Essays on Maritime Transport and International Trade
by
Philip Economides
A dissertation accepted and approved in partial
fulfillment of the requirements for the degree of Doctor of
Philosophy in Economics
Dissertation Committee:
Woan Foong Wong Co–Chair
Bruce Blonigen Co–Chair
Anca Cristea Core Member
Nicole Ngo Institutional Representative
University of Oregon
Spring 2024
© Copyright by Philip Economides 2024.
This work is licensed under Attribution-NonCommercial 4.0 International.
All rights reserved.
ii
DISSERTATION ABSTRACT
Philip Economides
Doctor of Philosophy in Economics
Department of Economics
Title: Essays on Maritime Transportation and International Trade
This dissertation considers topics which dovetail studies of maritime trade and trans-
portation. Using theoretical models, empirical identification and structural analysis, I
provide novel evidence on three key facts; (i) the repositioning of empty container units
is a key logistical practice in maritime shipping that enables the sustained service of global
trade imbalances, (ii) advancements in container shipping technology through increased
vessel capacity between 1977-2023 have introduced negative spillovers on cargo handling
times at port, and (iii) the newly introduced estimated time of arrival (ETA) based port
queuing system has contributed to decarbonization in the maritime shipping sector.
In the first substantive chapter, I develop a model of containerized trade and trans-
portation which embeds the logistical practice of container repositioning by transport
operators. This involves bringing equipment to where it is most needed for further trans-
port service, and may necessitate the transportation of empty containers when servicing
commerce between countries with particularly large trade imbalances. I contrast the com-
parative statics of this model with novel container traffic data, collected individually from
the key US ports. These reduced-form analyses demonstrate that the balanced exchange
of container units can only be revealed upon accounting for empty units. Motivated by
the recent passing of the Ocean Shipping Reform Act of 2022, I use a structural approach
to examine the implications of restricting empty container outflows from the US in order
to stimulate US exports. The results of this exercise suggest the policy backfires for
the broader public. Although exports are stimulated by policymaker action, transport
operators respond to this form of unconventional policy intervention by adjusting freight
rates bilaterally. The resulting increase in freight rates for shipping routes destined for
the US contributes to an overall reduction in trade activity and a pronounced decline in
vessel capacity allocated towards the US containerized shipping market.
iii
In the second substantive chapter, Woan Foong Wong, Simon Fuchs and I explore
how technological innovation and port conditions contribute to variation in individual
containership dwell time events across the US. Our data documents vessel size, container
capacity, and port concentration from January 1977 to December 2023. We observe
a four-fold increase in US port visits, peaking in 2010, followed by a downward trend
until 2023. This pattern coincides with an accelerated rate of entry among the largest
categories of containership classes. We suggest that transport operators are increasingly
relying on improved vessel technology to meet growing demand for trade, rather than by
supplying more vessels. Despite volume growing over time, average dwell times across US
ports have remained centered around 2.4 days. Our empirical results suggest that this
status quo is maintained by offsetting mechanisms; (i) larger vessels representing greater
unloading efficiencies, and (ii) increased port traffic volumes introduce stronger negative
spillover effects on visiting vessels.
In the third substantive chapter, I examine how logistical practices by port authorities
can influence vessel emissions. I use the case study of San Pedro Bay, California, which
introduced a new vessel queuing system. Under the former system, vessels would be
required to enter within 25 nautical miles of the ports of Los Angeles and Long Beach
before being eligible to be admitted to the vessel queue. Additionally, those awaiting
service could anchor near the port area or drift nearby. After observing a swelling of
anchorage zone and drift areas use, authorities introduced a queuing system in which
each vessel’s calculated time of arrival determined their queue position and mandated no
idling within a 150 nautical mile area of the ports. I find evidence which suggests that
the policy slowed down inbound vessels, reduced idling time prior to port admittance,
but increased the extent to which vessels would reposition while waiting. Accounting for
all three factors, I find that the policy contributed to a 30.2% decline in containership
emissions relative to control ports along the US West Coast.
The dissertation includes previously unpublished co-authored research.
iv
CURRICULUM VITAE
NAME OF AUTHOR: Philip Economides
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR
University College Dublin, Dublin, Ireland
Trinity College Dublin, Dublin Ireland
DEGREES AWARDED:
Doctor of Philosophy, Economics, 2024, University of Oregon
Masters of Science, Economics, 2020, University of Oregon
Masters of Science, Applied Economics, 2017, University College Dublin
Bachelors of Arts, Economics, 2014, Trinity College Dublin
AREAS OF SPECIAL INTEREST:
International Economics, Transport Economics, Applied Microeconomics
PUBLICATIONS:
Economides, P., and G. Nikolaishvili (2023) “Measuring Economic Activity in the Pres-
ence of Superstar MNEs”, Economics Letters, Vol. 226, 111077
Economides, P., K. McQuinn, and C. O’Toole (2021) “Household Savings Constraints,
Uncertainty, and Macroprudential Policy”, Scottish Journal of Political Economy, Vol.
68(2), pp. 238-260
GRANTS, AWARDS, AND HONORS:
PhD Scholars Fund, RES 2024 Conference (Queen’s University Belfast), 2024
Best Graduate Student Paper Award, West Coast Trade Workshop (UC Berkeley), 2024
Gerlof Homan Scholarship in International Economics, University of Oregon, 2023
Dissertation & Thesis Award, Division of Graduate Studies, University of Oregon, 2023
WEAI Graduate Workshop 2023, University of Oregon Nominee, 2022
Kleinsorge Summer Fellowship, University of Oregon, 2021
Edward G. Daniel Scholarship, University of Oregon, 2020
Graduate Teaching Fellowship, University of Oregon, 2019 – 2024
v
ACKNOWLEDGEMENTS
I wish to express my gratitude for the support, guidance and patience I have received
from Woan Foong Wong, Bruce Blonigen, Anca Cristea and Nicole Ngo. I am grateful
for the generous times and many fruitful discussions my peers have afforded me too.
A special thanks to Mark Colas, Gio Nikolaishvili, Tami Ren, Emmett Saulnier, and
Brock Wilson for their dedicated input. I am very thankful to the participants at the
University of Oregon International Trade group. Furthermore, versions of Chapter II
were presented at the 2023 WEAI Graduate Workshop, the 2023 Midwest International
Trade Conference, the 2024 West Coast Trade Workshop, the 2024 RES Conference and
the Atlanta FED. I thank the audience members and, in particular, Jefferey Bergstrand,
Carsten Eckel, Benjamin Faber, Simon Fuchs, David Hummels, Nayul Kim, Pearl Li,
James Markusen, Adam Pfander, Andrés Rodŕıguez-Clare, Thomas Sampson, and Larry
White for their many helpful comments. Any remaining errors are my own.
The financial support I received through the Gerlof Homan Graduate Scholarship in
International Economics and Dissertation & Thesis Award, which funded data require-
ments for Chapters II and IV, respectively, are also gratefully acknowledged.
vi
To my wife, Alina, and our parents, Carol, Irina, and Vitaly, for their unwavering
support.
vii
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Unconventional Protectionism in Containerized Shipping . . . . . . . 4
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.2 Weakly Imbalanced Trade . . . . . . . . . . . . . . . . . . . . . . 15
2.3.3 Comparative Statics . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Containerized Goods . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.2 Container Traffic . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.3 Auxiliary Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Stylized Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5.1 Empty Repositioning & Trade Balance Asymmetry . . . . . . . . 22
2.5.2 Balanced Container Flows . . . . . . . . . . . . . . . . . . . . . . 24
2.5.3 Port Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 Counterfactual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6.1 US-RoW Baseline . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6.2 Multi-Country Container Flows . . . . . . . . . . . . . . . . . . . 33
2.6.3 Solution Method and Model Calibration . . . . . . . . . . . . . . 40
2.6.4 Counterfactual Policy Background . . . . . . . . . . . . . . . . . . 41
2.6.5 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3 Container Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2.1 Ships and Port Data . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2.2 Port Composition . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.3 Stylized Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.4.1 Dwell Time on Vessel Size . . . . . . . . . . . . . . . . . . . . . . 71
3.4.2 Instrumental Variable Approach . . . . . . . . . . . . . . . . . . . 73
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4 Cargo Ships & Coastal Smog: A Case Study of San Pedro Bay . . . . 78
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.2 Data & Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
viii
4.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.2.2 Fuel Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.2.3 Vessel Emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3 Empirical Strategy & Results . . . . . . . . . . . . . . . . . . . . . . . . 92
4.3.1 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.3.2 Local Emission Effects . . . . . . . . . . . . . . . . . . . . . . . . 95
4.3.3 Voyage Emission Effects . . . . . . . . . . . . . . . . . . . . . . . 99
4.3.4 Queuing Emission Effects . . . . . . . . . . . . . . . . . . . . . . 102
4.3.5 Global Emission Effects . . . . . . . . . . . . . . . . . . . . . . . 105
4.4 Discussion & Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
List of Figures
2.1 Balanced National Container Flows by Time Window . . . . . . . . . . . 26
2.2 Imbalanced National Loaded Container Flows by Time Window . . . . . 27
2.3 Port Specialization by Net Inflow Status (2017) . . . . . . . . . . . . . . 28
2.4 Port Specialization by Total Container Thruflow (2012-2021) . . . . . . . 30
2.5 Empty Share of Container Movement by Year-Month . . . . . . . . . . . 32
2.6 Loading Factor Estimates by Commodity . . . . . . . . . . . . . . . . . . 36
2.7 Model Fit – Loaded Container Ratios by Region (2012-2021) . . . . . . . 38
2.8 Estimated Container Flows by Country and Direction . . . . . . . . . . . 39
2.9 US Import Value by Net Exporter (2017) . . . . . . . . . . . . . . . . . . 46
2.10 US Export Value by Net Exporter (2017) . . . . . . . . . . . . . . . . . . 47
2.11 US Import Price Inflation by Net Exporter (2017) . . . . . . . . . . . . . 48
2.12 Change in Trade Partner Shares of US imports (2017) . . . . . . . . . . . 48
3.1 Coastal Shares of Containerized Trade Value (2003-2023) . . . . . . . . . 64
3.2 Total Number of Port Visits by Containership . . . . . . . . . . . . . . . 65
3.3 Total Number of Unique Port Visits by Containers . . . . . . . . . . . . 65
3.4 Variation in Containership Capacity . . . . . . . . . . . . . . . . . . . . . 67
3.5 Changes in Container Capacity Shares . . . . . . . . . . . . . . . . . . . 68
3.6 Port Volumes of Vessel TEU Capacity by Year-Month . . . . . . . . . . . 68
3.7 Average Visits per Containership by Year . . . . . . . . . . . . . . . . . . 69
3.8 Average Dwell Times of Containership Visits . . . . . . . . . . . . . . . . 70
3.9 Average Dwell Times of Containership Visits by Port . . . . . . . . . . . 70
4.1 2005-2021 POLA OGV NOx Emissions (tons), by mode . . . . . . . . . . 82
4.2 2020 & 2021 POLA OGV NOx Emissions (tons), by mode . . . . . . . . 82
4.3 POLA Number of Container Ships at Anchorage, Daily . . . . . . . . . . 84
4.4 Match Status by Origin and Group . . . . . . . . . . . . . . . . . . . . . 88
4.5 Fuel Consumption by Speed and TEU Capacity . . . . . . . . . . . . . . 90
4.6 Parallel Trends in Average Voyage Speeds . . . . . . . . . . . . . . . . . 94
ix
4.7 AQI Monitor Local Sites . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.8 Event Study (TWFE) - Voyage Speed . . . . . . . . . . . . . . . . . . . . 101
4.9 Event Study (TWFE) - Voyage Emissions . . . . . . . . . . . . . . . . . 102
4.10 CO2 emissions by hour across queued vessels . . . . . . . . . . . . . . . . 104
4.11 Queuing Share of Global Containership Transit for US West Coast . . . . 105
5.1 Balanced Port Container Flows . . . . . . . . . . . . . . . . . . . . . . . 122
5.2 Counterfactual Outcomes by Empty Outflow Tax, 2017 . . . . . . . . . . 127
5.3 European Specialization by Net Flow Status (2017) . . . . . . . . . . . . 136
5.4 Input Price by Loaded Container . . . . . . . . . . . . . . . . . . . . . . 139
5.5 Vessel Category Shares of Coastal Containership Activity . . . . . . . . . 152
5.6 Container Capacity of Vessel Visits by Region . . . . . . . . . . . . . . . 152
5.7 Match Rate, Panjiva Vessel Offloading to Port Call Records . . . . . . . 155
5.8 Matching AIS positions to Port Call Records . . . . . . . . . . . . . . . . 158
5.9 Fuel Consumption Function . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.10 Speed by Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.11 Emissions by Speed-Stage . . . . . . . . . . . . . . . . . . . . . . . . . . 162
List of Tables
2.1 Trade Flow Ratio & Empty Shares . . . . . . . . . . . . . . . . . . . . . 23
2.2 Net-Gross Ratio & Empty Shares . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Empty Container Elasticity with Respect to Trade Flows . . . . . . . . . 24
2.4 Balanced National Container Flows . . . . . . . . . . . . . . . . . . . . . 25
2.5 Average Containership Gross Tonnage by Port Size . . . . . . . . . . . . 29
2.6 Average Empty Share of Container Outflows by Port-Year (%) . . . . . . 30
2.7 Disaggregated Counterfactual Outcomes . . . . . . . . . . . . . . . . . . 45
3.1 Port Rankings by Containerized Trade Value (2003-2023) . . . . . . . . . 63
3.2 OLS Vessel Dwell Time by Ship Category . . . . . . . . . . . . . . . . . 72
3.3 OLS Vessel Efficiency by Ship Category . . . . . . . . . . . . . . . . . . . 73
3.4 OLS Elasticity of Vessel Efficiency with respect to Port Traffic . . . . . . 74
4.1 Summary Statistics of Vessel Voyages . . . . . . . . . . . . . . . . . . . . 88
4.2 Estimate of Fuel Consumption by GT-Dwell Hours . . . . . . . . . . . . 91
4.3 Emissivity Indices for Selected Marine Fuels . . . . . . . . . . . . . . . . 91
4.4 Difference-in-Difference, Control: San Pedro Bay, Zone-II . . . . . . . . . 97
4.5 Difference-in-Difference, Control: San Pedro Bay, Zone-II . . . . . . . . . 97
4.6 Difference-in-Difference, Control: Seattle/Oakland, Zone-I . . . . . . . . 98
4.7 Difference-in-Difference, Control: Seattle/Oakland, Zone-I . . . . . . . . 98
4.8 Difference-in-Difference Estimates – Voyage Speed . . . . . . . . . . . . . 100
4.9 Difference-in-Difference Estimates – Voyage Emissions . . . . . . . . . . . 101
4.10 Difference-in-Difference Estimates – Queuing Emissions . . . . . . . . . . 105
4.11 Difference-in-Difference Estimates – Global Emissions . . . . . . . . . . . 106
x
5.1 Sample Representation - US Total Container Throughput . . . . . . . . . 120
5.2 Empty Container Elasticity with Respect to Trade Flows (kg) . . . . . . 121
5.3 (Ports) Trade Flow Ratio & Empty Shares . . . . . . . . . . . . . . . . . 121
5.4 (Ports) Empty Container Elasticity w.r.t. Opposite-Direction Trade Flows 122
5.5 Key Parameters, 2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.6 National Counterfactual Outcomes . . . . . . . . . . . . . . . . . . . . . 127
5.7 Jointly Estimated Loading Factors . . . . . . . . . . . . . . . . . . . . . 129
5.8 Import-Specific Loading Factors . . . . . . . . . . . . . . . . . . . . . . . 130
5.9 Export-Specific Loading Factors . . . . . . . . . . . . . . . . . . . . . . . 130
5.10 Performance Diagnostics by Methodology . . . . . . . . . . . . . . . . . . 131
5.11 Jointly Estimated Geographic Loading Factors . . . . . . . . . . . . . . . 131
5.12 Import-Specific Geographic Loading Factors . . . . . . . . . . . . . . . . 132
5.13 Export-Specific Geographic Loading Factors . . . . . . . . . . . . . . . . 132
5.14 Jointly Estimated Income-based Loading Factors . . . . . . . . . . . . . 133
5.15 Import-Specific Income-based Loading Factors . . . . . . . . . . . . . . . 133
5.16 Export-Specific Income-based Loading Factors . . . . . . . . . . . . . . . 134
5.17 RMSE of US - E. Asian Container Flow Ratios . . . . . . . . . . . . . . 134
5.18 RMSE of US-European Container Flow Ratios . . . . . . . . . . . . . . . 135
5.19 Manually Entered IMO–TEU Capacity Pairs . . . . . . . . . . . . . . . . 151
5.20 OLS Vessel Efficiency by Ship Category – Panjiva . . . . . . . . . . . . . 156
5.21 IV Elasticity of Vessel Efficiency with respect to Port Traffic – Panjiva . 156
5.22 Difference-in-Difference Estimates – Queuing Emissions . . . . . . . . . . 163
xi
Chapter 1
Introduction
1
In the past four years, transport services around the world have been tested to their
limits. The increased volatility in the demand for final goods imports, partly attributed
to bouts of precautionary accumulations of inventory and repeated cases of COVID-19
shutdowns, has led to the emergence of widespread log-jams across international trans-
port networks. The resulting disruptions in these supply chains have exposed a number
of key vulnerabilities across transport networks, particularly with respect to maritime
transport, which services over 80% of traded goods volumes.1 My dissertation seeks to
provide insight into critical policy questions related to a reliance on available transport
equipment, potential limitations in port infrastructure and concerns surrounding the re-
sulting pollution of vessel congestion. This dissertation contains previously published and
unpublished co-authored material. Chapter 3 is joint work with Woan Foong Wong and
Simon Fuchs.
