005) (0.004) (0.029) (0.011)
San Francisco, CA 0.652*** -0.388*** -0.164*** -0.185*** -0.595*** 0.210***
Continued on next page
199
Table 33 – Continued
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
(0.024) (0.008) (0.006) (0.005) (0.041) (0.016)
Port Hueneme, CA -0.017*** -0.281*** 0.944*** -1.047*** 1.357*** -0.677***
(0.006) (0.011) (0.003) (0.002) (0.026) (0.005)
Los Angeles, CA -0.141*** -0.022*** 0.181*** 0.001 0.243*** 0.084***
(0.016) (0.006) (0.005) (0.003) (0.073) (0.002)
Long Beach, CA -0.132*** 0.033*** 0.239*** -0.015*** 0.302*** -0.047***
(0.016) (0.000) (0.004) (0.002) (0.065) (0.004)
San Diego, CA 0.733*** 0.527*** -0.435*** 0.608*** -0.594*** 0.324***
(0.023) (0.005) (0.006) (0.001) (0.078) (0.004)
Chicago, IL 0.131*** -0.370*** 0.205*** -0.004 -2.503*** 0.360***
(0.048) (0.008) (0.002) (0.009) (0.035) (0.092)
Port Huron, MI 0.060 -0.268*** -0.285*** -0.090*** -0.108* -0.025
(0.043) (0.003) (0.003) (0.006) (0.057) (0.060)
Detroit, MI 0.089** 0.210*** 0.002 0.125*** -0.791*** 0.237***
(0.042) (0.004) (0.003) (0.006) (0.029) (0.049)
Portland, ME -0.336*** 0.140*** -0.312*** -1.462*** -1.961*** -4.011***
(0.034) (0.002) (0.005) (0.001) (0.004) (0.005)
Boston, MA 0.078 -0.140*** 0.297*** 0.092*** 0.078*** 0.238***
(0.051) (0.001) (0.004) (0.010) (0.009) (0.035)
New York, NY 0.093* -0.049*** 0.269*** -0.051*** 0.323*** 0.248***
(0.050) (0.001) (0.006) (0.009) (0.010) (0.033)
Newark, NJ 0.038 -0.045*** 0.389*** -0.023** 0.393*** 0.112***
(0.051) (0.004) (0.007) (0.009) (0.022) (0.033)
Perth Amboy, NJ -0.387*** -0.547*** 0.515*** -1.468*** 0.807*** 1.004***
(0.037) (0.000) (0.009) (0.001) (0.040) (0.005)
Philadelphia, PA 0.152*** -0.083*** 0.367*** 0.162*** 0.267*** 0.207***
(0.047) (0.004) (0.009) (0.005) (0.047) (0.012)
Chester, PA -0.103*** -0.140*** 0.536*** 0.042*** 0.468*** 0.182***
(0.025) (0.005) (0.008) (0.004) (0.051) (0.019)
Baltimore, MD -0.081*** -0.051*** 0.217*** 0.065*** 0.242*** 0.218***
(0.025) (0.006) (0.009) (0.004) (0.048) (0.017)
Newport News, VA 0.014 0.013*** 0.237*** 0.051*** 0.229*** 0.107***
(0.020) (0.001) (0.008) (0.002) (0.046) (0.013)
Wilmington, NC 0.994*** -0.196*** 0.324*** 0.272*** 0.867*** 0.881***
(0.055) (0.000) (0.006) (0.002) (0.040) (0.010)
Charleston, SC -0.036 -0.032*** 0.101*** -0.005*** -0.047 0.030***
(0.032) (0.001) (0.006) (0.001) (0.049) (0.005)
Savannah, GA -0.042 0.071*** 0.348*** 0.201*** 0.599*** 0.505***
(0.032) (0.001) (0.005) (0.003) (0.041) (0.014)
Brunswick, GA -0.825*** -0.340*** -0.269*** -0.750*** 0.312*** -1.049***
(0.016) (0.005) (0.008) (0.001) (0.046) (0.003)
Jacksonville, FL 0.167*** 0.104*** 0.283*** 0.114*** 0.368*** 0.504***
(0.052) (0.002) (0.005) (0.004) (0.036) (0.028)
Palm Beach, FL 0.065 -0.140*** 0.310*** -0.423*** 0.084* -0.914***
(0.053) (0.008) (0.006) (0.002) (0.044) (0.003)
Port Everglades, FL -0.406*** 0.145*** 0.433*** 0.026** 0.486*** -0.046
(0.098) (0.029) (0.038) (0.012) (0.138) (0.039)
Continued on next page
200
Table 33 – Continued
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
Miami, FL -0.611*** -0.037 0.083** -0.148*** -0.048 -0.114*
(0.119) (0.041) (0.041) (0.016) (0.153) (0.063)
Corpus Christi, TX -0.082*** -0.564*** -0.259*** 2.331*** 2.954*** 1.382***
(0.010) (0.011) (0.003) (0.001) (0.032) (0.006)
Houston, TX 0.177* 0.015 0.324*** 0.139*** 0.421*** 0.194***
(0.091) (0.044) (0.005) (0.004) (0.053) (0.037)
Freeport, TX 0.080*** -0.134*** 0.152*** -0.786*** 0.017 -0.674***
(0.026) (0.009) (0.007) (0.004) (0.062) (0.016)
Galveston, TX 0.700*** 0.017 -0.069*** 0.250*** 0.582*** 0.441***
(0.060) (0.017) (0.007) (0.001) (0.054) (0.028)
Port Arthur, TX -0.545*** 0.162*** 1.736*** -1.525*** 0.217*** -2.986***
(0.074) (0.033) (0.003) (0.005) (0.038) (0.005)
Tampa, FL -0.702*** 0.231*** 0.480*** 0.415*** 0.437*** 0.913***
(0.094) (0.016) (0.037) (0.012) (0.136) (0.038)
Panama City, FL 7.625*** 2.102*** 10.578*** 5.230*** 10.202*** 5.036***
(0.090) (0.028) (0.004) (0.003) (0.036) (0.003)