Chapter 2 considers a setting in which roundtrip transport services, operated by
containerships, routinely travel back and forth between origin–destination pairs. Due to
potential imbalances in shipping volumes and a persistent need for available transport
equipment on the larger-volume leg of a given round trip, vessel operators must include
empty containers on the low-volume leg. Using a novel data set of bilateral container
flows, which separately reports empty and loaded container traffic across key US ports,
I document a system of balanced container flows between the US its trade partners. I
develop a partial equilibrium model of round trip trade & container redistribution to
assess a counterfactual of restrictions to empty container outflows, in favor of making
more equipment available for US exporters.
I find this unconventional form of protectionism lowers vessel capacity on a given
roundtrip by almost 20%, which contributes to higher import prices (+1.75%) and a
reduction in the overall value and volume of trade between the US and the rest of the
world (-8.5%). My findings suggest that interference with the manner in which container
equipment is freely redistributed may backfire, depending on which party of interest is
focused upon. From the policymaker perspective, exports have been successfully grown.
However, I show evidence that policymaker concerns may need to be extended to effects
on the opposite return-leg of trade, which reveals an overall decline in US access to
containerized goods, lower available transport capacity and associated inflation.
Chapter 3 shows that vessels of relatively larger size have developed economies of
scale in container handling at the port-level, although negative externalties of increased
vessel mass have contributed to negative spillovers for the existing fleet. Despite contain-
ership mass growing 5-fold between January 1977 and December 2023, port visit times
have remained largely stable at 2.4 days in duration due to these offsetting factors. We
1UNCTAD (2021) Review of Maritime Transport 2021, Challenges faced by seafarers in view of the
COVID-19 crisis. https://unctad.org/publication/review-maritime-transport-2021
2
document the evolution of containership technology and its adoption among the major
ports of the US.
We find monotonic efficiency gains across containership categories, with some of the
largest vessels servicing the merchant fleet in the US offering a 47.8–55.7% reduction
in dwell times per container. Given the recent trends in weaker market competition in
containerized shipping and skewedness in the extent to which large operators can build
new ships relative to smaller operators, this mechanism may be propagating a faster
rate of market concentration. We conclude with remarks on how to approach the next
stage of this paper, in which we develop a micro-model of containership queuing theory
and explore means by which to estimate exogenous model primitives for the purpose of
evaluating counterfactuals related to port infrastructure and vessel technology.
In Chapter 4, I use the case study of the new containership queuing system in San
Pedro Bay, California, to assess how an estimated time of arrival based queue position
influences vessel emissions both locally and globally. To recover fuel consumption and
emissions of individual vessels, I first combine the use of MarineTraffic records port of
origins and departure timestamps of vessels inbound for the US West Coast with highly
granular vessel position data for US waters. I then use prevailing approaches in the
maritime engineering and transportation strands of literature to impute fuel and emission
levels. This allows me to identify individual vessels repeatedly over time and control for
strategic responses to the policy, such as lane switching, market exits and market entry.
I find that the policy featured offsetting effects on the voyage and queuing stages of
transit for individual vessels. Using a difference-in-difference approach for this quasi-
experimental setting, I find that the policy reduced associated vessel CO2 levels by 30.2%
relative to control ports along the US West Coast. While the policy did somewhat
slowdown vessel travel speeds, the biggest boon in emission reductions was achieved
through reduced idling in the port vicinity. Now that containership operators had certain
queue positions upon departure from a prior port, rather than receiving that information
within 20 nautical miles of arrival, they could minimize their associated idle time and
better purpose the use of their transport equipment. Use of this logistical practice across
ports may be of interest to policymakers and port authorities seeking to decarbonize
ports, particularly for facilities subject to extensive queuing experiences.
3
Chapter 2
Unconventional Protectionism in
Containerized Shipping
4
Unconventional Protectionism in Containerized
Shipping
Philip Economides∗
April 15th, 2023
Abstract
The containerized shipping market operates similarly to a bus system, where
vessels maintain round trip transport services between origin-destination
pairs. Intermediary transport operators must commit to sufficient shipping
capacity, while accounting for bilateral shipping imbalances. To ensure nec-
essary transport equipment availability, vessel owners reposition empty con-
tainer units on the low-volume leg of a round trip, from net importer origins
to net exporter destinations. I provide evidence of bilateral US container
traffic being consistently balanced – only when accounting for empty con-
tainer repositioning. Motivated by the Ocean Shipping Reform Act of 2022,
I explore the effects of a US restriction to empty container outflows in fa-
vor of stimulating US exports. This form of policy intervention – although
helpful in meeting policymaker goals of stimulating exports – backfires for
the broader public through elevated import prices, lower transport capacity,
and reduced trade activity.
JEL classification: F14, F18, F64, Q52, R49
Keywords: trade costs, transport costs, transportation, trade policy, container shipping
∗Department of Economics, University of Oregon.
I am indebted to Woan Foong Wong, Bruce Blonigen, Anca Cristea, and Nicole Ngo for their advice and
guidance. I would like to thank Mark Colas, Mike Kuhn, Benjamin Hansen, Kathleen Mullen, David
Evans, and seminar participants at the University of Oregon for suggestions and comments. I am also
thankful to the conference organizers at the WEAI and my cohort advisor, Prof. Lawrence J. White, for
the opportunity to present this work at the 2023 WEAI graduate workshop. All remaining errors of this
draft are my own.
Unconventional Protectionism
2.1 Introduction
Approximately 70% of international trade values travels via maritime transport,
two-thirds of which is attributed to containerized shipping (Notteboom et al.,
2022). These services specialize in providing round trip transport, where ports are
routinely visited back-and-forth between specific origin-destination combinations.
Containers are repositioned within these continuous loops of transport services,
creating a persistent circulation of transport equipment. In cases of imbalanced
demand and asymmetric shipping volumes, repositioning includes empty contain-
ers. This phenomenon introduces the empty container repositioning problem for
transport operators – a need to relocate empty containers on the low-volume leg
of a given round trip, from net importer countries back to net exporter coun-
tries (Song, 2021). The repositioning of empty containers is estimated to repre-
sent 20% of total ocean container movements and 15% of fleet management costs
(Drewry, 2006; Rodrigue, 2020). This implies that variation in repositioning in-
fluences vessel-owning intermediaries’ costs, which leads to changes in allocated
vessel capacity, freight rate pricing and trade outcomes on round trip routes. Al-
though container repositioning has been well-documented in the maritime logistics
literature (Crainic et al., 1993; Lee and Song, 2017; Song, 2007), little is known of
how frictions in container availability affect trade outcomes. The recent passing
of the Ocean Shipping Reform Act, henceforth OSRA22, embodies an example
of a restriction to container repositioning. Under this bill, the Federal Maritime
Commission (FMC) has been tasked with limiting the extent to which transport
operators can refuse allocating portions of vessel capacity to US containerized
exports in favor of transporting additional empty container units.
In this paper, I examine container repositioning under round trip trade and
quantitatively evaluate how policy restrictions to empty container outflows, such
as OSRA22, may influence US trade outcomes. My main findings suggest that
empty container repositioning is key in sustaining prevailing trade imbalances and
existing transport capacity levels. When empty repositioning is restricted in favor
of stimulating domestic exports, shipping supply declines, which in turn leads to
added inflationary pressure and an overall reduction in bilateral trade activity.
I first build a quantitative model of round trip trade based on Armington
(1969), which is capable of featuring both balanced and imbalanced exchanges of
goods, and includes a richer specification of endogenous trade costs. A representa-
tive exporter faces both the domestic cost of producing a good and the freight rate
6
Unconventional Protectionism
issued by a transport operator. The transport operator maintains bilateral round
trip services between two countries. Price setting for these services accounts for
differences in demand between regions and partly reflects the cost of repositioning
empty containers on the low-volume leg of a given round trip. Should the cost of
handling empty container units rise, a transport operator lowers their exposure to
trade volume asymmetries through bilateral freight rate adjustments and reduced
shipping capacity. From the perspective of a net importer country, such as the
US, the model predicts that when the import-export ratio rises, resulting empty
container traffic as a proportion of total outbound container units must rise too.
Using novel port-level loaded & empty container traffic data1, I empirically
examine the validity of these comparative statics and establish three key facts; (i)
the scale of the empty container repositioning problem grows as asymmetries in
shipping volumes intensify, (ii) balanced exchanges of national bilateral flows of
total container flows are evident only when accounting for empty container repo-
sitioning across these US ports, and (iii) the relative size of a port is predictive of
the role each location plays – large ports such as Los Angeles & New York gener-
ate persistent net inflows of containers while mid-tier US ports are sources of net
outflows. Findings (ii) and (iii) suggest that the US maintains an interdependent
container repositioning system between US ports and the hinterland, indicating a
reliance on the accessibility of inter-modal transport. Only upon a national aggre-
gation across US ports does the model’s constraint of a balanced container flow
network appear evident.
In preparing a quantitative analysis of OSRA22, I combine my measures of
container traffic with US census data on monthly port-level bilateral containerized
trade flows (by product type, value, and weight) and auxiliary country-level data.
This allows me to calibrate and estimate model primitives of the baseline scenario
of my model through a two-stage estimation strategy.
The first stage estimates bilateral loaded container flows between US ports
and the main trading partners of the US. This is achieved by exploiting variation
in metric tonne weights of 2-digit Harmonized System (HS2) goods shipped on
these same trade routes across each year-month of the sample. Suppose that for
a given shipping lane, there is a marginal increase in the metric tonnes of a prod-
uct’s weight. Given that each container maintains a weight capacity, a greater
amount of a given good suggests an increased number of containers allocated for
transport. Furthermore, the rate at which each product’s weight increases total
1This balanced panel represents over 80% of US container throughput for 2012–2021.
7
Unconventional Protectionism
container count usage varies due to the volume constraint each container repre-
sents. For example, a metric ton of sheet metal likely takes up far less volume in a
container unit compared to a metric ton of furniture. By estimating each product’s
“loading factor” – the rate at which weight contributes to loaded container flows
– I recover origin-destination loaded container flows between US ports and key US
trade partners. I provide evidence of a striking fit between country-specific esti-
mated loaded container flows and UNCTAD data of East Asian–North American
and European–North American bilateral loaded container traffic.
The second stage uses a Generalized Method of Moments (GMM) approach
to recover four model primitives for each shipping route – the underlying pair
of preference parameters each country’s consumer base maintains for their trade
partner’s manufactured goods, as well as per-unit costs of handling empty and
loaded container units. The remaining primitives are calibrated using a combi-
nation of public data sourced from the International Labor Organization, OECD,
and World Bank. Estimated primitives align well with what is known of shipping.
For example, depending on the lane, my estimate of empty container handling
costs varies between 14.9% and 21.3% of total fleet management costs, which is
rather close to the 15% share reported in Rodrigue (2020). Furthermore, implied
freight rates are consistently higher on the higher-volume lanes of a given round
trip, as established in Hummels et al. (2009).
To capture the intent of OSRA22’s unconventional trade policy, I consider the
effects of an empty container outflow (ECO) quota, which effectively reallocates
vessel space towards US exporters. I specifically consider a moderate regime, where
the policymaker seeks to return to a status-quo represented by the 40% long-run
average of empty container outflows as a percentage of total container outflows
originating from the US. I find that restricting the return of empty transport
equipment meets the sole objective for higher exports for the US policymaker,
but conflicts with the broader interests of the public once accounting for the full
roundtrip effect. Constraining repositioning contributes to an 18.6% decline in
round trip shipping capacity, a 17.7% decline in US containerized imports and an
8.5% reduction in the total value of US containerized trade. Additionally, imported
inflation grows by just under 2 percentage points.
To the best of my knowledge, this paper is the first to provide empirical evidence
of the affect of empty container repositioning in round trip transport services on
trade outcomes. Additionally, the micro-founded model of this paper enables the
assessment of a relatively modern and unique trade policy concern, represented by
8
Unconventional Protectionism
OSRA22. The results of this paper contribute to several strands of the literature.
First, this paper adds to the international trade literature on endogenous trade
costs. Transport costs represent an increasingly prominent factor in determining
overall trade costs. For example, Hummels (2007) finds that for every $1 ex-
porters paid in tariff duties to send goods to the US, $9 was paid in transportation
costs. Although earlier studies used ad-hoc transport costs,2 more recent theoret-
ical frameworks use a variety of endogenous approaches (Irarrazabal et al., 2015;
Hayakawa et al., 2020; Bonadio, 2022). Atkin and Donaldson (2015), Brancaccio
et al. (2020) and Ignatenko (2023) use differences in market power across inter-
mediary transport service operators for variation in transport costs. Allen and
Arkolakis (2022) and Wong and Fuchs (2022) highlight how the quality of infras-
tructure and traffic congestion across regions can also explain variation in transport
costs. Using bilateral container traffic data at the port level, I document how the
cost of servicing imbalanced trade routes through empty container repositioning
affects round trip trade flows.
Secondly, this paper is closely related to studies focused on particular facets
of maritime transport. These technological and logistical innovations play impor-
tant roles in influencing key economics outcomes. Bernhofen et al. (2016) suggests
container technology introductions between 1962-1990, on average, contributed to
a 85% higher trade ten years later. Brooks et al. (2021) highlights how container
technology led to substantial population and employment growth in US counties
near containerized ports. Following the 2016 Panama Canal expansion, Heiland
et al. (2022) estimates an average increase in trade of 9-10% across affected ship-
ping lanes. Ganapati et al. (2021) provides evidence of logistical hubs known
as entrepôts fostering advancements in vessel technology and size, which lowered
transport costs. Carreras-Valle (2022) shows that technological innovations re-
duced internationally-sourced input costs.3 Through the novel container traffic
data available to me, I demonstrate a joint dependency on the logistical practice
of empty container repositioning on both legs of round trip services between the
US and the rest of the world. I find that limitations on this practice may under-
mine the aforementioned benefits of containerization. Furthermore, routes that
maintain particularly high asymmetries in trade volume, such as shipping lanes
2Transport costs are often treated as an exogenous model primitive, commonly referred as
an iceberg cost, which represented a fixed percentage of value-attrition while a good is in transit
(Samuelson, 1952).
3These cost saving measures also coincided with greater precautionary inventory management
and higher delivery time volatility.
9
Unconventional Protectionism
between the US and China or Japan, are far more exposed to the malaise effects
from intervention in empty repositioning.
Third, this paper adds to the literature examining the motivation and effects of
resurgent trade protectionism. Such decisions are largely a reflection of the state
of policymakers’ underlying constituent bases, which are subject to adverse de-
velopments in social identification patterns (Grossman and Helpman, 2021; Bom-
bardini et al., 2023). While resurgent protectionism often leads to welfare losses
(Sampson, 2017; Fajgelbaum et al., 2020; Bown, 2021; Fajgelbaum and Khan-
delwal, 2022), infant industries may find themselves on more favourable growth
trajectories (Juhász, 2018). While there is a well-documented understanding of
how demand-side interventions influence trade outcomes (e.g., tariffs and quotas),
OSRA22 relates to supply-side elements of trade by constraining the use and avail-
ability of transport equipment. This study represents the first and only paper to
consider this unconventional form of protectionism. I find that although exports
are stimulated by these restrictions, overall trade activity declines – suggesting
that the policy is protectionist in nature. My results also suggest that this new
tool is precise in targeting net exporters, particularly those with a greater reliance
on empty containers from the US.
Lastly, this paper relates to the theoretical literature of round trip transport
services. Given that the volumes of transported goods between two locations are
often imbalanced, shipping capacity on the lower volume ‘backhaul’ route is un-
derutilized. As Demirel et al. (2010) demonstrates, the ‘backhaul’ freight rate
drops to zero under perfect competition and perfect information. Both Demirel
et al. (2010) and Wong (2022) remedy this deviation from observed freight rates
by either (i) enforcing balanced trade flows across round trips, or (ii) introducing
imperfect information and a matching process into the model. Ishikawa and Tarui
(2018) solves for positive bilateral freight rates by introducing imperfect competi-
tion. I approach this challenge by instead using physical equipment as inputs in
a joint profit function of round trip transport services. To ensure the continued
service of the high-volume leg of an imbalanced round trip, a transport operator
redistributes empties. Under imbalanced trade, the marginal revenue of shipping
an additional loaded container on the high-volume route is equal to the cost of
handling that loaded unit plus the cost of returning one empty container. In con-
trast, transporting one additional loaded unit on the low-volume leg of a round
trip occupies an existing empty, resulting in a freight rate equal to the loaded
handling cost less the cost of returning an empty unit. Under specific assump-
10
Unconventional Protectionism
tions, bilateral freight rates are both positive and the low-volume route maintains
a relatively lower freight rate, as predicted in Hummels et al. (2009). This pricing
scheme under asymmetric volumes relates closely to peak load pricing strategies
featured in round trip passenger flights and the dynamic pricing on highway toll
lanes (Williamson, 1966; Cooks and Li, 2023).