Mobile, AL -0.179*** 0.522*** 0.596*** 0.422*** 1.308*** 1.077***
(0.044) (0.009) (0.040) (0.007) (0.125) (0.014)
New Orleans, LA -1.757*** -1.908*** 0.041 -0.464*** 0.125 0.034
(0.088) (0.019) (0.041) (0.010) (0.144) (0.025)
Gulfport, MS -1.729*** -1.788*** -0.566*** -0.761*** -0.815*** -0.682***
(0.073) (0.020) (0.045) (0.012) (0.152) (0.038)
Observations 1,120 1,120 1,960 1,960 4,840 4,840
R-squared 0.000 0.014 0.000 0.012 0.000 0.064
Port FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Spatial- 3rd 3rd 3rd 3rd 3rd 3rd
Weighting Order Order Order Order Order Order
Convergence yes yes yes yes yes yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
201
TABLE 34. Port Characteristics Legend Key
Entrance Restrictions
Harbor Type Tide Other
0 - Coastal (Breakwater)
1 - Coastal (Natural) 0 - No 0 - No
2 - Coastal (Tide Gates) 1- Yes 1- Yes
3 - Canal or Lake
4 - River (Basin)
5 - River (Natural)
6 - River (Tide Gates)
Source: 2015 WPI - National Geospatial-Intelligence Agency
TABLE 35. Influence of Port Characteristics on Treatment Effects - SAC
(1) (2) (3) (4) (5) (6)
Short-Run Medium-Run Long-Run
VARIABLES Export Import Export Import Export Import
HT (=1) -0.682 -0.069 -1.568** -0.112 -1.640** -0.423
(0.551) (0.203) (0.666) (0.528) (0.665) (0.565)
HT (=2) 0.011 -0.284** -0.475* 0.244 -3.389*** 0.289
(0.213) (0.103) (0.278) (0.405) (0.360) (0.421)
HT (=3) -2.481** -0.257 -1.764 -2.889*** -3.846*** -4.723***
(1.166) (0.353) (1.433) (0.851) (1.323) (0.915)
HT (=4) -0.449 -0.061 -0.860* -0.100 -1.135** -0.296
(0.351) (0.131) (0.444) (0.447) (0.479) (0.475)
HT (=5) -0.394 -0.158 -1.100** -0.200 -1.274** 0.026
(0.396) (0.151) (0.470) (0.479) (0.491) (0.493)
HT (=6) -1.570 -0.351 -2.882** -1.056 -3.311** -1.369
(1.056) (0.322) (1.298) (0.788) (1.205) (0.848)
ER - Tide -0.630* -0.005 -0.643 -0.298 -0.721* -0.835*
(0.350) (0.124) (0.396) (0.341) (0.401) (0.450)
ER - Other -1.026 -0.199 -1.383* -0.156 -0.692 0.160
(0.660) (0.206) (0.785) (0.456) (0.726) (0.641)
Inv. Dist. 985.466 274.106 1,530.182** 874.352** 1,723.503** 903.609**
(584.907) (173.251) (714.277) (375.725) (644.449) (394.423)
Observations 37 37 37 37 37 37
R-squared 0.508 0.403 0.618 0.509 0.656 0.540
Std. Errors Robust Robust Robust Robust Robust Robust
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
202
TABLE 36. Treatment Effect over Distance
(1) (2) (3) (4) (5) (6) (7) (8)
PPML SAR SEM SAC
VARIABLES Export Import Export Import Export Import Export Import
Core (LA, MS) 0.286*** -0.145*** 0.366*** -0.128 0.174* -0.200* 0.965** -0.098
(0.025) (0.008) (0.123) (0.113) (0.089) (0.106) (0.453) (0.147)
Core*Inv. Dist. -17.140*** -8.688*** -15.179*** -7.531*** -15.542*** -7.778*** -13.822*** -7.430***
(0.000) (0.000) (0.274) (0.066) (0.763) (0.575) (3.865) (0.415)
Periphery -0.207 -0.153 -3.869*** -1.838*** -3.996*** -1.819*** -4.214*** -1.851***
(GA, FL, SC, TX) (0.173) (0.184) (1.144) (0.658) (1.243) (0.668) (1.052) (0.677)
Periphery* 189.283* 158.799 2,757.220*** 1,359.084*** 2,807.832*** 1,343.031*** 2,946.219*** 1,367.465***
Inv. Dist. (107.726) (135.222) (740.791) (414.671) (801.494) (421.506) (649.930) (420.849)
Observations 1,960 1,960 1,960 1,960 1,960 1,960 1,960 1,960
R-squared 0.989 0.996 0.010 0.015 0.011 0.016 0.009 0.015
Port FE Yes Yes Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes Yes Yes
Spatial- None None 3rd 3rd 3rd 3rd 3rd 3rd
Weighting Order Order Order Order Order Order
Convergence yes yes yes yes yes yes yes yes
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
203
−
10
−
5
0
5
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
2nd Order Neighbor − West (Port Arthur (TX))
(a) Exports
−
10
−
5
0
5
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
2nd Order Neighbor − West (Port Arthur (TX))
(b) Imports
FIGURE 18. Dynamic Treatment Effects - Port Arthur, TX
−
1
0
1
2
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
3rd Order Neighbor − West (Houston (TX))
(a) Exports
−
1.
5
−
1
−
.
5
0
.
5
1
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
3rd Order Neighbor − West (Houston (TX))
(b) Imports
FIGURE 19. Dynamic Treatment Effects - Houston, TX
−
4
−
2
0
2
4
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
4th Order Neighbor − West (Galveston (TX))
(a) Exports
−
2
−
1
0
1
2
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
4th Order Neighbor − West (Galveston (TX))
(b) Imports
FIGURE 20. Dynamic Treatment Effects - Galveston, TX
204
−
1
0
1
2
3
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
4th Order Neighbor − East (Tampa (FL))
(a) Exports
−
2
−
1
0
1
2
3
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
4th Order Neighbor − East (Tampa (FL))
(b) Imports
FIGURE 21. Dynamic Treatment Effects - Tampa, FL
−
2
−
1
0
1
2
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
5th Order Neighbor − West (Freeport (TX))
(a) Exports - Freeport, TX
−
2
−
1
0
1
2
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
5th Order Neighbor − West (Freeport (TX))
(b) Imports - Freeport, TX
FIGURE 22. Dynamic Treatment Effects - Freeport, TX
−
1
−
.
5
0
.
5
1
1.
5
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
5th Order Neighbor − East (Miami (FL))
(a) Exports
−
1
−
.
5
0
.
5
1
1.
5
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
5th Order Neighbor − East (Miami (FL))
(b) Imports
FIGURE 23. Dynamic Treatment Effects - Miami, FL
205
0
5
10
15
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
6th Order Neighbor − West (Corpus Christi (TX))
(a) Exports
−
15
−
10
−
5
0
5
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
6th Order Neighbor − West (Corpus Christi (TX))
(b) Imports
FIGURE 24. Dynamic Treatment Effects - Corpus Christi, TX
−
.
5
0
.
5
1
1.
5
2
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
6th Order Neighbor − East (Port Everglades (FL))
(a) Exports
−
1
−
.
5
0
.
5
1
1.
5
Tr
ea
tm
en
t E
ffe
ct
−50 0 50 100
Month relative to Katrina
6th Order Neighbor − East (Port Everglades (FL))
(b) Imports
FIGURE 25. Dynamic Treatment Effects - Port Everglades, FL
206
REFERENCES CITED
Altay, N. and Ramirez, A. (2010). Impact of disasters on firms in different sectors:
implications for supply chains. Journal of Supply Chain Management, 46(4):59–80.
Anderson, J. E. (2011). The gravity model. Annual Review of Economics, 3(1):133–160.
Anderson, J. E. and van Wincoop, E. (2003). Gravity with gravitas: A solution to the
border puzzle. American Economic Review, 93(1):170–192.
Anderson, J. E. and van Wincoop, E. (2004). Trade costs. Journal of Economic
Literature, 42:691–751.
Anderton, C. H. and Carter, J. R. (2001). The impact of war on trade: An interrupted
times-series study. Journal of Peace Research, 38(4):445–457.
Ando, M. and Kimura, F. (2012). How did the japanese exports respond to two crises in
the international production networks? the global financial crisis and the great east
japan earthquake. Asian Economic Journal, 26(3):261–287.
Andrade da Silva, J. and Cernat, L. (2012). Coping with loss: the impact of natural
disasters on developing countries’ trade flows. Technical report, Directorate General
for Trade, European Commission.
Andriamananjara, S. (2004). Trade and international transport services: an analytical
framework. Journal of Economic Integration, pages 604–625.
Armington, P. S. (1969). A theory of demand for products distinguished by place of
production (une the´orie de la demande de produits diffe´rencie´s d’apre`s leur
origine)(una teor´ıa de la demanda de productos distinguie´ndolos segu´n el lugar de
produccio´n). Staff Papers-International Monetary Fund, pages 159–178.
Auffret, P. (2003). High consumption volatility: The impact of natural disasters? World
Bank Policy Research Working Paper, (2962).
Baier, S. L. and Bergstrand, J. H. (2001). The growth of world trade: tariffs, transport
costs, and income similarity. Journal of international Economics, 53(1):1–27.
Baldwin, R. and Taglioni, D. (2006). Gravity for dummies and dummies for gravity
equations. Technical report, National Bureau of Economic Research.