The remainder of the paper proceeds as follows. I detail how container redis-
tribution operates and outline the factors which contribute to empty container re-
distribution. Section 3 outlines a partial equilibrium model of containerized trade.
Section 4 provides a brief description of the novel data that I have collected, and
Section 5 presents stylized facts of containerized trade. In Section 6, I calibrate
and estimate the exogenous parameters of the empty container model and consider
the counterfactual effects of government intervention aimed at limiting the outflow
of empty container units from the US. Section 7 concludes.
2.2 Background
Since the emergence of container technology, this form of transport equipment
has grown to become a worldwide norm. As Levinson (2016) explains, container
unit standardization was the key development that led to the modern day scale of
intermodal transportation. This challenge, starting in the late 1950s, represented
ten years of negotiations in which time the industry determined that the standard
containers would be 20-ft & 40-ft in length. Additionally, corner fittings used to lift
individual units and interlock units together were also agreed upon. These efforts
resulted in a flexible, harmonized system in which transport equipment could be
freely redistributed back and forth within a given round trip. The subsequent
global adoption of container technology across ports has yielded a complex network
of supply chains which operates at lower costs but represents greater risks through
increased uncertainty surrounding delivery times (Carreras-Valle, 2022).
Although container shipping and the repositioning of empty containers have
been a long-held practice in international trade, it is important to understand
why economic agents coordinate in this manner. For a transport operator, bilat-
eral transport service demand within a given round trip can differ. This would
contribute towards net exporters shipping more loaded container units out to a
given destination than those that make their way back from the net importer. To
accommodate required container inventory across ports, container repositioning
11
Unconventional Protectionism
features empty units on the backhaul (lower volume) leg of a given round trip.
In essence, this behaviour reflects an inventory management problem in which a
cost–minimizing assignment of container capacity and flows must be determined.4
Lee and Song (2017) describes two considerations that transport operators
face under imbalanced round trip trade; (i) a quantity decision, in which the firm
decides how many empty containers to store at each port, and when and how many
to move between ports, and (ii) a cost estimation of empty repositioning, which
contributes to how freight rate prices are determined. Regarding the quantity
decision, Song and Dong (2015) refers to two key considerations. Upon adopting
a network flow model, origin-destination based matrices specify the quantity of
empty containers to be moved from one node to another. The goal of this decision is
to satisfy flow balancing, where container flows between two nodes should be equal.
The second item addresses uncertainties by adopting inventory control models to
produce decision-making rules which dynamically determine the amount of empty
repositions in and out of a node. I incorporate the associated contribution of empty
container repositioning costs to freight rates and enforce a balanced container flow
constraint between nodes such that combinations of loaded and empty container
units can be accounted for on the backhaul (lower volume) leg of a given round
trip. However, given that I use a static model, I do not feature decision-making
rules and uncertainty for individual firms.
The transport logistics literature therefore recognizes the scale of the empty
container repositioning problem to be a product of underlying asymmetries in im-
port demand volumes between service nodes and uncertainty surrounding vessel
delivery times, inter-reliances on other modes of transport and demand volatility.
For the purposes of this paper, I focus on the long-term determinants of variation
in empty container repositioning through imbalanced trade. The greater the asym-
metry in loaded container flows within a given round trip, the larger the volume
of empty container repositioning. Furthermore, the empty container repositioning
problem should be considered a longstanding and necessary feature of container-
ized trade rather than a specific byproduct of recent episodes of port congestion
and delays.
4As Lee and Song (2017) highlights, empty container repositioning functions similarly to
conventional manufacturing logistics in which firms strategically relocate their inventory in order
to meet consumer demand. In the case of containerized round trip shipping, exporters consume
transport services from transport operators and container units are redistributed in order to be
readily available for further shipping service demand. When volumes of service demand differ
on these continuous loops of transportation, firms strategically relocate empty container units to
sustain the service of their larger export volume destination.
12
Unconventional Protectionism
2.3 Model
In this section, I specify the empty container repositioning problem in an aug-
mented Armington model based on Hummels et al. (2009) and Wong (2022). I
include three representative agents: consumers, producers and transport opera-
tors. Endogenous transport costs are a function of per-unit loaded and empty
container handling costs. I first outline key assumptions and then solve the model.
Lastly, I establish a set of comparative statics that explain variation in empty
repositioning.
2.3.1 Assumptions
I consider an international economy of round trip containerized trade that features
J heterogeneous countries, where each country produces a unique variety of a
↔
tradeable good. The term ij denotes a round trip route that services trade between
countries i and j. Consumers in country j are endowed with one unit of labor that
is supplied elastically, exhibit a love of variety across consumable goods and are
geographically immobile. A representative consumer at location j is assumed to
maximize a quasi-linear utility function:
∑M
(ϵ−1)/ϵ
max Uj = lj0 + aijlij , ϵ > 1, (2.1){lj0,...,lij}
i=1
where lj0 represents the quantity of the numeraire good consumed in country j
and lij represents the quantity of a tradeable variety sourced from country i.
5 Het-
erogeneous countries maintain route-specific preference parameters, aij, for each
tradeable variety. A single unit of a good is associated with one unit of transport
equipment utilized. Therefore, lij is equivalent to the number of loaded containers
shipped from i to j. The price elasticity of demand, ϵ, is common across varieties
and routes.
Producers are perfectly competitive and produce variety j using inputs of labor.
I assume that the price of transported goods from i to j increases through the
following components; (i) the domestic wage rate, wi; (ii) the specific tariff rate of
the given ij leg of the round trip, τij; and (iii) the per-container freight rate, T
6
ij.
5The numeraire good is traded at no cost and maintains a unit price of 1.
6Holmes and Singer (2018) highlights an indivisibility of transport costs due to per-container
freight rates not varying based on variation in the usage of containers’ cubic volume capacity.
13
Unconventional Protectionism
pij = wiτij + Tij (2.2)
Intermediary transport operators are perfectly competitive and service a given
↔
bilateral trade route, ij. The profit maximization problem for the transport oper-
↔
ator servicing route ij is a joint-profit function that considers the optimal bundle
of container inputs. This is a variation of the joint-profit function featured in
Behrens and Picard (2011), in which I add a balanced container flow constraint.
max π↔ =Tijlij + Tjilji − c↔(lij + lji)− r↔(eij + eji) (2.3)
{lij ,lji,eij ,e } ij ij ijji
s.t. lij + eij = lji + eji
↔
Revenue generated from servicing route ij is the sum of each leg’s respective
freight rate times the loaded container quantity. Costs are determined by loaded
and empty states of container inputs used to provide services. The costs of per-
unit loaded and empty container handling is represented, respectively, by the set
{c↔, r↔}.7 Due to equidistant travel across routes ij and ji and the minimal atten-
ij ij
tion that incoming empty containers {eij, eji} require to be repurposed, I assume
that handling costs are invariant to voyage direction and empties are cheaper to
handle.8 Bilateral flows of container units, irrespective of their state, are balanced
as a result of transport operators needing to sustain container inputs on both sides
of a given round trip. This constraint is affirmed in the first stylized fact in Section
3 (Figure 2.1).
In the next subsection, I depict the profit maximization problem under weakly
imbalanced trade. In the case of balanced trade, Eq. (2.3) is subject to a constraint
of equivalent bilateral flows of loaded container units and the empty container
redistribution problem is nonexistent. The resulting system of equations are solved
for in Appendix II and mirror the balanced container case featured in Wong (2022).
7Following Notteboom et al. (2022), I attribute container handling costs to the transport
operator. This study highlights that operators spend, on average, 15% of fleet management
costs on empty repositioning.
8Appendix I considers homogeneous input prices. Similarly to a footloose capital model of
Behrens and Picard (2011), this specification yields zero freight rates on low volume legs of round
trip trade. Given that I do not observe zero empty container flows, nor zero freight rates across
observed data, I conclude that there must be differences input prices across containers which
vary by their state.
14
Unconventional Protectionism
2.3.2 Weakly Imbalanced Trade
↔
Suppose country j is a weak net importer of route ij, where lij ≥ lji. This leads to
a prevailing empty redistribution problem, and the profit function is subject to a
balanced container flow constraint, lij = lji+eji, where maximum service capacity
is pinned down by max{lij, lji}. This is consistent with other imbalanced trade
models under a round trip setting (Ishikawa and Tarui, 2018). To ensure positive
bilateral freight rates under imbalanced trade, I assume that the per-unit shipment
↔
cost of empties is cheaper than loaded handling on every route: c↔ > r↔ ∀ ij.
ij ij
The profit maximization problem is expressed as
max π↔ = Tijlij + Tjilji−c↔(lij + lji)− r↔(0 + eji) (2.4)
{l ij ij ijij ,lji,eji}
s.t. eji = lij − lji
Upon substituting the balanced container flow constraint into the profit maxi-
↔
mization problem, freight rates for both legs of a given round trip ij are determined.
Due to the price-taking nature of this perfectly competitive transport operator,
these prices are underpinned by the marginal costs of container redistribution.
T ∗ij = c↔ + r↔ , T
∗
ji = c↔ − r↔ (2.5)ij ij ij ij
These first order conditions intuitively state that the marginal benefit of an
additional loaded container on the larger volume leg, from net exporter i to net
importer j, is equal to the the direct per unit shipping cost, c↔, and the cost
ij
of an additional empty container on the return trip, r↔. An additional loaded
ij
↔
container transported from j to i represents one less empty on route ij, which
implies the added cost of c↔ being partially compensated for by a cost reduction
ij
of r↔. Expressions for these bilateral freight rates can be substituted into Eq.
ij
(2.2).
p∗ij = wiτij + c
∗
↔ + r↔, pji = wjτji + c↔ − r↔ (2.6)ij ij ij ij
To solve for {l∗ , l∗ij ji}, I insert Eq. (2.6) into the demand function for imported
varieties.
15
Unconventional Protectionism
( )−ϵ
ϵ 1 ( )−ϵ
l∗ij = ( wiτij + c↔ + r↔ϵ− 1 aij) ij ij−ϵ
∗ ϵ 1
( )−ϵ
lji = w− j
τji + c↔ − r↔
ϵ 1 aji ij ij
The net difference in flows determines the empty container flow quantity and
direction of flow. In this case l∗ij = max{lij, lji}geql∗ji, which implies that empties
will trave(l on th)e low(er volume backhaul route ji.−ϵ
ϵ 1 −ϵ ( )−ϵ 1 −ϵ ( ) )−ϵ
e∗ji = wiτij + c↔ + r↔ − wjτ− ji
+ c↔ − r↔
ϵ 1 aij ij ij aji ij ij
(2.7)
The resulting equilibrium trade quantities, {lij, lji}, and values, {Xij, Xji}, on
↔
route ij are decreasing in the marginal cost of loaded container transport, local
wages, and import tariffs i(mposed by)the destination country.−ϵ
ϵ 1 ( )1−ϵ
X∗ij =( − ) (wiτij + c↔ + r↔ϵ 1 aij ij ij−ϵϵ 1 )1−ϵ
X∗ji = wjτji + c↔ − r↔ (2.8)ϵ− 1 aji ij ij
However, variation in empty container handling costs, r↔, will have counteract-
ij
ing effects on outcome variables for a given round trip, highlighting a round trip
effect in the model. For example, suppose the cost of empty outflows from country
j rises. Not only does this stimulate j’s exports, as existing cargo space on leg
ji is reallocated from empty repositioning to exports, but in addition, transport
↔ ↔
capacity of route route ij, relfected by l∗ij, declines. Transport services on route ij
decline due to the associated cost of maintaining imbalanced container flows.
2.3.3 Comparative Statics
Consider first a set of demand shocks to consumer preferences {aij, aji} and import
tariff adjustments {τij, τji}. In each case, a marginal change implies the follow-
↔
ing adjustments to the trade outcomes for route ij. Recall for the trade value
expression that we assume ϵ > 1.
16
Unconventional Protectionism
∂T ∗ ∗ ∗ ∗ij ∂Tji ∂pij ∂pji
( = 0 ,) ( = 0 , ) = wi > 0 , = 0∂τij ∂τij ∂τij ∂τij∂l∗ −ϵ −(ϵ+1) ∗ij ϵ 1 ∂lji
= −ϵwi w
∂τij ϵ−( 1 aij∗ ) i∂X −ϵϵ 1 (
τij + c↔ + r↔
ij ij ) < 0 , = 0∂τij−ϵ
ij ∂X
∗
− ji= (1 ϵ)(wi w τ + c↔ + r↔ < 0 , = 0∂τij ϵ− ) i ij1 aij ( ij∂e∗ −ϵ ) ij ∂τij−(ϵ+1)ji
= − ϵ 1ϵwi w τ− i ij
+ c↔ + r↔ < 0
∂τij ϵ 1 aij ij ij
A preference shock in country j for goods from country i would be represented
by aij increasing. The resulting adjustments to outcome variables in this model
are as follows.
∂T ∗ ∗ ∗ ∗ij ∂Tji ∂pij ∂pji
( = 0 , = 0 , = 0 , = 0∂aij ) (∂aij ∂)aij ∂aij∂l∗ − ϵ−1ij ϵ 1 ϵ− 1 −ϵ ∂l∗ji
= ϵ ( aij) ( wiτij + c
↔ + r↔ > 0 , = 0
∂aij ϵ ϵ ij ij ∂aij
∂X∗ ϵ−1 )1−ϵ ∗ij ϵ− 1 ϵ− 1 ∂Xji
= ϵ ( aij ) (wiτij + c
↔ + r↔ ) > 0 , = 0∂aij ϵ ϵ ij ij ∂aij∂e∗ ϵ−1ϵ− 1 ϵ− 1 −ϵji
= ϵ aij wiτij + c↔ + r↔ > 0
∂aij ϵ ϵ ij ij
Since these are perfectly competitive firms providing transport services, quan-
tity supplied and freight rates are unresponsive to demand-side adjustments. How-
ever, when underlying costs of these services adjust, the corresponding freight rates
charged will be adjusted uniformly. Endogenous transport costs are simply a linear
function of underlying costs of shipping the required container inputs. Suppose
the underlying cost of repositioning empty containers increases. This will make
the existing trade balance less viable to manage. In response, firms must exhibit a
widening of the freight rate ‘gap’ between ij and ji, where the net exporter coun-
tries sees freight rates of outgoing goods increase and net importer countries see
freight rates of outgoing goods decline. This results in the trade balance narrowing
and the ‘backhaul’ problem shrinking in scale.
17
Unconventional Protectionism
∂T ∗ij ∂p
∗
ij ∂T
∗ ∗
ji ∂pji
= > 0 , = < 0
∂r↔ ( ∂r↔ ) ∂r↔ ∂r↔ij ij ij ij
∂l∗ −ϵ ( )−ϵ−1ij ϵ 1
= −ϵ( ) wiτij + c
↔ + r↔ < 0 ,
∂r↔ ϵ− 1 aij ( ij ijij∂l∗ −ϵϵ 1 )−ϵ−1ji
= ϵ( ) wjτji + c
↔ − r↔ > 0 ,
∂r↔ ϵ− 1 aji ij ij
ij
∂X∗ −ϵ ( )−ϵij
= (1− ϵ 1ϵ) ↔( ) wiτij + c + r
↔ < 0 ,
∂r↔ ϵ− 1 aij ij ij
ij
∂X∗ −ϵϵ 1 ( )−ϵji
= (ϵ− 1) wjτji + c↔ − r↔ > 0,
∂r↔ ϵ− 1 aji ij ij
ij
∂e∗
( )−ϵ
ϵ 1 ( )−ϵ−1ji
= −ϵ( w− iτij + c
↔ + r↔ −
∂r↔ ϵ 1 aij
ij )−ϵ
ϵ 1 (
ij ij )−ϵ−1
ϵ wjτji + c↔ − r↔ < 0
ϵ− 1 aji ij ij
Proposition 1. Under the assumption of competitive transport firms and imbal-
anced trade,
(i) When transport costs are endogenous and constrained under balanced con-
tainer flows, an increase in the tariff rate of imports from i to a net importer
country j, τij, reduces the scale of the backhaul problem destined for the
∂e∗
partner net exporter country i: ji < 0
∂τij
(ii) When transport costs are endogenous and constrained under balanced con-
tainer flows, an increase in j’s preferences for variety i, aij, increases the
scale of the backhaul problem destined for the partner net exporter country
∂e∗
i: ji > 0
∂aij
(iii) When transport costs are endogenous and constrained under balanced con-
tainer flows, an increase in the per unit cost of empty container inputs, r↔,
ij
reduces the scale of the backhaul problem, given that freight rates resultingly
∂T ∗ ∂T ∗
rise on the full route ij and lessen on the return route ji: ij > 0, ji <
∂r↔ ∂r↔
ij ji
∂e∗
0, ji < 0
∂r↔
ij
18
Unconventional Protectionism
The relationship between the scale of the empty container redistribution prob-
lem and the skewness of the existing trade balance can be examined in a pro-
portional manner. These expressions simplify otherwise non-linear relationships
between outcome variables to a reduced linear relationship that can be taken di-
rectly to the surrounding data, should one be equipped with bilateral container
traffic flows, as well as containerized trade values. I represent the scale of the
empty container redistribution problem with Eji, which indicates the share of
empties as a percentage of total container outflows from a net importer country j
to net exporter i. ( )
e∗
( )ϵ w τ + c + r ϵa i ij ↔ ↔∗ ji ji ij ijEji = ∗ ∗ = 1− (2.9)lji + eji aij wjτji + c↔ − r↔ij ij
Proposition 2. Under the assumption of competitive transport firms and imbal-
anced trade,
(i) When transport costs are endogenous and constrained under balanced con-
tainer flows, an increase in the tariff rate of imports from i to a net importer
country j, τij, reduces the scale of the backhaul problem destined for the
∂E∗
partner net exporter country i: ji < 0
∂τij
(ii) When transport costs are endogenous and constrained under balanced con-
tainer flows, an increase in j’s preferences for variety i, aij, increases the
scale of the backhaul problem destined for the partner net exporter country
∂E∗
i: ji > 0
∂aij
(iii) When transport costs are endogenous and constrained under balanced con-
tainer flows, an increase in the per unit cost of empty container inputs, r↔,
ij
reduces the scale of the backhaul problem, given that freight rates rise on
∂T ∗ ∂T ∗ ∂E∗
the full route ij and lessen on the return route ji: ij > 0, ji > 0, ji < 0
∂r↔ ∂r↔ ∂r↔
ij ji ij
Examining the skewness of the trade balance using an import-export ratio from
X
j’s perspective: ji
Xij
( ) ( )∗ −ϵ w τ + c + r 1−ϵX i ij ↔ ↔ij aji ij ij
∗ = (2.10)Xji aij wjτji + c↔ − r↔ij ij
Using Eq. (2.9) and (2.10), I find that any exogenous shock to empty outflows
will adjust the import-export ratio in the same sign direction for trade route ij.