Basemann, R. and Daugherty, A. (1977). Complementarity in production and empty
backhauls in transportation. In Western Economics Association annual meeting.
207
Baumol, W. J. and Vinod, H. D. (1970). An inventory theoretic model of freight transport
demand. Management science, 16(7):413–421.
Behar, A. and Venables, A. J. (2011). 5 transport costs and international trade. A
Handbook of Transport Economics, page 97.
Behrens, K., Gaigne´, C., Ottaviano, G. I., and Thisse, J.-F. (2006). How density
economies in international transportation link the internal geography of trading
partners. Journal of Urban Economics, 60(2):248–263.
Behrens, K. and Picard, P. M. (2011). Transportation, freight rates, and economic
geography. Journal of International Economics, 85(2):280–291.
Belasen, A. R. and Polachek, S. W. (2008). How hurricanes affect wages and employment
in local labor markets. The American Economic Review, 98(2):49–53.
Bensassi, S., Martinez-Zarzoso, I., and Sua´rez, C. (2014). The effect of maritime transport
costs on the extensive and intensive margins: Evidence from the europe–asia trade.
Maritime Economics & Logistics, 16(3):276–297.
Bergstrand, J. H. (1985). The gravity equation in international trade: some
microeconomic foundations and empirical evidence. The review of economics and
statistics, pages 474–481.
Bergstrand, J. H. and Egger, P. (2011). Gravity equations and economic frictions in the
world economy. Palgrave Handbook of International Trade, pages 532–570.
Bernard, A. B., Eaton, J., Jensen, J. B., and Kortum, S. (2003). Plants and productivity
in international trade. The American Economic Review, 93(4):1268–1290.
Bernhofen, D. M., El-Sahli, Z., and Kneller, R. (2016). Estimating the effects of the
container revolution on world trade. Journal of International Economics, 98:36–50.
Bertrand, M., Duflo, E., Mullainathan, S., et al. (2004). How much should we trust
differences-in-differences estimates? The Quarterly Journal of Economics,
119(1):249–275.
Blonigen, B. A. and Wilson, W. W. (2008). Port efficiency and trade flows*. Review of
International Economics, 16(1):21–36.
Blonigen, B. A. and Wilson, W. W. (2013). The growth and patterns of international
trade. Maritime Policy & Management, 40(7):618–635.
Bosch, P., Coenen, P., Fridell, E., A˚stro¨m, S., Palmer, T., and Holland, M. (2009). Cost
benefit analysis to support the impact assessment accompanying the revision of
directive 1999/32/ec on the sulphur content of certain liquid fuels. AEA Report to
European Commission, Didcot.
208
Bougheas, S., Demetriades, P. O., and Morgenroth, E. L. (1999). Infrastructure, transport
costs and trade. Journal of International Economics, 47(1):169–189.
Brander, J. A. and Spencer, B. J. (1984). Trade warfare: tariffs and cartels. Journal of
International Economics, 16(3):227–242.
Burrus Jr, R. T., Dumas, C. F., Farrell, C. H., and Hall Jr, W. W. (2002). Impact of
low-intensity hurricanes on regional economic activity. Natural Hazards Review,
3(3):118–125.
Carrere, C. (2006). Revisiting the effects of regional trade agreements on trade flows with
proper specification of the gravity model. European Economic Review, 50(2):223–247.
Caruso, R. (2005). The impact of international economic sanctions on trade empirical
evidence over the period 1960-2000. Rivista Internazionale di Scienze Sociali, pages
41–66.
Cavallo, E. A. and Noy, I. (2009). The economics of natural disasters: a survey.
Chang, S. E. (2000). Disasters and transport systems: loss, recovery and competition at
the port of kobe after the 1995 earthquake. Journal of transport geography,
8(1):53–65.
Cheng, I.-H. and Wall, H. J. (2005). Controlling for heterogeneity in gravity models of
trade and integration. FEDERAL RESERVE BANK OF ST. LOUIS REVIEW,
87(1):49–63.
Choi, I. (2001). Unit root tests for panel data. Journal of international money and
Finance, 20(2):249–272.
Clark, X., Dollar, D., and Micco, A. (2004). Port efficiency, maritime transport costs, and
bilateral trade. Journal of development economics, 75(2):417–450.
Clausing, K. A. (2001). Trade creation and trade diversion in the canada–united states
free trade agreement. Canadian Journal of Economics/Revue canadienne
d’e´conomique, 34(3):677–696.
Cliff, A. D., O. K. (1969). The problem of spatial autocorrelation. A. J. Scott, ed., Studies
in Regional Science, pages 25–55.
Combes, P.-P. and Lafourcade, M. (2005). Transport costs: measures, determinants, and
regional policy implications for france. Journal of Economic Geography,
5(3):319–349.
CRED, T. C. f. R. o. E. o. D. (2015). The human cost of weather-related disasters
1995-2015.
209
Crespo Cuaresma, J., Hlouskova, J., and Obersteiner, M. (2008). Natural disasters as
creative destruction? evidence from developing countries. Economic Inquiry,
46(2):214–226.
Cullinane, K. and Bergqvist, R. (2014). Emission control areas and their impact on
maritime transport. Transportation Research Part D: Transport and Environment,
28:1–5.