19
Unconventional Protectionism
For example, should US(preferenc)es for goods from China rise, the existing trade
X
deficit would increase ∆ ij > 0 and the associated scale of empty container
Xji
redistribution originating from the US would rise (∆Eji > 0).
9
2.4 Data
The main data set of the paper combines monthly US port samples of containerized
trade and associated container traffic flows, both for empty and loaded units.
Auxiliary tariff and wage data is used for the calibration of exogenous parameters
throughout the counterfactual analyses of this study.
2.4.1 Containerized Goods
I use monthly trade data from the US Census Bureau, which details the imports
and exports of containerized goods at the US port level by value and weight for
each US trade partner. The available sample period begins with January 2003 and
provides commodity-level stratification down the to six-digit Harmonized System
(HS) level. Using this data, I form a balanced panel of the top 14 port locations
for containerized trade flows.10 In cases of port alliances, I assume that port
infrastructure is jointly utilized between ports. The ports of Seattle & Tacoma as
well as New York & Newark are each combined into two unique port authorities,
the NWSA and PANYNJ, respectively.
2.4.2 Container Traffic
Using this informed shortlist of the top containerized US ports, I approached each
respective port authority individually and retrieved monthly 20-foot equivalent
unit (TEU) traffic flow data. I received four separate series: (i) inbound loaded
containers, (ii) outbound loaded containers, (iii) inbound empty containers, and
(iv) outbound empty containers. To my knowledge, this is the first study in
international economics to document and use novel empty container repositioning
data. Unlike containerized goods flows, I do not observe the origin or ultimate
9I test this identity empirically in Subsection 5.1 and find significance at a monthly frequency.
10These individual ports include New York (NY), Los Angeles (CA), Houston (TX), Long
Beach (CA), Norfolk (VA), Savannah (GA), Charleston (SC), Oakland (CA), Newark (NJ),
Seattle (WA), Tacoma (WA), Baltimore (MD), New Orleans (LA) and Jacksonville (FL).
20
Unconventional Protectionism
destination of container traffic flows. A 40-foot intermodal container is counted as
two TEUs. To ensure a balanced and representative panel of data, I have limited
container traffic flows to those observed between January 2012 and December 2021
of 12 key ports, which represents approximately 80% of national container unit
thruflows. For more details on the wider time series of port data made available
for this study, see Appendix III.
2.4.3 Auxiliary Data
For the quantitative exercises detailed in Section 6, I calibrate observable param-
eters of wages and tariffs through the use of monthly manufacturing wages and
specific tariff rates data. Time series of monthly wages between 2012 and 2021
are sourced from the International Labor Organization (ILO), which specifies an-
nual averages of manufacturing wages in USD value. To account for unreported
wage values for specific years of the data, I use OECD annualized growth rates
of average monthly manufacturing wages and infer the associated level amounts.
I use the U.S. Bureau of Labor Statistics’ “Consumer Price Index for All Urban
Consumers”, which excludes contributions made by food and energy, to deflate
these series. I leverage use of the UNCTAD Trade Analysis Information System
(TRAINS) database for effective tariff rates on manufactured goods between the
US and its trade partners. ‘Manufactures’ are an SITC 4 product group predefined
on the World Integrated Trade Solution (WITS) platform of the World Bank.
2.5 Stylized Facts
In this section, I present two stylized facts which test the validity of the balanced
container flow constraint and the hypothesized negative relationship between the
share of empty container outflows and the export-import value ratio of container-
ized goods. While many of these facts have previously been theorized, this study
is the first to directly document the responsiveness of the empty container redis-
tribution problem to variation in the US trade balance. Additionally, I provide
port-level evidence which suggests that the volume of container traffic at a given
port is a strong predictor of whether said port acts as a net inflow or net out-
flow in terms of its contribution to nationally balanced container flows. I use this
third stylized fact to motivate my treatment of the European Custom Area as a
single entity, which at only this scale of operations maintains balanced container
21
Unconventional Protectionism
redistribution comparable to the US.
2.5.1 Empty Repositioning & Trade Balance Asymmetry
Stylized Fact 1. A positive deviation in country j’s export-import ratio with
country i is correlated with a negative deviation in the volume of empty container
units shipped from j to i as a share of total container units shipped from j to i.
When combined, Eq.(2.9) and Eq.(2.10) imply that a higher export-import
ratio of a net importer, the US in this case, implies lower empties as a percentage
of total container outflows. As US imports from a net exporter country rises
(Xji/Xij ↓), the asymmetry in trade volumes between these two countries grows,
which implies that the logistical burden in servicing imbalanced trade – through
the repositioning of empty container un
( )(
its – has grown (E)ji ↑).
∗ X
∗ wjτji + c↔ − r↔
ji ij ij
Eji =1− ∗ (2.11)Xij wiτij + c↔ + r↔ij ij
Given that I do not observe container flows between the US and individual
countries, I instead aggregate across US ports and I test this negative relationship
empirically through variation in trade and container flows between the US (j) and
the rest of the world ((i), ) ( )
∗ X
∗ ∗
jit Xijt
Ejit = α + β ∗ + µjit , E
∗
ijt = α + β ∗ + µijt, (2.12)Xijt Xjit
where β < 0 is my proposed null hypothesis. I use four measures of trade
balance skewness: the export-import ratio, Exports , a net-gross ratio featured in
Imports
Brancaccio et al. (2020), Exports - Imports , and their respective opposites of Imports
Total Trade Exports
and Imports - Exports when addressing inflows of empties. As displayed in Table
Total Trade
2.1, a relatively smaller US trade deficit is associated with a lower scale of empty
redistribution. This highlights adjustments in the empty repositioning burden that
transport operators face, given variation in bilateral trade volumes across round
trips. In Table 2.2, I use the Net-Gross ratio featured in Brancaccio et al. (2020),
and observe further support for this proposed relationship between the prevailing
trade imbalance and the size of the empty container redistribution problem.
I next examine co-movement between empty container outflows and opposite-
direction US imports of containerized goods. As featured in Table 3.2, empty
22
Unconventional Protectionism
Table 2.1: Trade Flow Ratio & Empty Shares
Dependent Variable: Empty Container Share of Total Flows
Outbound Inbound
Model: (1) (2) (3) (4)
Export/Import (USD) -0.9575∗∗∗
(0.0687)
Export/Import (kg) -0.3909∗∗∗
(0.0288)
Import/Export (USD) -0.0253∗∗∗
(0.0062)
Import/Export (kg) -0.0327∗∗∗
(0.0097)
Mean Dep. Var 43.51% 7.47%
Mean Regressor 0.322 0.711 3.143 1.427
n-obs 120 120 120 120
Within R2 0.58 0.68 0.30 0.15
Heteroskedasticity-consistent ‘White’ standard-errors. Codes: ***: 0.01, **: 0.05, *: 0.1. Examines variation
empty containers as a share of total container outflows, given variation in the skewedness of the trade balance. I
use month and year fixed effects to control for influences of the US business cycle and seasonality.
Table 2.2: Net-Gross Ratio & Empty Shares
Dependent Variable: Empty Container Share of Total Flows
Outbound Inbound
(Model: ) (1) (2) (3) (4)
Net Exports USD -0.8510∗∗∗ 0.2322∗∗∗(Gross Trade) (0.0703) (0.0428)
Net Exports KG -0.5756∗∗∗ 0.1121∗∗∗
Gross Trade
(0.0514) (0.0308)
Mean Dep. Var 43.51% 7.47%
Mean Regressor -0.514 -0.172 -0.514 -0.172
n-obs 120 120 120 120
Within R2 0.57 0.65 0.37 0.21
Heteroskedasticity-consistent ‘White’ standard-errors. Codes: ***: 0.01, **: 0.05, *: 0.1. Examines variation
empty containers as a share of total container outflows, given variation in the net-to-gross trade balance. I use
month and year fixed effects to control for influences of the US business cycle and seasonality.
container flows are highly correlated with opposite-end flows of US trade. My
results suggest that a 1 percent rise in US imports is, on average, associated with a
1.45% increase in outflows of empty container units from the US. In sharp contrast,
significant unilateral responses are not detected for given roundtrips between the
23
Unconventional Protectionism
US and the rest of the world. I assume this co-movement is primarily driven by
exogenous variation in route-specific unobservables such as preference parameter
shifts across consumer bases, wage variation, container unit handling costs and
tariff rate adjustments. Examining the robustness of these results in Appendix IV,
I find that variation in the weight of opposite-end trade flows is also predictive of
adjustments in empty container repositioning. Additionally, upon disaggregating
to within-port variation I find similar patterns of positive co-movement between
trade flows and the opposite-end empty container repositioning problem.
Table 2.3: Empty Container Elasticity with Respect to Trade Flows
Dependent Variable: Empty Container Flows (TEU)
ln(Outbound) ln(Inbound)
Model: (1) (2) (3) (4)
ln(Inbound Trade) 1.450∗∗∗ -0.0903
(0.0917) (0.2267)
ln(Outbound Trade) 0.3229 0.8409∗∗∗
(0.2489) (0.2559)
n-obs 120 120 120 120
Within R2 0.59 0.02 0.002 0.16
Heteroskedasticity-consistent ‘White’ standard-errors. Codes: ***: 0.01, **: 0.05, *: 0.1. US empty container
flows are regressed on US containerized trade flows, expressed in terms of deflated USD. For example, a one
percent increase in ‘Inbound Trade’ is associated with a 1.45% rise in outbound empty container flows. I use
month and year fixed effects to control for influences of the US business cycle and seasonality.
2.5.2 Balanced Container Flows
Stylized Fact 2. A positive deviation from the total container units transported
from i to j is correlated with a positive deviation from the total container units
transported from j to i.
Thus far I have shown that trade balances are strongly indicative of the scale of
the empty container redistribution. Upon aggregating across US ports, evidence
suggests that national levels of container inflows and outflows appear largely bal-
anced, but only when incorporating contributions made by empty container redis-
tribution. This lends strong support for the balanced container flow constraint,
which underpins my partial equilibrium model of empty container redistribution.
In Table 2.4, I regress the total number of inbound container units on the total
number of outbound containers at the national level. These results suggest that
24
Unconventional Protectionism
a system of balanced container exchanges exists even within a given month of
containerized transport, as highlighted by the reported coefficient not statistically
differing from 1 at a 99% confidence level. In contrast, when focusing on only
loaded container exchanges, a far more commonly reported measure of container
traffic at the port level, this balance in the exchange of transport equipment is left
completely obscured.
Table 2.4: Balanced National Container Flows
Dependent Variable: ln(Inbound Container Flows)
Total Loaded Empty
Model: (1) (2) (3)
ln(Total Outbound) 1.012∗∗∗
(0.0210)
ln(Loaded Outbound) -0.0913
(0.2841)
ln(Empty Outbound) -0.4641***
(0.0314)
Observations 120 120 120
Within R2 0.94 -0.007 0.62
Heteroskedasticity-consistent ‘White’ standard-errors. Codes: ***: 0.01, **: 0.05, *: 0.1. Container flows inbound
to the US are regressed on outbound container flows. Results are consistent with estimates using month and year
fixed effects.
Figure 2.1 suggests that container flows remain balanced across widenining
windows of time. As I expand the relevant time interval, through backward sums
of loaded and empty container unit flows, noise surrounding these estimations
lessens and levels remain approximately balanced at the 1-to-1 percentage point
ratio. Although larger aggregations of container flow statistically deviate from the
1-to-1 ratio of balanced container flows, these deviations are low in power, only
ranging between 1 to 2 percent in size.11
Container unit measures have largely been focused on loaded traffic flows and
often rely on imputed measures available through third-party private entities such
as S&P Panjiva (Flaaen et al., 2021; Steinbach, 2022; Ardelean et al., 2022). Fo-
11This is likely a symptom of my sample of ports being based on the largest ports in the US.
As I highlight in my next stylised fact, although my container flow data represents over 80% of
total container traffic in the US, the smaller ports that I exclude from my sample most likely
function as net outflows of container units. With their inclusion, and a full representation of the
population of container flows, I hypothesize that my mild deviations from the 1-to-1 balanced
container exchange would reduce in size.
25
Unconventional Protectionism
Figure 2.1: Balanced National Container Flows by Time Window
1.10
1.05
1.00
0.95
0.90
Time Interval
Heteroskedasticity-consistent ‘White’ standard-errors. Both the dependent variable and regressor are log-
transformed. Total inbound containers are summed across a balanced panel of 12 US ports and represent both
loaded and empty containers, is regressed on total outbound containers for these same set of ports. Sums are taken
across windows of varying lengths of time, ranging from bilateral exchanges within a single month to exchanges
across 12 month backward sums.
cusing only on loaded container units – whether reported directly by ports or
estimated using Bill of Laden Records from the U.S. Customs and Border Pro-
tection (CBP) agency – conceals logistical efforts between the US and the rest of
the world. No semblance of balanced container flow patterns are present when ex-
cluding empty container repositioning and focusing only on loaded container flows
(Figure 2.2).
These findings, when jointly considered, suggest that the system of intermediate
transport equipment present in the US achieves a balanced exchange of transport
equipment, only when taking into account empty units. In the next subsection, I
consider how individual ports contribute to nationwide balanced container flows.
2.5.3 Port Heterogeneity
Stylized Fact 3. A positive deviation in the total volume of container inflows
and outflows of port p is correlated with a positive deviation from the net volume
in container inflows less outflows of port p.
26
Change in Total Inbound Container Units (%)
1 month
2 months
3 months
4 months
5 months
6 months
7 months
8 months
9 months
10 months
11 months
12 months
Unconventional Protectionism
Figure 2.2: Imbalanced National Loaded Container Flows by Time Window
1
0
-1
Time Interval
Heteroskedasticity-consistent ‘White’ standard-errors. Both the dependent variable and regressor are log-
transformed. Total inbound containers are summed across a balanced panel of 12 US ports and represent both
loaded and empty containers, is regressed on total outbound containers for these same set of ports. Sums are taken
across windows of varying lengths of time, ranging from bilateral exchanges within a single month to exchanges
across 12 month backward sums.
Although total container flows – both loaded and empty containers – are bal-
anced at the national level, patterns in port-level container flows highlight that
the largest ports in the US function as net inflows of total containers, while mid-
tier sized ports act as net outflows of total container units. This suggests that
an interdependence exists across ports, which maintains balanced container flows
at a national level. To the best of my knowledge, these statuses across ports
have not yet been documented in the transport economics literature. In Figures
2.3a and 2.3b, I display annual net differences in total container flows by port for
2017 along with the geographic dispersion of these key entry and exit points for
container equipment.
These statuses are consistent across time. Los Angeles, Long Beach, PANYNJ,
and NWSA act as net inflows whereas the remaining set of mid-tier ports are net
outflows. As displayed in Figure 2.4, the total thruflow of loaded and empty con-
tainers at a given port is highly predictive directional status. This pattern likely
relates to comparative advantages in handling vessels of varying sizes. Larger ports
may attract net inflows due to their relatively higher efficiency in handling arriving
27
Change in Loaded Inbound Container Units (%)
1 month
2 months
3 months
4 months
5 months
6 months
7 months
8 months
9 months
10 months
11 months
12 months
Unconventional Protectionism
Figure 2.3: Port Specialization by Net Inflow Status (2017)
Panel A: Net Inflow of Total Container Units by Port
400
200
0
-200
gin
ia nd ton illea an
s
nah de
s ore tons s SAl v le n la m  N
J
eac
h les
ir ak earl son Or va rg i
u W & ge
f V O h ck w Sa ve Ba
lt Ho N NY
 g B An
rt o C a e  E f
 n  
o J N ort rt o L
o osL
P P Po
Panel B: Geographic Dispersion of Net Inflows
Northwest Seaport Alliance
P.A. of NYNJ
Baltimore
Oakland
Norfolk
LA & LB
Charleston
Savannah
New Orleans Jacksonville
Houston
Port Everglades
Net Inflow Net Outflow
goods (Blonigen and Wilson, 2008). This pattern may also be partly explained by
the ‘hub and spokes’ mechanism in which larger vessels travel between port hubs in
28
Net Flow, '000 TEU
Unconventional Protectionism
order to exploit lower per-unit transport costs (Ganapati et al., 2021). Addition-
ally, one may levy use of a proximity-concentration argument, in which case the
best of both worlds would be for imports to arrive at ports positioned closely to
high density population centers such as California and New York (Ducruet et al.,
2018). Upon examining average vessel sizes between these port groups, I find that
larger vessels arrive at larger net inflow ports, where per-unit import prices are
likely cheaper (Table 2.5).
Table 2.5: Average Containership Gross Tonnage by Port Size
Ports 2014 2015 2016 2017 2018 2019
Major Ports 31,558 32,990 34,790 36,569 38,141 39,241
Mid-tier Ports 26,564 27,999 29,639 31,637 32,784 33,407
Note: Reports the average gross tonnage, a nonlinear measure of a ship’s overall internal volume, weighted by
the number of vessel visits in each port. Source: US Army Corp of Engineers, Port Clearance data.
Given that national bilateral container flows are balanced, yet individual ports
act as either net inflows or outflows of container units, I suggest that an interdepen-
dence across ports which has persisted since at least January 2003. As highlighted
in Wong and Fuchs (2022), shipments arriving at major ports see some portion
of goods, along with intermodal transport equipment, be transported across the
US hinterland. While some container units may return to their US port of origin,
my findings suggest that many units of equipment depart from the US through
alternative ports around the country, particularly through mid-tier sized ports.
Rather than treating each port’s trade with the world as an isolated bilateral set
of roundtrip trade routes, this container traffic data exhibits signs of a national-
level round trip effect which permeates across ports. Containers are redistributed
across US ports and collectively form a balanced container flow system necessary to
support round trip containerized trade. This motivates my counterfactual analysis
of balanced container flow trade at the country rather than port level.
The intensity of involvement in empty container repositioning also varies widely
across ports. In 2021, while larger ports adjacent to net exporter countries – such
as Los Angeles or Long Beach – shipped out 70–80% of containers completely
empty, the southern ports of Houston (TX), New Orleans (LA) and Jacksonville
(FL) have maintained historical averages of 6–22%. As displayed in Table 2.6,
while differences in these shares are longstanding, many of the larger ports and
their respective transport operators have been shouldering an increasing burden of
the growing US trade deficit and resulting rise in empty container repositioning.
29
Unconventional Protectionism
Figure 2.4: Port Specialization by Total Container Thruflow (2012-2021)
500
250
0
-250
0 3,000 6,000 9,000
Thruflow, '000 TEU
Major Ports Mid-tier Ports
Table 2.6: Average Empty Share of Container Outflows by Port-Year (%)
Port 2013 2015 2017 2019 2021
Los Angeles, CA 48.69 56.87 57.55 60.61 76.94
Long Beach, CA 46.70 55.52 58.77 61.11 69.05
Port of NY & NJ 44.68 55.34 56.96 60.31 68.89
Savannah, GA 21.91 34.61 32.50 36.00 49.93
Norfolk, VA 15.82 27.93 33.14 37.56 41.18
Charleston, SC 23.19 30.32 30.32 36.61 41.09
2.6 Counterfactual
I use the quantitative model featured in Section 3 to consider the policy implica-
tions of OSRA22. I first outline a simple two-country baseline scenario of US-RoW
(Rest of the World) round trip containerized trade. I then illustrate the flaws as-
sociated with this approach, and motivate the estimation of bilateral loaded con-
tainer flows by US trade partner. By separately representing countries, I include
two key features of round trip containerized trade; (i) bilateral flows of empty con-
tainer units between the US and RoW, and (ii) heterogeneity across trade partners’
varying extensive and intensive margins of reliance on empty container outflows
from the US. I provide a diagnostic assessment of these estimates, identify the
key set of restrictions and assumptions necessary to yield the most compelling fit
30
Net Flow, '000 TEU
Unconventional Protectionism
to UNCTAD regional container traffic data and proceed with a calibration and
estimation of model primitives. Upon establishing this multi-country baseline sce-
nario, I then introduce the counterfactual policy measure – an empty container
outflow (ECO) quota, applied through a specific per-unit tax on outgoing empty
containers. Accounting for trade partners’ varying degrees of reliance on empty
containers provides the same qualitative result of policy backfiring on the import
leg of US round trips, but introduces quantitatively larger bilateral adjustments
in containerized trade.
2.6.1 US-RoW Baseline
In a simple two-entity representation of US containerized trade, a single round trip
services all of US containerized trade. Given that I do not observe the origin or
destination of port-level container traffic in the US, this is a natural starting point
for examining how market intervention would affect containerized trade outcomes.
According to the no excess capacity constraint featured in Equation (2.4), empty
container flows can only be present on one leg of a round trip route. However,
as displayed in Figure 2.5, the US maintains positive bilateral flows of empty
containers with the rest of the world. For example, at the height of the COVID-19
supply chain crisis approximately 63% of outbound containers left the US empty
and less than 4% of incoming container units were empty. In order to reconcile this
disparity between observed data and a baseline scenario of containerized trade, I
use the net difference in empty container flows to represent the scale of the empty
container repositioning problem.
In this setting, I establish a baseline scenario of the model using trade and
container traffic data specific to average monthly levels reported in 2017.12 Using
a generalized method of moments (GMM) estimator in which a system of trade and
container flow equations, {Xij, Xji, lij, lji}, featured in Equations (2.7) and (2.8), I
can represent the endogenous set of moments in the data and exactly identify four
unknown model primitives. I reduce the number of unknown exogenous parameters
to four by calibrating observable parameters based on a trade-weighted average
of tariffs on manufactures, a trade-weighted average of monthly manufacturing
wages, and an elasticity of demand of 20.96, represented by ϵ in Eq. (3.1) and
estimated using monthly data by Wong (2022).13
12This choice of year avoids any complications that later periods associated with the China-US
Trade War and COVID-19 epidemic would introduce.
13See Appendix section V for a detailed description of baseline estimation, as well as an
31
Unconventional Protectionism
Figure 2.5: Empty Share of Container Movement by Year-Month
60
40
20
Share of container inflows Share of container outflows
While this remedy simplifies a representation of US round trip trade, the use
of net empty flows in a single round trip setting also introduces three drawbacks;
(i) an under-representation of the scale of the empty repositioning problem, (ii)
no distinguishing between net exporter and net importer statuses across US trade
partners, and (iii) no acknowledgement of differences in degrees of reliance on
the return of empty containers across net exporters. If this first point is left
unaddressed, my estimates may under–report both the substitution of transport
services from empty repositioning to US exports and the associated contraction
of vessel capacity. Secondly, no accounting of trade partners’ extensive margin of
reliance on empty container inflows from the US leads to policy effects being spread
across all participating countries. In order to determine where vessel capacity will
retract, these effects must instead be focused on the net exporter subset of trade
partners, which rely on these equipment flows. Lastly, the intensive margin of trade
partners’ reliance on empty container redistribution also needs to be represented
in this baseline scenario. Particular net exporters maintain notably more skewed
trade imbalances relative to other US trade partners, which deepens the effect
of ECO quotas on these round trips in particular. By accommodating for these
last two factors, adjustments in vessel capacity and consequential contractions in
import levels will be better reflective of particular vulnerability that net exporter
trade partners would exhibit.
assessment of model fit and depiction of the broad backfiring effect of ECO quotas under a
US-RoW setting.
32
%
2012
2014
2016
2018
2020
2022
Unconventional Protectionism
To incorporate these key features of containerized trade, I prepare a multi-
country baseline scenario, which uses observed country-level containerized goods
flows by value and estimated volumes of container unit flows to identify a full set
of unobserved exogenous parameters via GMM. In the next section, I detail how I
estimate loaded container flows by US trade partner.14
2.6.2 Multi-Country Container Flows
To establish a baseline scenario of multiple countries, I require two components;
(i) a set of calibrated parameters for each country’s round trip with the US, which
consists of the real wage and tariff rate for 2017, {wj, wi, τij, τji}, and (ii) a set of
observable trade outcomes of each round trip, which reports levels of US imports,
exports, loaded container inflows and loaded container outflows with each country,
represented by {Xij, Xji, lij, lji}, respectively. Given that I do not observe country-
specific flows of loaded container units, I estimate these values using variation in
commodity-specific weights of containerized goods exchanged between specific US-
country pairs.15
2.6.2.1 Assumptions
Container units used in shipping include a set of operational characteristics which
define the maximum weight that each individual unit can carry. Therefore, a
positive relationship exists between the number of loaded container units used in
transport and the weight of goods shipped to a given country. This fact is well-
documented in Ardelean et al. (2022), which finds a consistent synchronization
of variation in per-unit freight rates of containerized goods imported to Chile
across per-kilogram and per-TEU measures. In support of this evidence, I find
that a simple log-log regression of US loaded container inflows on the weight of
containerized US imports yields a 1-for-1 co-movement between the two measures.
Individual container units not only feature an explicit weight limit, but also
report cubic volume capacity. Both the weight and the cubic volume of a partic-
ular set of goods determines how many container units are needed for transport.
14The results of the simple US-RoW baseline setup and counterfactual exercise are detailed
in Appendix V.
15The number of countries for which I can estimate container flows is larger than the set
featured in my baseline calibration of the model. This is due to only a subset of individual
countries having average monthly manufacturing wage data available from 2012 to 2021.
33
Unconventional Protectionism
As Holmes and Singer (2018) demonstrates, the binding constraint for a given
container unit is almost always volume, rather than weight. This introduces the
possibility that differences in the dimensionality of specific products may alter the
rate at which variation in weight contributes to the number of necessary container
units used. For example, a kilogram of wooden products may utilize more of a
given container’s cubic volume capacity when compared to a metallic product of
similar weight.
To estimate the number of TEU units utilized on a given US-trade partner
round trip, I exploit monthly commodity-level variation in the weight of container-
ized goods, which is observed at the US port to country-level. I incorporate both
weight and volume considerations in the decomposition of port-level US container
using
∑J ∑J ∑K
lfpt = l
f fjk f
pjt = β wpjkt, f ∈ {Imports,Exports}, (2.13)
j=1 j=1 k=1
where at US port p, in year-month t, the total number of loaded container units
lpt is the sum of containerized weights of country j for commodity k, w
f
pjkt, times
respective loading factors, βjk. Superscript f indicates the direction that con-
tainerized goods and their associated loaded containers are moving from the US
perspective. Using these population parameters, the data generating process for a
loaded container flows between the US and country j is
∑P ∑P ∑K
lf = lf = βfjkwfUS−j,t pt pjkt, (2.14)
p=1 p=1 k=1
where combinations of observed wpjkt, and estimated β̂
fjk allows me to construct
fitted values of national container units flows in each direction across J countries.
Using this proposed identity would imply a JK number of regressors, which is
infeasible even at the HS-2 commodity level aggregation. I assume that the manner
in which cubic volume capacity determines commodity-specific loading factors does
not vary across countries. For example, should workers at the port of Mumbai
fit three metric tonnes of furniture into a single container unit, I assume that, on
average, they use available cubic volume as efficiently as workers loading containers
in Rotterdam. Given my assumption of loading factor invariance with respect to
34
Unconventional Protectionism
the country of origin, my estimation is represented as
∑K ∑J
lf = βfk wf + εfpt pjt pt (2.15)
k=1 j=1
For a given commodity traded between the US and partner countries, the
use of available volume capacity may differ on either leg of a round trip, leading
to differences in commodity-specific loading factors. While restricting loading
factors β to be invariant by direction f would double the associated observation
count of this exercise and allows me to exploit wider variation in commodity-
specific volumes, this restriction may also inadvertently pool within-commodity
variation in too aggregated a manner. For example, consider HS item 68 which
includes articles of stone, plaster and similar materials. The US may be exporting
particularly low quality stone masonry (low loading factor) while more delicate,
higher mineral quality articles may originate from Japan (high loading factor).
Should these high quality materials be associated with relatively low volumes of
kilogram weight, while low quality US exports of stone articles are associated
with high volumes of weight, this restriction would inadvertently yield a negative
coefficient in which for HS-68, as weight increases, the loading factor associated
with this shipments lowers.16
Lastly, while I do estimate loading factors across 97 HS2 commodity-level
goods, I use only the 72 HS2 products featured in the UNCTAD’s Trade Analysis
Information System (TRAINS) SITC product group of ‘manufactures’ in estab-
lishing a multi-country baseline scenario of the model. This is due to my reliance
on manufacturing wage data in the calibration of the model.
2.6.2.2 Loading Factor Estimates
Under these assumptions, I regress Eq.(2.15) to generate loading factor estimates
across a variety of fixed effects combinations, which control for differences in the
scale of container flow operations at each port, local industry compositions in
each port’s surrounding area and potential biases in loading factors attributed
16To address these potential sources of bias, I have explored country-groupings for a given
commodity which potentially limits product differentiation, reducing the influence the differences
across quality within a product group may have on the estimation of an appropriate loading
factor. These geographic and income-based country groupings for specific commodity weights
have been evaluated in Appendix VI and generally contribute little towards improving loading
factor estimates.
35
Unconventional Protectionism
to the seasonality of within-commodity variation. To assess the importance of
composition differences in commodities by direction at HS2 level, I have estimated
both direction-invariant (joint) and f -specific (separate) loading factors.
Across both ‘joint’ and ‘separate’ loading factor exercises, I find that port
fixed effects are key in minimizing the number of negative coefficients that crop up
among the 97 HS2 products included. These negative coefficients would suggest
that, all else controlled for, the higher the weight of goods loaded into containers,
the lower the number of containers necessary to ship said goods. A rather salient
objective therefore is to use the specification which yields the most plausible set
of coefficient estimates.
Figure 2.6: Loading Factor Estimates by Commodity
0.4 0.75
0.3
0.50
0.2
0.1 0.25
0.0
0.00
-0.1
Clustered (port) standard-errors. Regresses monthly port-level total loaded container inflows (outflows) on a set
of commodity-specific weights of containerized US imports (exports), expressed in metric tons. Each coefficient
can be interpreted as the average loaded container unit volume occupied by a metric ton of commodity k. Re-
sults displayed for top 16 manufactured commodities by value. Observed total container levels and associated
containerized weights of goods are observed between Jan-2012 and Dec-2021 and use port & year-month fixed
effects. Point sizes vary based on share of associated trade flow.
These estimates are generally significant and positive in value.17 Combinations
of port, year and month fixed effects yield within R2 values ranging from 0.78-0.97
for imported goods and 0.59-0.98 for goods exports. Furniture, paper articles and
electrical machinery are found to be the most demanding commodities on incoming
container volumes. For example, a single metric ton of furniture is estimated to
17Appendix VI provides diagnostic tables which highlights that commodities associated with
negative loading factors are generally traded in particularly low volumes. Given that each exercise
estimates 97 commodity coefficients under varying combinations of fixed effects, it us unsurprising
that some false-positive findings of negative coefficients would populate overall results.
36
Import Coefficient Estimates
94 Furniture & Fixtures
48 Paper Articles
85 Electric Machinery
95 Toys & Games
73 Articles of Iron & Steel
39 Plastic Articles
84 Nuclear Reactors, Boilers
72 Iron & Steel
40 Rubber Articles
87 Vehicles & Parts
22 Beverages
44 Wooden Articles
69 Cermaic Products
68 Articles of Stone, Plaster, etc.
29 Organic Chemicals
25 Salt & Sulfur
Export Coefficient Estimates
84 Nuclear Reactors, Boilers
34 Soap
40 Rubber Articles
74 Articles of Copper
38 Misc Chemicals
48 Paper Articles
52 Cotton Fabric
87 Vehicles & Parts
76 Articles of Aluminium
44 Wooden Articles
39 Plastic Articles
72 Iron & Steel
29 Organic Chemicals
25 Salt & Sulfur
22 Beverages
28 Inorganic Chemicals
Unconventional Protectionism
take up one third of a container unit whereas a metric ton of iron & steel is
estimated to take up only a tenth of a container unit. US exports of nuclear
reactors, boilers, soap and rubber articles are estimated to be the most demanding
on container volumes whereas plastic articles, iron & steel occupy far less loaded
container unit volume.