Deardorff, A. (1998). Determinants of bilateral trade: Does gravity work in a neoclassical
world? In The Regionalization of the World Economy, pages 7–32. University of
Chicago Press.
Deardorff, A. V. (2014). Local comparative advantage: Trade costs and the pattern of
trade. International Journal of Economic Theory, 10(1):9–35.
Demirel, E., Van Ommeren, J., and Rietveld, P. (2010). A matching model for the
backhaul problem. Transportation Research Part B: Methodological, 44(4):549–561.
Dickey, D. A. and Fuller, W. A. (1979). Distribution of the estimators for autoregressive
time series with a unit root. Journal of the American statistical association,
74(366a):427–431.
Dickey, D. A. and Said, S. E. (1981). Testing arima (p, 1, q) against arma (p+ 1, q). In
Proceedings of the Business and Economic Statistics Section, American Statistical
Association, volume 28, pages 318–22.
Dolfman, M. L., Wasser, S. F., and Bergman, B. (2007). Effects of hurricane katrina on
the new orleans economy, the. Monthly Lab. Rev., 130:3.
Donald, S. G. and Lang, K. (2007). Inference with difference-in-differences and other panel
data. The review of Economics and Statistics, 89(2):221–233.
Dornbusch, R., Fischer, S., and Samuelson, P. A. (1977). Comparative advantage, trade,
and payments in a ricardian model with a continuum of goods. The American
Economic Review, pages 823–839.
Eaton, J. and Kortum, S. (2002). Technology, geography, and trade. Econometrica, pages
1741–1779.
Egger, P. (2000). A note on the proper econometric specification of the gravity equation.
Economics Letters, 66(1):25–31.
Egger, P. (2002). An econometric view on the estimation of gravity models and the
calculation of trade potentials. The World Economy, 25(2):297–312.
Egger, P. and Gassebner, M. (2015). International terrorism as a trade impediment?
Oxford Economic Papers, 67(1):42–62.
210
Egger, P. and Pfaffermayr, M. (2003). The proper panel econometric specification of the
gravity equation: A three-way model with bilateral interaction effects. Empirical
Economics, 28(3):571–580.
Egger, P. H. and Lassmann, A. (2012). The language effect in international trade: A
meta-analysis. Economics Letters, 116(2):221–224.
Elliott, J. R. and Pais, J. (2006). Race, class, and hurricane katrina: Social differences in
human responses to disaster. Social Science Research, 35(2):295–321.
Emanuel, K. (2005). Increasing destructiveness of tropical cyclones over the past 30 years.
Nature, 436(7051):686–688.
(EMSA), E. M. S. A. (2010). The 0.1% sulphur in fuel requirement as from 1 January
2015 in SECAs: An assessment of available impact studies and alternative means of
compliance.
Endresen, Ø., Bakke, J., Sørg˚ard, E., Berglen, T. F., and Holmvang, P. (2005). Improved
modelling of ship so 2 emissionsa fuel-based approach. Atmospheric Environment,
39(20):3621–3628.
European Commission (2009). Commission recommendation of 21 december 2009 on the
safe implementation of the use of low sulphur fuel by ships at berth in community
ports.
European Commission (2015). Emissions from maritime transport; the directive
1999/32/EC.
Fan, L., Luo, M., and Wilson, W. W. (2014). Pricing joint products in liner shipping.
International Journal of Shipping and Transport Logistics 5, 6(4):371–386.
Federal Maritime Commission et al. (2000). The ocean shipping reform act: An interim
status report. FMC, Washington, DC.
Feenstra, R. C. (2002). Border effects and the gravity equation: consistent methods for
estimation. Scottish Journal of Political Economy, 49:491–506.
Feenstra, R. C. and Romalis, J. (2014). International prices and endogenous quality*.
Quarterly Journal of Economics, 129(2).
Felton, J. R. (1981). The impact of rate regulation upon icc-regulated truck back hauls.
Journal of Transport Economics and Policy, pages 253–267.
Francois, J. F. and Wooton, I. (2001). Trade in international transport services: the role of
competition. Review of International Economics, 9(2):249–261.
Frankel, J. and Rose, A. (2002). An estimate of the effect of common currencies on trade
and income. Quarterly Journal of economics, pages 437–466.
211
Friedberg, L. (1998). Did unilateral divorce raise divorce rates? evidence from panel data.
The American Economic Review, 88(3):608–627.
Friedt, F. L. (2016). The varying effectiveness of trade policy in the presence of
endogenous transport costs. mimeo.
Friedt, F. L. and Wilson, W. W. (2015). Trade, transportation and trade imbalances: An
empirical examination of international markets and backhauls. Available at SSRN
2636983.
Fusillo, M. (2013). The stability of market shares in liner shipping. Review of Industrial
Organization, 42(1):85–106.
Gassebner, M., Keck, A., and Teh, R. (2010). Shaken, not stirred: the impact of disasters
on international trade. Review of International Economics, 18(2):351–368.
Getis, A. (2009). Spatial weights matrices. Geographical Analysis, 41(4):404–410.
Glick, R. and Taylor, A. M. (2010). Collateral damage: Trade disruption and the economic
impact of war. The Review of Economics and Statistics, 92(1):102–127.
Grenzeback, L. R., Lukman, A. T., and Systematics, C. (2008). Case Study of the
Transportation Sector’s Response to and Recovery from Hurricane’s Katrina and
Rita. Transportation Research Board.
Griffith, D. A. (1996). Some guidelines for specifying the geographic weights matrix
contained in spatial statistical models. Practical handbook of spatial statistics, pages
65–82.
Guha-Sapir, D., Hoyois, P., and Below, R. (2015). Annual disaster statistical review 2014:
The numbers and trends. Centre for Research on the Epidemiology of Disasters
(CRED), pages 1–33.
Hallegatte, S. (2008). An adaptive regional input-output model and its application to the
assessment of the economic cost of katrina. Risk analysis, 28(3):779–799.
Hallstrom, D. G. and Smith, V. K. (2005). Market responses to hurricanes. Journal of
Environmental Economics and Management, 50(3):541–561.
Hamilton, J. D. (1994). Time series analysis, volume 2. Princeton university press
Princeton.
Harrigan, J. (1996). Openness to trade in manufactures in the oecd. Journal of
international economics, 40(1):23–39.
Head, K. and Mayer, T. (2013). Gravity equations: Workhorse, toolkit, and cookbook.
Handbook of International Economics, 4.
212
Head, K., Mayer, T., and Ries, J. (2010). The erosion of colonial trade linkages after
independence. Journal of International Economics, 81(1):1–14.
Heaver, T. D. (1973). The structure of liner conference rates. The Journal of Industrial
Economics, pages 257–265.
Heckscher, E. F. and Ohlin, B. G. (1991). Heckscher-Ohlin trade theory. The MIT Press.
Holtz-Eakin, D. (2005). Macroeconomic and budgetary effects of Hurricanes Katrina and
Rita. HeinOnline.
Hornok, C., Koren, M., et al. (2011). Administrative barriers and the lumpiness of trade.
EFIGE WP, (36).
Hummels, D. (2001). Toward a geography of trade costs.
Hummels, D. (2007). Transportation costs and international trade in the second era of
globalization. The Journal of Economic Perspectives, pages 131–154.
Hummels, D., Lugovskyy, V., and Skiba, A. (2009). The trade reducing effects of market
power in international shipping. Journal of Development Economics, 89(1):84–97.
Hummels, D. and Skiba, A. (2004). Shipping the good apples out? an empirical
confirmation of the alchian-allen conjecture. Journal of Political Economy,
112(6):1384–1402.
Hummels, D. L. and Lugovskyy, V. (2006). Are matched partner trade statistics a usable
measure of transportation costs? Review of International Economics, 14(1):69–86.
Ishikawa, J. and Tarui, N. (2015). Backfiring with backhaul problems.
Ivanov, N. C. (2010). The EU directive on low sulphur fuel (2005/33/EC): A legal and
insurance industry perspective. Assuranceforeningen SKULD.
Jacks, D. S., Meissner, C. M., and Novy, D. (2008). Trade costs, 1870-2000. The American
Economic Review, 98(2):529–534.
Jonkeren, O., Demirel, E., van Ommeren, J., and Rietveld, P. (2011). Endogenous
transport prices and trade imbalances. Journal of Economic Geography,
11(3):509–527.
Jordan, W. C. (1987). Truck backhauling on networks with many terminals.
Transportation Research Part B: Methodological, 21(3):183–193.
Kahn, M. E. (2005). The death toll from natural disasters: the role of income, geography,
and institutions. Review of economics and statistics, 87(2):271–284.
213
Kao, C. and Chiang, M.-H. (1999). On the estimation and inference of a cointegrated
regression in panel data. Available at SSRN 1807931.
Kleinert, J. and Spies, J. (2011). Endogenous transport costs in international trade.
Technical report, IAW-Diskussionspapiere.
Korinek, J. (2011). Clarifying trade costs in maritime transport. Working Party of the
Trade Committee.
Krugman, P. (1980). Scale economies, product differentiation, and the pattern of trade.
The American Economic Review, 70(5):950–959.
Krugman, P. R. (1979). Increasing returns, monopolistic competition, and international
trade. Journal of international Economics, 9(4):469–479.
Kunkel, K. E., Pielke Jr, R. A., and Changnon, S. A. (1999). Temporal fluctuations in
weather and climate extremes that cause economic and human health impacts: A
review. Bulletin of the American Meteorological Society, 80(6):1077.
LeSage, J. P. and Pace, R. K. (2009). Introduction to Spatial Econometrics (Statistics,
textbooks and monographs). CRC Press.
Levin, A., Lin, C.-F., and Chu, C.-S. J. (2002). Unit root tests in panel data: asymptotic
and finite-sample properties. Journal of econometrics, 108(1):1–24.
Levinson, M. (2010). The box: how the shipping container made the world smaller and the
world economy bigger. Princeton University Press.
Li, Q. and Sacko, D. H. (2002). The (ir) relevance of militarized interstate disputes for
international trade. International Studies Quarterly, 46(1):11–43.
Limao, N. and Venables, A. J. (2001). Infrastructure, geographical disadvantage, transport
costs, and trade. The World Bank Economic Review, 15(3):451–479.