Upon predicting port-level container flows & aggregating across US ports, I
compare these US estimates to observed national loaded container flows. I find
that predicted values using ‘separate’ loading factors are associated with lower
root mean square error values and higher correlation score compared to ‘joint’
estimates.18 I therefore focus attention on loading factor estimates specific to the
direction in which goods are flowing and generate country–level loaded container
flows,
∑P ∑P ∑K
Container Inflows: l̂ = l̂ Imp,kj−US,t jpt = β̂ wjpkt,
∑p=1 p=1 k=1P ∑P ∑K
Container Outflows: l̂ Exp,kUS−j,t = l̂pjt = β̂ wpjkt, (2.16)
p=1 p=1 k=1
where these bilateral volumes are determined by the product of commodity k’s con-
tainerized weight at time t and a corresponding time-invariant estimated loading
factor, βfk, summed across P ports and K commodities.
2.6.2.3 Container Flow Estimates
Estimates of loaded container flows are sensitive to the assumptions and meth-
ods used in identifying loading factors across commodities. While the ‘separate’
estimation of loading factors by direction yields far stronger results, the precise
set of fixed effects appears open to multiple combinations, so long as port fixed
effects are included. To determine which fixed effects yield the best match and
quantify differences in performance, I compare estimated volumes and bilateral
ratios of loaded container flows to UNCTAD records of annual loaded containers
exchanged on US-East Asian & US-European routes (UNCTAD, 2022).
As highlighted previously, product quality likely plays a role in determining
the container volume capacity required for the transport of a given metric ton of
a specific commodity. To address this concern, I have also estimated loading fac-
18See Table 5.10 in Appendix VI for further details.
37
Unconventional Protectionism
tor estimates for each commodity specific to groupings of countries by continental
boundaries and by income per-capita. Product quality may be correlated across
the imports and exports of countries that share close proximity with one another.
Similarly, countries of similar wealth levels may trade in goods of a comparable
quality levels. Country-groups’ commodity-specific loading factors, each repre-
senting a regressor, would introduce a far more substantial extent of sparseness
data used for estimation. I address this concern by removing any country-group
commodity specific-regressor if less than 40% of its monthly port-level weight flows
are reported as positive values. This introduces a trade-off between added preci-
sion for key, actively traded commodities across each country-group at the loss of
broader commodity representation upon aggregating across port data. Following
a series of container flow diagnostics – outlined in Appendix VI – I use port &
year fixed effects with no geographic or income-based groupings. These relatively
less restricted empirical approaches often introduced greater uncertainty in loading
factor estimates without any notable improvement in fits to untargeted UNCTAD
measures of two of the US’s busiest bilateral lanes of regional loaded container
flows. This fit is depicted in Figure 2.7.
Figure 2.7: Model Fit – Loaded Container Ratios by Region (2012-2021)
4
3
2
1
Correlation: 0.98
00 1 2 3 4
UNCTAD Loaded Inflow-Outflow Ratio
East Asia Europe
Note: Observed levels originate from UNCTAD records on regional total loaded container flows by year and were
untargeted in the estimation of individual country container flow estimates.
While the loading factors and resulting country-level container flow estimates
are available across a wide range of countries, I limit the use of these estimates to
the subset of countries that report manufacturing wage measures needed for model
calibration between 2012 to 2021. Additionally, I introduce balanced container
38
Estimated Loaded Inflow-Outflow Ratio
Unconventional Protectionism
flow system that incorporates the entire European Single Market and exclude
both Mexico and Canada due to land borders with the US potentially limiting the
degree to which bilateral flows of containerized trade are fully serviced by maritime
transport operators.19 Lastly, given that the model is calibrated on manufacturing
wages, I restrict container flow estimates to levels associated with the weight of
containerized manufactures travelling between the US and its respective trade
partners.20
Upon accounting for these product and multi-country constraints, I generate
loaded container flow estimates specifically for manufactured goods across the
countries featured in Figure 2.8. This limits my use of multi-country estimated
bilateral container flows to represent 70% (50%) of containerized import (export)
value.
Figure 2.8: Estimated Container Flows by Country and Direction
10,000,000
7,500,000
5,000,000
2,500,000
0
Loaded Inflows Loaded Outflows
Note: Compared total US container traffic in 2017 across my 12 sampled ports, these disaggregated estimates
of manufactured goods flows across the choice subset of trade partners represents 70% (50%) of containerized
goods imports (exports) and 65% (43%) of loaded container inflows (outflows). The country group of ”other”
represents 10 additional lanes of round trip container traffic with the US – consisting of Argentina, Australia,
Chile, Columbia, Dominican Rep., Ecuador, Malaysia, the Philippines, Singapore and Turkey.
19See Appendix VII for evidence of balanced container flows only at the Single Market level.
20The contributing commodities are those featured in the TRAINS SITC-based product group
known as ‘Manufactures’. I use the United Nations Statistics Divisions’ correspondence tables,
HS - SITC/BEC, to convert SITC 4 codes belong to the manufactures product group on TRAINS
into a set of relevant HS 2017 codes. https://unstats.un.org/unsd/classifications/Econ Last
accessed as of March 17th 2023.
39
Total Containers
China
European CU
Other
Japan
Korea, South
Taiwan
Unconventional Protectionism
2.6.3 Solution Method and Model Calibration
To establish a baseline set of exogenous parameters, I first calibrate a select subset
of model primitives and then estimate the remaining set of unknown model prim-
itives using a Generalized Method of Moments (GMM) approach. For a given ij
round trip contain(erized shipping route, the s)et of unknown exogenous parame-
ters ρ is equal to aij, aji, wi, wj, τij, τji, c↔, r↔ and the elasticity of substitution
ij ij
measure is represented by ϵ.
For wages, I use an OECD index of monthly manufacturing income growth
rates and the International Labor Organization (ILO) annual measure of monthly
manufacturing income levels, which are available for a subset of trade partners.
For tariffs, I use the UNCTAD Trade Analysis Information System (TRAINS)
database on effective manufactured goods’ tariff rates, all of which are reported
across US trade partners.21 I deflate the value of trade flows and USD-converted
wage levels for each trade partner using the Bureau of Labor Statistics Consumer
Price Index for all urban consumers, which considers all final good items less food
and energy, averaged across major US cities.22 Lastly, I include an estimate of
price elasticity of demand provided by Wong (2022) and specific to containerized
trade, where ϵ̂ = 20.96 is assumed to be common across individual trade routes.
Using these calibrated parameters and a vector of country-level endogenous
trade outcomes, represented by Y data = {X
( ) ij
, Xji, l̂ij, l̂ji}, I estimate the remaining
set of unobserved preference parameters and route-specific per unit handling costs
of containers, ρ̃ = aij, aji, c↔, r↔ , via GMM.
23 I minimize the object function,
ij ij
R = dist′ × W̄ × dist, (2.17)
where dist represents the log difference in vectors of ‘observed’ and model-guess
trade outcomes between the US and a given trade partner, log(Y data) − log(Y G),
and W̄ is a weight matrix that assists in speeding the identification of ρ̃. I use
measures from 2017 to estimate these parameters of underlying long-run primitives
of containerized trade. This specific year allows me to avoid any complications or
21Upon establishing a login for http://wits.worldbank.org/, select ‘Advanced Query’ and then
the ’Tariff and Trade Analysis’ subsection. I use the SITC 4 product group labelled ‘manufac-
tures’ and the effective tariff rate measure.
22U.S. Bureau of Labor Statistics, Consumer Price Index for All Urban Consumers: All Items
Less Food and Energy in U.S. City Average [CPILFESL], retrieved from FRED, Federal Reserve
Bank of St. Louis; https://fred.stlouisfed.org/series/CPILFESL, November 1st, 2022.
23The respective outcome variables used are observed average monthly containerized imports
& exports (USD value) and estimated loaded container inflows and outflows.
40
Unconventional Protectionism
concerns that the use of data from the proceeding China-US trade war or period of
COVID-related port congestion could introduce. Given that for each round trip,
I estimate four unknowns across a system of four equations, my model is just-
identified and I exactly match the observed trade values and estimated loaded
container flows.
To assess the performance of this exercise on untargetted features and mo-
ments in the data, I provide three means of assessing model fit for this baseline
scenario; (1) the empty container redistribution share of container fleet manage-
ment costs averages between 14.9–21.3%, depending on the given year, which places
US-related costs of empty container redistribution relatively close to 15% share re-
ported by Rodrigue (2020); (2) the difference in pairs of preference parameters on
round trip routes attributes stronger tastes on the larger volume importing lane,
with ratios of tastes being highly predictive of the skewness prevailing in trade
imbalances; (3) using marginal costs of handling loaded, c↔, and empty container
ij
flows, r↔, implied freight rates are greater for portions of US round trips that fea-
ij
ture a full set of loaded containers, which is reflective of empirically documented
freight rate asymmetries under imbalanced trade (Hummels et al., 2009).
In order to address the importance of specifying country-specific container
flows, I also prepare a US-RoW calibration and estimation of exogenous model
primitives using these same inputs. This second baseline scenario, which represents
trade through a single round trip, under-represents empty container redisitribu-
tion and effectively spreads the reliance on the return of this transport equipment
from the US across all trade partners. By introducing both the US-RoW and
multi-country baseline scenarios to the same counterfactual change, I quantify
the importance of accounting for variation in extensive and intensive margins of
dependencies on empty container redistribution from the US.
2.6.4 Counterfactual Policy Background
In this subsection, I discuss recent changes to liner shipping regulation through
the Ocean Shipping Reform Act of 2022 (OSRA22), a portion of which aims to
limit empty container redistribution in favor of stimulating greater US exports.
To examine the consequences of restricting empty container outflows, I outline a
simplified version of this policy in which the policymaker has capped the share of
empty container outflows relative to total outflows from the US through a per-unit
tax rate.
41
Unconventional Protectionism
2.6.4.1 Pre-policy Conditions
Between October 2021 and November 2022, vulnerabilities in US transport services
became notably tangible. A resurgence of economic activity in the US contributed
to elevated import demand, which resulted in a widening of the US trade deficit.
The associated increase in the asymmetry of bilateral containerized trade volumes
coincided with record-high rates of empty container outflows. For example, ac-
cording to container traffic levels provided by the Port Authority of Los Angeles,
the percentage of empties featured on container outflows originating from LA in-
creased from a pre-COVID historical average of 50 percent to over 80 percent in
the latter half of 2021. As of 2022, for every five containers that entered the US
laden with goods, three of these containers leave the US empty.
These signs of elevated empty redistribution are the result of a sudden shift in
market conditions. For example, if US demand for Chinese manufactured goods
suddenly increased, a greater number of loaded container units would be trans-
ported to the US from China. Upon redistributing containers back to China –
to service subsequent Chinese exports – the percentage of empties featured on
outbound voyages from the US would rise. Log-jams of vessels and transport
equipment also made empty repositioning relatively more appealing. They require
less handling due to less time spent transporting goods within a given destination
country’s hinterland area, are readily usable upon arrival at a destination port and
relatively cheaper to transport due to their lower weight. These factors suggest
that in certain cases, it may be more profitable for a firm to transport an empty
container unit rather than service an additional loaded container unit that cannot
be repurposed as quickly.
These opportunity costs and existing differences in import demand between two
regions determine the scale of the empty container redistribution problem. Due
to the relatively higher opportunity costs of servicing loaded container units and
the increased volume of import traffic to the US, a greater percentage of shipping
capacity was reassigned to service empty container transport. However, short-run
adjustments to a new empty-loaded outflow equilibrium and the increased difficulty
for exporters in securing vessel allocated space contributed to a swift bipartisan
response from US policymakers.
42
Unconventional Protectionism
2.6.4.2 Ocean Shipping Reform Act 2022
In December 2021, the House of Representatives passed H.R.4996, the Ocean Ship-
ping Reform Act of 2021. This bipartisan bill sought to empower the Federal
Maritime Commission (FMC) by introducing legislation that prohibits the ‘unrea-
sonable’ refusal of vessel capacity from US exports. The stated intention of this
bill is to ensure fair trade by supporting good-paying American manufacturing
jobs and agricultural exports. Upon passing this proposed legislation on to the
Senate, lawmakers were explicit in further emphasizing the intent of this bill.
“The rulemaking under paragraph (1)24 shall address the unreasonableness
of ocean common carriers prioritizing the shipment of empty contain-
ers while excluding, limiting, or otherwise reducing the shipment of full,
loaded containers when such containers are readily available to be shipped
and the appurtenant vessel has the weight and space capacity available to
carry such containers if loaded in a safe and timely manner.”
H.R.4996, the Ocean Shipping Reform Act of 2021
In February 2022, the Senate passed OSRA22, which maintained this prohibi-
tion. This bill has since entered into public law as of June 16th 2022. However, the
bill did not specify how this restriction on prioritizing empties must be imposed
and instead delegated this task to the Federal Maritime Commission (FMC). The
first challenge for the FMC involves defining cases of ‘unreasonable refusals of ves-
sel capacity’ and then it must devise measures by which to punish any violators.
The FMC has since issued a Notice of Proposed Rulemaking (NPRM), which has
suggested that ‘unreasonable’ refusals must be determined on a case-by-case basis
(FMC, 2022). To judge reasonability, the FMC would require that ocean common
carrier provide a documented export strategy that enables the efficient movement
of export cargo.25
In response, the World Shipping Council (WSC), an association that represents
90% of transport operators, has clarified some of the operational and commercial
24This relates to Section 9 of the proposed bill, Prohibition on Unreasonably Declining Cargo,
where transport operators are warned against “engaging in practices that unreasonably reduce
shipper accessibility to equipment necessary for the loading or unloading of cargo”.
25No connection is provided in the NPRM between an “export strategy” document require-
ment and how this establishes a definition of how a transport operator may unreasonably refuse
to negotiate or deal with respect to vessel space accommodations. This has led to a second round
of public discourse by the FMC and an extension to these deliberations.
43
Unconventional Protectionism
realities that contribute to empty repositioning. A static export strategy is sug-
gested to not align with the business practices of the industry, which is “volatile
with rapidly changing factors that impact space availability on a daily basis.” Most
notably, the WSC goes on to highlight that “export trades cannot be considered
in isolation from import trades”. This important facet of containerized shipping
acts as the cornerstone of my container redistribution model.
Carriers use the same containers, ships, and marine terminals to handle
both import and export containers, and vessels operate on continuous loops,
not distinct import and export legs disconnected from one another. Addition-
ally, the proposed regulatory language does not address in any way the basic
reality that imbalanced trades (as reflected on in the preamble) require the
repositioning of equipment, which adds an additional dimension to planning
and operating vessel networks. It defies the reality of ocean transportation to
ignore these complexities and to treat the export and import legs of a trade
as unrelated.
World Shipping Council response to FMC (2022)
While the FMC continues to deliberate over these key details, I propose an
exercise which embodies policymakers’ intent of limiting empty redistribution in
favor of greater capacity allocation towards US exporters. To capture the potential
effects of this unconventional policy approach, I introduce a per-unit tax on empty
container outflows to the baseline model, where the tax rate is calibrated to target
a capped share of empties as a percentage of total container outflows. I consider a
restriction to transport equipment use by the US policymaker, where the expressed
goal is to return empty activity back to its historical share of 40% of total container
outflows.
To establish this counterfactual scenario, the US policymaker sets a per-unit
empty tax rate of γ on the outbound channel of round trip transport, which targets
the historical average of empty container share of container gross outflows, Ēji =
0.4. This moderate ECO quota scenario represents a case in which policymakers
are content with the prior long run average of the empty container redistribution
problem.26 Using the same tax rate, γĒ =0.4, on the US-Row version of the baselineji
26I have also examined an ‘extreme’ ECO quota, in which γĒ =0 is targeted and the practiceji
of empty container redistribution is eliminated. Similarly to the main results described in the
next section, I find that the policy contributes to a decline in vessel capacity on net exporter
trade routes and reduction in overall trade value and volume.
44
Unconventional Protectionism
scenario, I highlight the benefits of accounting for heterogeneous dependencies on
empty container redistribution from the US.
2.6.5 Main Results
As displayed Table 2.7, a moderate ECO quota stimulates export activity. US
exporters flock to relatively cheaper freight rates for round trip services to net
exporter countries, which results in a substitution from empty container redis-
tribution to additional loaded container servicing. The US containerized trade
deficit, represented by the import-export ratio, also declines by 37.3%. However, a
focus only on this outbound leg of US round trip transport ignores further market
developments, known as round trip effects, which may also be of interest to the
policymaker. Relative to the baseline scenario, a multi-country model of US con-
tainerized trade sees a 17.7% decline in the real value of imports. This is attributed
to the greater cost associated with returning the empties, which passes through
entirely to the price of US imports under this perfectly competitive setting. As
a result, the price of imported goods rises by 1.7% while US exporters see their
goods’ prices decline by 4.3%. The overall capacity of TEU services for round
trips between the US and individual countries declines by 18.6% due to policy
introducing an added friction servicing imbalanced volumes of trade. This leads
to a reduction in container redistribution. The scale of the empty container redis-
tribution problem as a percentage of total US container outflows falls by 37.4%.
Table 2.7: Disaggregated Counterfactual Outcomes
U.S. Measures Imports Exports Imp. Price Exp. Price Value Vol. Capacity
∆% -17.7 31.1 1.7 -4.3 -8.5 -4.4 -18.7
Note: These results reflect percentage changes from their respective 2017 baseline scenarios of
the partial equilibrium model and are based on estimates of loaded container flows & observed
levels of associated trade in containerized manufactured goods.