Lin, H., Kuo, Y., Shaw, D., Chang, M., and Kao, T. (2012). Regional economic impact
analysis of earthquakes in northern taiwan and its implications for disaster reduction
policies. Natural hazards, 61(2):603–620.
Lipsey, R. E. and Weiss, M. Y. (1974). The structure of ocean transport charges. In
Explorations in Economic Research, Volume 1, Number 1, pages 162–193. NBER.
Luo, Z. (2011). Transportation costs and trade imbalance: Theory and evidence.
Martincus, C. V. and Blyde, J. (2013). Shaky roads and trembling exports: Assessing the
trade effects of domestic infrastructure using a natural experiment. Journal of
International Economics, 90(1):148–161.
214
Mart´ınez-Zarzoso, I., Garc´ıa-Mene´ndez, L., and Sua´rez-Burguet, C. (2003). Impact of
transport costs on international trade: the case of spanish ceramic exports. Maritime
Economics & Logistics, 5(2):179–198.
Martinez-Zarzoso, I. and Nowak-Lehmann, F. D. (2007). Is distance a good proxy for
transport costs? the case of competing transport modes. The Journal of
International Trade & Economic Development, 16(3):411–434.
Mart´ınez-Zarzoso, I., Pe´rez-Garc´ıa, E. M., and Sua´rez-Burguet, C. (2008). Do transport
costs have a differential effect on trade at the sectoral level? Applied Economics,
40(24):3145–3157.
Mart´ınez-Zarzoso, I. and Sua´rez-Burguet, C. (2005). Transport costs and trade: empirical
evidence for latin american imports from the european union. Journal of
International Trade & Economic Development, 14(3):353–371.
Masozera, M., Bailey, M., and Kerchner, C. (2007). Distribution of impacts of natural
disasters across income groups: A case study of new orleans. Ecological Economics,
63(2):299–306.
McCallum, J. (1995). National borders matter: Canada-us regional trade patterns. The
American Economic Review, pages 615–623.
Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate
industry productivity. Econometrica, 71(6):1695–1725.
Meng, Y., Yang, S., Shi, P., and Jeager, C. (2015). The asymmetric impact of natural
disasters on china’s bilateral trade. Natural Hazards and Earth System Science,
15(10):2273–2281.
Mora, R. and Reggio, I. (2012). Treatment effect identification using alternative parallel
assumptions.
Nicholson, H. W. (1958). Motor carrier costs and minimum rate regulation. The Quarterly
Journal of Economics, pages 139–152.
Notteboom, T., Delhaye, E., and Vanherle, K. (2010). Analysis of the consequences of low
sulphur fuel requirements. ITMMA–Universiteit Antwerpen Transport&Mobility.
Noy, I. (2009). The macroeconomic consequences of disasters. Journal of Development
economics, 88(2):221–231.
Noy, I. and Vu, T. B. (2010). The economics of natural disasters in a developing country:
The case of vietnam. Journal of Asian Economics, 21(4):345–354.
Obstfeld, M. and Rogoff, K. (2001). The six major puzzles in international
macroeconomics: is there a common cause? In NBER Macroeconomics Annual 2000,
Volume 15, pages 339–412. MIT press.
215
Oh, C. H. and Reuveny, R. (2010). Climatic natural disasters, political risk, and
international trade. Global Environmental Change, 20(2):243–254.
Ord, K. (1975). Estimation methods for models of spatial interaction. Journal of the
American Statistical Association, 70(349):120–126.
Parsons, C. R. (2014). Disasters and trade: did hurricane katrina affect us imports?
International Economics and Economic Policy, pages 1–18.
Pedroni, P. (2000). Fully modified ols for heterogeneous cointegrated panels.
Pedroni, P. (2001). Purchasing power parity tests in cointegrated panels. Review of
Economics and Statistics, 83(4):727–731.
Pedroni, P. (2004). Panel cointegration: asymptotic and finite sample properties of pooled
time series tests with an application to the ppp hypothesis. Econometric theory,
20(03):597–625.
Pedroni, P. et al. (1999). Critical values for cointegration tests in heterogeneous panels
with multiple regressors. Oxford Bulletin of Economics and statistics, 61(s
1):653–670.
Pelli, M. and Tschopp, J. (2013). The creative destruction of hurricanes. Technical report,
Working Paper.
Pesaran, M. H., Shin, Y., and Smith, R. J. (2001). Bounds testing approaches to the
analysis of level relationships. Journal of applied econometrics, 16(3):289–326.
Phillips, P. C. and Perron, P. (1988). Testing for a unit root in time series regression.
Biometrika, 75(2):335–346.
Pielke Jr, R. A., Gratz, J., Landsea, C. W., Collins, D., Saunders, M. A., and Musulin, R.
(2008). Normalized hurricane damage in the united states: 1900–2005. Natural
Hazards Review, 9(1):29–42.
P&O Ferrymasters. New rules on maritime fuel: a short guide to the impact on you.
Port Finance International (2013). Worlds largest containership sets sail from busan. Last
accessed: 09/08/2015.
Raschky, P. A. (2008). Institutions and the losses from natural disasters. Natural Hazards
and Earth System Science, 8(4):627–634.