While adjustments in individual flow measures and the trade balance are of
interest, understanding changes to the scale of overall trade activity is of the
greatest importance in this setting. Should overall trade activity decline, so too
would the associated gains from trade. In the case of the multi-country setup, a
moderate ECO quota contributes to an 8.5% (4.4%) decline in the value (volume)
of containerized trade, which suggests a degradation in the gains to trade the US
and its trade partners would have otherwise been able to accrue.
45
Unconventional Protectionism
Across the subset of net exporters which engage in containerized trade with
the US, the pre-existing scale of the empty container repositioning problem acts
as a strong predictor of this policy’s effectiveness. Using US outflows of empties to
country i as a percentage of total US container outflows to country i, I find that
countries particularly reliant on empty repositioning yielded the largest declines in
imports. East Asian trade partners maintained the high empty container shares
in the predefined baseline scenario (Figure 2.9). Upon the introduction of a per-
unit tax on empty repositioning, these particularly asymmetric trade routes faced
the greatest contractionary pressure. Transport operators servicing these routes
respond by introducing larger contractions in vessel capacity, which in turn lowers
the value and volume of imports shipped from East Asia to the US. The greater
each country’s intensive margin of reliance on empty containers, represented by
the empty share term, the greater the decline in import levels.
Figure 2.9: US Import Value by Net Exporter (2017)
-8
Turkey Europe
-12
Malaysia
Korea, South
-16
Taiwan
Japan
-20
China
25 50 75
Empty Share of Containers Inbound from US
Note: The real value of imports is used, deflated by US CPI for urban areas, less food and energy. The empty
share represents 100 ∗ US-Country Empty Outflows , and reflects pre-policy shares of total container outflows.
Total US-Country Outflows
Given that the repositioning of empties has become more expensive, the un-
derlying costs of loaded container services are relatively more appealing. This is
reflected by a decline in the US-net exporter freight rate and a substitution into
increased US export activity across net exporter round trip trade routes. Countries
such as China and Japan yield greater changes due to their particularly significant
reliance on empty containers and greater declines in export prices (Figure 2.10).
46
% Change, Imports (Real Value)
Unconventional Protectionism
Figure 2.10: US Export Value by Net Exporter (2017)
China
Japan
60
50
Taiwan
40
Korea, South
30
Malaysia
20 Europe
Turkey
10
25 50 75
Empty Share of Containers Inbound from US
Note: The real value of exports is used, deflated by US CPI for urban areas, less food and energy. The empty
share represents 100 ∗ US-Country Empty Outflows , and reflects pre-policy shares of total container outflows.
Total US-Country Outflows
Lastly, the inflationary pressure generated by a tax on empty container units
appears to have particularly pronounced effects on endogenous import prices across
the net exporters that exhibit a greater reliance on empties. As displayed in Figure
2.11, Turkey and Europe yield relatively low pass-through of this new tax burden
on prevailing market prices. However, East Asia yet again yields evidence of greater
exposure to this form of protectionism, in which percentage point increases in price
levels are almost threefold larger.
The sizable loss in transport equipment accessibility and the acuteness of this
decline on routes with a particularly high dependencies on empty repositioning
leads to noteworthy changes in country shares of the US import market. As dis-
played in Figure 2.12, in some cases net exporters gain market shares despite being
reliant on empty container repositioning. China, which receives approximate four
empty returns for every five loaded containers shipped to the US, suffers a two
percentage point loss in its share of containerized US imports. Given Europe’s
relatively weaker dependency on empty container repositioning, although imports
do decline, the overall decline in total US containerized imports of manufactures
falls by a greater margin. This results in the European Custom’s Area developing
a larger share of overall US imports, despite being negatively affected by an ECO
quota.
47
% Change, Exports (Real Value)
Unconventional Protectionism
Figure 2.11: US Import Price Inflation by Net Exporter (2017)
China
1.25
Japan
Taiwan
1.00
Korea, South
Malaysia
0.75
0.50 Turkey Europe
25 50 75
Empty Share of Containers Inbound from US
Note: Real prices are reflective of the average value of containerized goods per loaded TEU. Deflated using US
CPI for urban areas, less food and energy. The empty share represents 100 ∗ US-Country Empty Outflows .
Total US-Country Outflows
Figure 2.12: Change in Trade Partner Shares of US imports (2017)
2
1
0
-1
-2
Note: Real values of imports are deflated using US CPI for urban areas, less food and energy.
2.7 Conclusion
This paper provides a quantitative approach towards understanding the novelties
of containerized trade and its reliance on the repositioning of physical transport
equipment. The first contribution of this study identifies how variation in the avail-
48
Percentage Point
% Change, Import Prices
Europe
Philippines
Korea, South
Singapore
Australia
Chile
Malaysia
Turkey
Argentina
Dom. Rep.
Colombia
Ecuador
Taiwan
Japan
China
Unconventional Protectionism
ability of transport equipment may feed into trade outcomes on the opposite leg
of a given round trip, and enriches means of incorporating endogenous transport
costs. In this particular case, I internalize the cost of repositioning container units
to associated transport operators and highlight how variation in such costs may
result in adjustments to the available transport capacity devoted to a particular
origin-destination pair. Using novel container traffic data provided by the largest
ports in the US, representative of 80% of gross container unit traffic, I directly
connect theory and empirics. Through this connection, I document a round trip
effect taking place in which adjustments in the prevailing trade balance of the US,
through larger trade deficits, enlarges the scale of the empty container reposition-
ing problem. Through supportive evidence, I argue that it is opposite-leg trade
outcomes that drives variation in the empty container repositioning problem of
the US.
I also contribute theoretically to the literatures of international trade and trans-
port economics through my partial equilibrium model of container repositioning.
This model that yields positive bilateral freight rates under a setting of perfectly
competitive transport operators with perfect knowledge, which as highlighted by
Demirel et al. (2010), normally introduces unintuitive and troublesome model pre-
dictions. By representing container units physically in the joint profit maximiza-
tion problem of transport operators, I circumvent a persistent challenge in mod-
elling imbalance round trip trade in which the lower volume leg of a given route
yields a freight rate of zero. Additionally, this challenge is not unique to maritime
commerce and can be considered applicable across multiple modes of transport.
In future work, it would be of interest to understand how this phenomenon inter-
acts with recent developments in market concentration across the global fleet of
transport operators. For example, does a greater extent of coordination through
cooperative shipping alliances across the global fleet limit container shortages?
Lastly, I quantitatively evaluate how interfering in the use of this transport
technology can affect trade flows. Although studies of trade conventionally con-
sider protectionism to occur through adjustments to tariff rates, goods quotas,
and other means of applying non-tariff measures, little is understood of how pol-
icymakers’ targeting of transport equipment could influence trade outcomes. I
highlight how a modern and unconventional form of protectionism may backfire
and negatively affect broader public. This specific form of policy is motivated
by the recently passed Ocean Shipping Reform Act of 2022 (OSRA22), in which
restrictions to empty container outflow activities were introduced in an effort to
49
Unconventional Protectionism
stimulate US exports. My findings suggest that government intervention in the
repositioning of empty container units may lead to unanticipated and adverse ef-
fects, in which overall vessel capacity servicing the US reduces due to the relatively
greater expense associated with servicing trade imbalances. Within trade lanes,
exports grow, but this minor boon are outscaled by a reduction in import activity
and increased price inflation for US consumers. Great care should be taken in
considering the joint-effects of liner shipping regulation, rather than focusing on
an export lane of round trip traffic in isolation. To quote the World Shipping
Council’s response to OSRA22, “It defies the reality of ocean transportation to
ignore these complexities and to treat the export and import legs of a trade as
unrelated.”
These considerations are important to consider, not only in maritime ship-
ping, but across trucking, rail and airline services. Each of these forms of round
trip service accommodate differences in trade volume. As I highlight, particu-
larly asymmetric round trip volumes are the most subject to malaise effects, given
the introduction of empty container repositioning regulation. Should we wish to
fully embrace trade flows, irrespective of differences in bilateral flows, this requires
particularly low costs of handling empty containers. According to these compara-
tive statics, developments in foldable container technology would dramatically cut
required dock space for storage and further expand global trade opportunities.
Going forward, I believe this study adds emphasis toward more granular data
on port traffic and container shipping details. I welcome the data provision require-
ments introduced through OSRA22, which enables the Federal Maritime Commis-
sion to publish a quarterly report detailing the total import and export tonnage
and the total loaded and empty 20-foot equivalent units per vessel operated by
ocean common carriers. Upcoming container tracking technology would be of great
interest and enable studies such as this paper to directly connect port container
traffic through the US hinterland. As my estimation country-specific container
flows may indicate, further studies of maritime transport would also be enhanced
by a greater knowledge of container origins, routes travelled upon and ultimate
destinations. These improvements in data availability would enhance the iden-
tification of key transport bottlenecks, allow for the accounting of transshipping
activity and better our understanding of countries’ joint dependency on efficient
transport equipment usage.
50
Unconventional Protectionism
Bridge
In this essay, I isolated the ‘backhaul problem’ faced across multiple transport
sectors, in which empty equipment needs to be returned to a location of relatively
greater transport demand. While asymmetries in trade are one means of influ-
encing the formation of endogenous transport costs, one must not only consider
the container but also the vessel carrying these containers. In the next substan-
tive essay of my dissertation, I explore the evolution of containership services in
the US, documenting patterns associated with individual US port visits between
January 1977 and December 2023. Underlying container handling costs shaped
asymmetries in my first substantive essay. In this next piece, the efficiency of
handling containers will be influenced both by the size of incoming vessels and the
surrounding port conditions they are exposed to.
Bibliography
Allen, T. and C. Arkolakis (2022). The Welfare Effects of Transportation Infras-
tructure Improvements. Review of Economic Studies 89 (6), 2911–2957.
Ardelean, A. T., V. Lugovskyy, A. Skiba, and D. M. Terner (2022, April). Fathom-
ing Shipping Costs : An Exploration of Recent Literature, Data, and Patterns.
Policy Research Working Paper Series 9992, The World Bank.
Armington, P. S. (1969). A Theory of Demand for Products Distinguished by
Place of Production. IMF Staff Papers 16 (1), 159–178.
Atkin, D. and D. Donaldson (2015). Who’s Getting Globalized? The Size and Im-
plications of Intra-national Trade Costs. NBERWorking Papers 21439, National
Bureau of Economic Research, Inc.
Behrens, K. and P. Picard (2011). Transportation, freight rates, and economic
geography. Journal of International Economics 85 (2), 280–291.
Bernhofen, D., Z. El-Sahli, and R. Kneller (2016). Estimating the effects of the
container revolution on world trade. Journal of International Economics 98 (C),
36–50.
Blonigen, B. and W. Wilson (2008). Port Efficiency and Trade Flows. Review of
International Economics 16 (1), 21–36.
51
Unconventional Protectionism
Bombardini, M., B. Li, and F. Trebbi (2023). Did US Politicians Expect the China
Shock? American Economic Review 113 (1), 174–209.
Bonadio, B. (2022). Ports vs. Roads: Infrastructure, Market Access and Regional
Outcomes. Technical report, Working Paper.
Bown, C. P. (2021). The US–China Trade War and Phase One Agreement. Journal
of Policy Modeling 43 (4), 805–843.
Brancaccio, G., M. Kalouptsidi, and T. Papageorgiou (2020). Geography, trans-
portation, and endogenous trade costs. Econometrica 88 (2), 657–691.
Brooks, L., N. Gendron-Carrier, and G. Rua (2021). The local impact of con-
tainerization. Journal of Urban Economics 126 (C), S009411902100070X.
Carreras-Valle, M.-J. (2022). Increasing Inventories: The Role of Delivery Times.
Technical report, Working Paper.
Cooks, C. and P. Li (2023). Value Pricing or Lexus Lanes? Winners and Losers
from Dynamic Toll Pricing. Job Market 2024 Paper, Institute for Economic
Policy Research (SIEPR), Stanford University, Work in Progress.
Crainic, T. G., M. Gendreau, and P. Dejax (1993). Dynamic and Stochastic Models
for the Allocation of Empty Containers. Operations Research 41 (1), 102–126.
Demirel, E., J. van Ommeren, and P. Rietveld (2010). A matching model for
the backhaul problem. Transportation Research Part B: Methodological 44 (4),
549–561.
Drewry (2006). Annual Container Market Review & Forecast 2006/07. Technical
report, Drewry Shipping Consultants Ltd: London.
Ducruet, C., S. Cuyala, and A. El Hosni (2018). Maritime networks as systems of
cities: The long-term interdependencies between global shipping flows and urban
development (1890–2010). Journal of Transport Geography 66 (C), 340–355.
Fajgelbaum, P. D., P. Goldberg, P. J. Kennedy, and A. Khandelwal (2020). The
Return to Protectionism. The Quarterly Journal of Economics 135 (1), 1–55.
Fajgelbaum, P. D. and A. K. Khandelwal (2022). The economic impacts of the
us–china trade war. Annual Review of Economics 14, 205–228.
52
Unconventional Protectionism
Flaaen, A., F. Haberkorn, L. Lewis, A. Monken, J. Pierce, R. Rhodes, and M. Yi
(2021). Bill of Lading Data in International Trade Research with an Application
to the COVID-19 Pandemic. Finance and Economics Discussion Series 2021-066,
Board of Governors of the Federal Reserve System (U.S.).
FMC (2022). Definition of Unreasonable Refusal To Deal or Negotiate With Re-
spect to Vessel Space Accommodations Provided by an Ocean Common Carrier.
Federal Register 87 (182), 57674–57679.
Ganapati, S., W. F. Wong, and O. Ziv (2021). Entrepôt: Hubs, Scale, and Trade
Costs. NBER Working Papers 29015, National Bureau of Economic Research,
Inc.
Grossman, G. M. and E. Helpman (2021). Identity politics and trade policy. Review
of Economic Studies 88 (3), 1101–1126.
Hayakawa, K., J. Ishikawa, and N. Tarui (2020). What goes around comes around:
Export-enhancing effects of import-tariff reductions. Journal of International
Economics 126 (C), S0022199620300787.
Heiland, I., A. Moxnes, K. H. Ulltviet-Moe, and Y. Zi (2022). Trade From Space:
Shipping Networks and The Global Implications of Local Shocks. (Work in
Progress).
Holmes, T. J. and E. Singer (2018). Indivisibilities in Distribution. Working Paper
24525, National Bureau of Economic Research.
Hummels, D. (2007). Transportation Costs and International Trade in the Second
Era of Globalization. Journal of Economic Perspectives 21 (3), 131–154.
Hummels, D., V. Lugovskyy, and A. Skiba (2009). The Trade Reducing Ef-
fects of Market Power in International Shipping. Journal of Development Eco-
nomics 89 (1), 84–97.
Ignatenko, A. (2023). Price Discrimination and Competition in International
Transportation. Working paper, Norwegian School of Economics.
Irarrazabal, A., A. Moxnes, and L. D. Opromolla (2015). The Tip of the Iceberg: A
Quantitative Framework for Estimating Trade Costs. The Review of Economics
and Statistics 97 (4), 777–792.
53
Unconventional Protectionism
Ishikawa, J. and N. Tarui (2018). Backfiring with Backhaul Problems. Journal of
International Economics 111 (C), 81–98.
Juhász, R. (2018). Temporary Protection and Technology Adoption: Evidence
from the Napoleonic Blockade. American Economic Review 108 (11), 3339–76.
Lee, C.-Y. and D. Song (2017). Ocean container transport in global supply
chains: Overview and research opportunities. Transportation Research Part
B-methodological 95, 442–474.
Levinson, M. (2016). The Box: How the Shipping Container Made the World
Smaller and the World Economy Bigger (REV - Revised, 2 ed.). Princeton
University Press.
Notteboom, T., A. Pallis, and J.-P. Rodrigue (2022, January). Port Economics,
Management and Policy. Routledge.
Rodrigue, J.-P. (2020, May). The Geography of Transport Systems. Fifth edition.
— Abingdon, Oxon ; New York, NY : Routledge, 2020.: Routledge.
Sampson, T. (2017). Brexit: The Economics of International Disintegration. Jour-
nal of Economic perspectives 31 (4), 163–84.
Samuelson, P. A. (1952). The Transfer Problem and Transport Costs: The Terms
of Trade When Impediments are Absent. The Economic Journal 62 (246), 278–
304.
Song, D. (2021). A Literature Review, Container Shipping Supply Chain: Planning
Problems and Research Opportunities. Logistics 5 (2), 1–26.
Song, D.-P. (2007). Characterizing optimal empty container reposition policy in
periodic-review shuttle service systems. Journal of the Operational Research
Society 58 (1), 122–133.
Song, D.-P. and J.-X. Dong (2015). Empty Container Repositioning. Chapter 6,
pp. 163–208. Springer.
Steinbach, S. (2022). Port congestion, container shortages, and U.S. foreign trade.
Economics Letters 213, 110392.
UNCTAD (2022). Review of Maritime Transport 2022, Navigating Stormy Waters.
United Nations Publications.
54
Unconventional Protectionism
Williamson, O. E. (1966). Peak-load pricing and optimal capacity under indivisi-
bility constraints. The American Economic Review 56 (4), 810–827.
Wong, W. F. (2022). The Round Trip Effect: Endogenous Transport Costs and
International Trade. American Economic Journal: Applied Economics 14 (4),
127–66.