Rasmussen, M. T. N. (2004). Macroeconomic implications of natural disasters in the
Caribbean. Number 4-224. International Monetary Fund.
216
Rietveld, P. and Roson, R. (2002). Direction dependent prices in public transport: A good
idea? the back haul pricing problem for a monopolistic public transport firm.
Transportation, 29(4):397–417.
Rodrigue, J.-P., Comtois, C., and Slack, B. (2013). The geography of transport systems.
Routledge.
Rose, A. (2007). Economic resilience to natural and man-made disasters: Multidisciplinary
origins and contextual dimensions. Environmental Hazards, 7(4):383–398.
Rose, A., Benavides, J., Chang, S. E., Szczesniak, P., and Lim, D. (1997). The regional
economic impact of an earthquake: Direct and indirect effects of electricity lifeline
disruptions. Journal of Regional Science, 37(3):437–458.
Rose, A. and Wei, D. (2013). Estimating the economic consequences of a port shutdown:
the special role of resilience. Economic Systems Research, 25(2):212–232.
Rua, G. (2012). Fixed costs, network effects and the international diffusion of
containerization. Unpublished, October.
Samuelson, P. A. (1952). Spatial price equilibrium and linear programming. The American
economic review, pages 283–303.
Samuelson, P. A. (1954). The transfer problem and transport costs, ii: Analysis of effects
of trade impediments. The Economic Journal, pages 264–289.
Santos Silva, J. and Tenreyro, S. (2006). The log of gravity. The Review of Economics and
statistics, 88(4):641–658.
Schrooten, L., De Vlieger, I., Panis, L. I., Styns, K., and Torfs, R. (2008). Inventory and
forecasting of maritime emissions in the belgian sea territory, an activity-based
emission model. Atmospheric Environment, 42(4):667–676.
Sjostrom, W. (1992). Price discrimination by shipping conferences. Logistics and
Transportation Review, 28(2):207.
Skidmore, M. and Toya, H. (2002). Do natural disasters promote long-run growth?
Economic Inquiry, 40(4):664–687.
Smith, A. (1776). An inquiry into the nature and causes ofthe wealth ofnations. London:
George Routledge and Sons.
Strobl, E. (2011). The economic growth impact of hurricanes: evidence from us coastal
counties. Review of Economics and Statistics, 93(2):575–589.
Strobl, E. (2012). The economic growth impact of natural disasters in developing
countries: Evidence from hurricane strikes in the central american and caribbean
regions. Journal of Development economics, 97(1):130–141.
217
Stro¨mberg, D. (2007). Natural disasters, economic development, and humanitarian aid.
The Journal of Economic Perspectives, 21(3):199–222.
Takahashi, T. (2011). Directional imbalance in transport prices and economic geography.
Journal of urban Economics, 69(1):92–102.
Talley, W. K. and Pope, J. A. (1985). Determinants of liner conference rates under
containerization. International Journal of Transport Economics/Rivista
internazionale di economia dei trasporti, pages 145–155.
Tanaka, K. and Tsubota, K. (2014). Directional imbalance of freight rates: evidence from
japanese inter-prefectural data.
Thursby, J. G. and Thursby, M. C. (1987). Bilateral trade flows, the linder hypothesis,
and exchange risk. the Review of Economics and Statistics, pages 488–495.
Tinbergen, J. (1962). Shaping the world economy; suggestions for an international
economic policy. Books (Jan Tinbergen).
Toya, H. and Skidmore, M. (2007). Economic development and the impacts of natural
disasters. Economics Letters, 94(1):20–25.
Transportation Research Board (2006). Tr news september-october 2006: The intermodal
container era: History, security, and trends. page 246.
UNCTAD, Secretariat (1979-2014). Review of Maritime Transport 2014. United Nations
Publications.
Vigdor, J. (2008). The economic aftermath of hurricane katrina. The Journal of Economic
Perspectives, 22(4):135–154.
Wallace, R. (n.d.). Definitions of eia distillate categories and fuels contained in the
distillate grouping. Department of Energy.
Wilmsmeier, G., Hoffmann, J., and Sanchez, R. J. (2006). The impact of port
characteristics on international maritime transport costs. Research in Transportation
Economics, 16:117–140.
Wilson, W. W. (1987). Transport markets and firm behavior: the backhaul problem. In
Journal of the Transportation Research Forum, volume 28.
Wilson, W. W. (1994). Market choice, entry regulation, and joint production: Market
access and market service in motor carrier markets. Review of Industrial
Organization, 9(6):791–812.
Wilson, W. W. and Beilock, R. (1994). Market access in regulated and unregulated
markets: The continuing cost of interstate motor carrier regulation. Journal of
Regulatory Economics, 6(4):363–379.
218
Wolfers, J. (2006). Did unilateral divorce laws raise divorce rates? a reconciliation and
new results. American Economic Review, 96(5):1802–1820.
Xiao, Y. and Nilawar, U. (2013). Winners and losers: analysing post-disaster spatial
economic demand shift. Disasters, 37(4):646–668.
Yang, D. (2008). Coping with disaster: The impact of hurricanes on international financial
flows, 1970-2002. The BE Journal of Economic Analysis & Policy, 8(1).
219