Wong, W. F. and S. Fuchs (2022). Multimodal Transport Networks. FRB Atlanta
Working Paper 2022-13, Federal Reserve Bank of Atlanta.
55
Chapter 3
Container Ports
56
Container Ports
Philip Economides∗, Simon Fuchs† and Woan Foong Wong∗
April 29th, 2023
Abstract
The majority of international trade is transported by ocean-going
ships, where ports serve as important interchanges facilitating the flow
of these goods. We study the long run interaction between contain-
erships and top US ports over the past 47 years by compiling and
combining three data sets. We establish that while the number of ship
visits have more than tripled and the average ship size quadrupled,
the average amount of time a ship dwells at a port has stayed roughly
the same at about 2.5 days. Ongoing work includes developing a shift-
share instrument to identify the impact of port traffic on ship dwell
times and a model of capacity reallocation at ports, in order to quan-
titatively study the impact of changing shipping technologies on port
traffic and wait times.
JEL Codes: D62, F18, F64, L51, Q52, Q53, R41
Keywords: transport externalities, port logistics, vessel queues, congestion, mar-
itime trade.
∗Department of Economics, University of Oregon.
†Federal Reserve Bank of Atlanta.
Contact: economi@uoregon.edu, simon.fuchs@atl.frb.org, and wfwong@uoregon.edu. The views
in this paper are solely the responsibility of the authors and should not necessarily be interpreted
as reflecting the views of the Board of Governors of the Federal Reserve System or of any other
person associated with the Federal Reserve System. All errors are our own. We thank the
NBER/NSF/DOT for project support.
Container Ports
The essay below originated from a project I contributed to as a research assis-
tant for Professor Woan Foong Wong. While Professor Wong and our co-author,
Dr Simon Fuchs, suggested the topic of the paper and provided initial points of
reference in the literature, I have refined the research questions addressed in this
essay and developed supporting geo-spatial algorithms used to clean and merge
three separate large datasets on vessel positioning at ports in order to conduct the
analyses in this essay. Following consistent meetings with my co-authors, I have
also been supported in compiling an extensive review of existing data sources on
maritime trade and transportation that is suitable for the analysis in this essay,
which has allowed us to pursue access to these data sets. Referring to the out-
lined specifications I discussed with my co-authors, I have prepared each of the
appropriate estimations in the empirical section of this essay and implemented a
preliminary methodology for applying an instrumental variable of interest. Going
forward, I expect further changes in future iterations of this study, attributable
to my co-authors and I. These changes include further refinement of our empiri-
cal approach, the introduction of a theory model, combining our empirical results
with counterfactual analysis to examine the influence of port infrastructure adjust-
ments and technological innovations in vessel technology on US trade outcomes.
While both Professor Wong and Dr Fuchs have contributed to various parts of
this project, I have contributed strongly to the progress of this research. From the
data collection, cleaning, initial analysis, up to the current draft, I have worked
directly on each of the sections of this study.
3.1 Introduction
The majority of international trade is transported via ocean on ships. This places
a great deal of importance on the state of port facilities, both with respect to each
port’s scale of operations and efficiency in servicing vessels. Over recent years,
increasingly larger and larger containerships are built to take advantage of scale
economies in transportation (Cullinane and Khanna, 2000; Tran and Haasis, 2015).
A concerted effort to pursue these benefits has contributed to an effective arms
race across the merchant fleet, in which container ship capacities have increased
nearly six-fold. As a result, ports and terminals – seeking to remain competitive
destinations for waterborne commerce – are incentivized to make “large and rapid
investments in infrastructures” to meet the needs of larger vessel sizes (Imai et al.,
2006; Parola et al., 2017). While these large ships carry significantly more vol-
58
Container Ports
ume, vessels are also consequentially spending longer times at ports unloading and
loading. These longer ship dwell times, all else equal, introduce increased negative
externalities for other vessels using the same port. However, larger ships does also
translate into fewer port visits overall, since these ships are able to transport much
more goods each visit.
Using historical records of ship tracking and port call data, we capture the
long-run evolution of ship visits and dwell times at US ports and examine the
factors that contribute to ship dwell times. We compiled and combined datasets
from four different sources in order to put together 47 years of panel data on
container port visits across the US. From 1977-2002, we use private data from
Lloyds List Intelligence, which at the time employed numerous agents across the
world to manually track vessel arrivals and departures from ports through on-site
observations and paper-based port records. From 2002-2016, we obtained port
entrance and clearance records from the US Army Corps of Engineers. From
2016-2023, we obtained automatic identification system (AIS) data for ships in US
waters which captures ship movements down to the minute-level. For each of these
datasets, we observe the unique vessel identifiers for each ship which we merge to
their characteristics including containership capacity (in twenty-foot equivalent
units, TEUs) and year they were built using a fourth dataset.
Although port operations are key to securing access to foreign goods for many
countries, there is a dearth of port performance measures. To address this need,
experts in maritime transport and international trade have sought to estimate port
efficiency. Both survey responses (Clark et al., 2004) and estimates using fixed ef-
fects models using variation in handling costs (Blonigen and Wilson, 2008; Ducruet
et al., 2020) have been used to yield port efficiency measures. We provide a rela-
tively more disaggregated measure of performance, which allows for the assessment
of individual berth locations within port facilities at minute-level intervals.
Furthermore, this growing volume of maritime transport services has incen-
tivized the construction of larger containerships and development of complex ship-
ping networks that exploit the increasing returns of scale that larger vessels pro-
vide. Ganapati et al. (2021) documents a hub-and-spoke system of global trade,
finding that 80% of trade is shipped indirectly via entrepôts – major hubs that
facilitate trade between many origins and destinations. In particular, routes with
lower per-unit trade costs feature larger ships. A 10% decrease in estimated ice-
berg trade costs is associated with 6% larger ship sizes. Carballo et al. (2023) finds
evidence that reduced processing time for shipments at the border contributes sig-
59
Container Ports
nificantly to elevated trade outcomes. While larger vessels are associated with
efficiency through lower costs, we ask whether the increase in size could also con-
tribute to faster rates of TEU handling. Our study focuses on port-level container
handling and vessel berth use across US ports, allowing us to incorporate poten-
tial negative spillovers of prevailing port congestion on surrounding vessel services.
By identifying dwell events at the individual vessel level, we are able to track the
adoption of larger containership classes over time and identify granular adjust-
ments in vessel services of specific berth locations within ports. This also allows
us to decompose how adjustments in technology and congestion contribute to the
efficient use of port resources.
Our results show that there have been large increases in the amount of contain-
ership traffic at US ports: the total number of containership visits has increased
by about three times since 1977 while the total TEU capacity of these ships has
increased sevenfold. Second, the average size of these containerships has increased
dramatically as well: average TEU capacity of containerships have increased by
about 5 times, from averages of approximately 1,300 TEUs in 1977 to almost 8,000
TEUs by 2023. Third, these large containerships have been making less port visits:
the average number of annual port visits per containership have halved from 24 in
the 1970’s to 12 in recent years. At the same time, the average amount of time a
ship spends at port has stayed steady at around 2.4 days (with large spikes during
strike times and during the pandemic period). We categorized vessels using their
time invariant characteristics of year built, length, container capacity and gross
tonnage to document increasing container handling efficiencies as vessels scale up
in size. Our analysis suggests that a one percent increase port traffic contributes
to a 0.09–0.17 percent increase in vessel dwell times per TEU, suggesting that
elevating congestion at port leads to negative spillovers for surrounding vessels.
The rest of this chapter proceeds as follows. In Section 3.2, we describe our
data and provide key details that have guided our choice set of US ports. Section
3.3 provides stylized facts on the evolution of container shipping for 1977–2023
servicing the US. In Section 3.4, we outline empirical strategies for identifying the
effect of vessel size and congestion on port dwell times. Section 3.5 concludes.
60
Container Ports
3.2 Data
In this section, we describe the multiple data sources we utilize to capture traffic
activity at US container ports between January 1977 and December 2023. We then
discuss how we compile and combine 47 years of port visits and vessel-positioning
data. Next, we explain how we match this data to ship registries, geographic areas
of ports, and international trade flows using the unique International Maritime
Organization (IMO) code assigned to each vessel. Lastly, we describe some general
summary statistics.
3.2.1 Ships and Port Data
Port Visits We utilize two sources of data to construct the first 39 years of
port visits data. For the first 25 years, 1977–2002, we utilize data from Lloyd’s
List Intelligence (Lloyds). Lloyds collected details on port visits at the day-level
through on-site agents situated at ports and paper records of vessel activity. This
data reports the unique ship IMO code, arrival date, departure date, gross tonnage
and name of each visiting vessel. For the next 13 years, We use Port Entrances
and Clearances data from the United States Army Corp of Engineers (USACE).
In addition to the details in the LLoyds data, these records has information on
the type of vessel, which allows us to identify which of these are containerships
(International Classification of Ships by Type, ICST).
Vessel-Positioning Data Our most granular data is based on an automatic
identification system (AIS) of vessels in which transponders fixed to ships reports
the geo-location, speed, status, vessel name and unique identifier for each vessel
present in US waters. These signals are stored and updated every minute by the
US Coast Guard and made publicly available on the MarineCadastre platform.
We observe minute-level individual vessel movements between January 2016 and
December 2023. Prior to 2016, usage of AIS transponders was not mandatory for
commercial vessels in US waters. Due to this limitation, we elect use port call
data for 1977–2015 which yields less precise durations of port visits – by calendar
date count – across vessels.
Each individual vessel location is tracked across time at a minute-level resolu-
tion. We pair these geo-spatial coordinates with spatial polygons of port statistical
areas. Any vessels observed within or in close proximity to a port is retained in
61
Container Ports
the data set. The focus of this study is placed specifically on containerships. We
therefore limit our use of AIS coverage to the top containership ports in the US.1
An individual port dwell event begins the moment a vessel stops within a port
polygon (signals a speed of zero) and begins broadcasting that it is engaged in the
loading/unloading of cargo (a status code of 5). The moment the vessel switches
to a positive speed and updates its status code back to being underway, we note
this as the end of a port dwell event. The passage of time between the start and
end of this time spent mooring at port represents the duration a vessel dwells at
port, loading and unloading cargo. Upon arriving at a resulting set of port dwell
events, we apply post-processing filtration to mirror procedures applied by the
Bureau of Transport Statistics.2 We provide a full account of the steps we have
devised in the Data Appendix for this chapter.
Ports In order to define the geographic areas of the 28 ports, we use geospatial
data from the U.S. Army Corps of Engineers.3 After some modifications, this
allows us to use polygon regions of US port areas and interact them with vessel
position data. In combination, these items allow us to identify the exact minute
at which port entrances and departures occur across key port areas of interest.
International Trade We source US trade data from the USACE and US Census
Bureau. The USACE provides records of US waterborne commerce between 1940–
2021, which details the volume of imports and exports by US port and year. The
US Census Bureau provides monthly containerized trade flows from January 2003
to December 2023 by port, country, and commodity type in both weight and value.
Vessel Registry We use the VesselTracking platform to update our historical
record of port calls with further information about individually identified vessels.
Using the IMO code of each vessel, VesselTracking provides the year built, max-
1These ports include the Port Authority of New York and New Jersey (NY, NJ), Savannah
(GA), Virginia (VA), Oakland (CA), Everglades (FL), the North Western Sea Alliance (WA), Los
Angeles (CA), Miami (FL), Long Beach (CA), Charleston (SC), Jacksonville (FL), Philadelphia
Regional Port Authority (PA), San Juan (PR), Baltimore (MD), Mobile (AL), Tampa Port
Authority (FL), Boston (MA), Wilmington (NC), Gulfport (MS), Wilmington (DE), Palm Beach
(FL), Hueneme (CA), San Diego (CA), and Everett (WA).
2For further details on the Bureau’s methodology in recording vessel dwell times, see
https://www.bts.gov/PPFS-Tech-Docs
3Our selection of ports is informed by 2020 Panjiva rankings of container handling volumes.
See https://www.logisticsmgmt.com/article/top 30 u.s. ports big ports got bigger in 2020, last
accessed on May 30th 2024.
62
Container Ports
imum container (TEU) capacity and length of each vessel. Given that this data
covers both existing and decommissioned vessels, we are able to apply full cover-
age across our sample. We limit our sample to vessels associated with non-zero
container capacities, if not already flagged as a containership by USACE port call
data.
3.2.2 Port Composition
Our study focuses on the top-ranked ports for containerized shipping activity, as
displayed in Table 4.1. In cases of close geographic proximity, aggregated data
reporting for sets of ports, or strategic alliances across individual port authorities,
we combine port areas together into single entities.4 In total, these ports repre-
sent 90.7% (88.4%) of the value (volume) of US containerized trade over the past
two decades. The 17 trillion USD amount of containerized trade we account for
represents 23% of trade value upon including all modes of transportation.
Table 3.1: Port Rankings by Containerized Trade Value (2003-2023)
Rank Port Value ($bn) Rank Port Value ($bn) Rank Port Value ($bn)
1 Los Angeles 4685.75 10 Seattle 573.96 19 Mobile 106.82
2 New York 2973.94 11 Baltimore 443.37 20 Boston(USA) 90.70
3 Long Beach 1617.12 12 Miami 418.82 21 Gulfport 47.60
4 Savannah 1376.86 13 Port Everglades 317.64 22 Palm Beach 35.29
5 Norfolk 1177.66 14 New Orleans 200.29 23 San Diego 28.33
6 Houston 1134.07 15 Philadelphia 195.86 24 Everett(WA) 26.23
7 Charleston 1096.86 16 Jacksonville 149.51 25 Port Hueneme 23.49
8 Oakland 797.44 17 Wilmington(NC) 123.75 26 Tampa 17.61
9 Tacoma 694.93 18 Chester(PA) 113.80 27 Freeport(Texas) 14.59
Source: US Census Bureau, HS Port-level Data. Values reflect total imports and exports summed across 2003 to
2023.
By region, Gulf and East coast ports have been increasing their shares of overall
waterborne commerce between 2003 and 2023. This is attributed to factors such
as the widening of the Panama Canal, which lessened East Asia’s reliance on the
4We have combined Tampa and Manatee ports. Similarly, we have also merged Seattle &
Tacoma ports into a single North Western Alliance (NWSA) entity and New York & Newark
ports, which are maintained by the Port Authority of New York and New Jersey (PANYNJ).
Lastly, due to LLoyd’s records combining Los Angeles and Long Beach, we have maintained this
combination throughout our sample into a single port category. In practice, the two San Pedro
Bay ports share strong competition between one another and unique business models for securing
long-term warehouse use/leasing.
63
Container Ports
US West Coast for major vessel operations, and the spatial market appeal of tran-
shipping to the US through highly modernized European ports. In recent years,
this attrition faced by the US West Coast has been accelerated by contentious rail
and port labor negotiations alongside severe challenges in port congestion.
Figure 3.1: Coastal Shares of Containerized Trade Value (2003-2023)
100
75
50
25
0
Gulf Coast West Coast East Coast
3.3 Stylized Facts
In this section, we document a series of stylized facts regarding container shipping
activity trends at US ports over the past 47 years. Overall, we observe 4,522
unique containers making a total of 624,570 port visits at the top 28 US container
ports over this time period.
Frequency of Port Visits We find that the total number of port visits by con-
tainerships per month has risen steadily, almost quadrupling in count between 1977
and 2010 (Figure 3.2). However, this trend stagnates after the Great Recession
and starts to decrease afterwards to around 1200 ships per month.
These port visits could be made by the same ship visiting more than once
in a month, or they could be made by multiple different ships. This distinction
is important since the frequency of port visits is often influenced by vessel size,
wherein smaller vessels normally service shorter distance routes and therefore are
more likely to repeatedly visit the US within a given month. Restricting our
number of visits to unique port-vessel events, Figure 3.3 shows that the increasing
64
Share (%)
2005
2010
2015
2020
Container Ports
Figure 3.2: Total Number of Port Visits by Containership
m1 m1 m1 m1 m1 m1 m1 m1 1 1
77 2 7 2 7 2 7 2 7
m 2m
19 19
8 98 99 99 00 00 011 1 1 2 2 2 20
1 022
Notes: This figure presents the total number of port visits by each containership per month-year
in our combined data set for the top 28 US ports spanning 47 years.
trend of port visits are not solely driven by the same ships visiting ports with
higher frequency, although the overall magnitude is down by more than half. The
total number of unique port visits by ships still increased by approximately three
times from 1977 to a peak of 600 visits around 2007 onwards. This allows us to
put together our first stylized fact:
Figure 3.3: Total Number of Unique Port Visits by Containers
m1 m1 m1 m1 m1 m1 m1 m1 m1 m1
779 98
2 879 99
2 97 02 7 2 7 2
1 1 1 1 19 20 20
0 012 20
1 2
20
Notes: This figure presents the total number of unique containership-port visits per month-year
in our combined data set for the top 28 US ports spanning 47 years.
65
Vessel Count Vessel Count
100 200 300 400 500 600 700 200 400 600 800 1000 1200 1400 1600
Container Ports
Stylized Fact 1. The number of port visits by containerships has increased by
more than three times over the past 47 years. The number of visits quadrupled at
its peak around the Great Recession and then declined after that. The increase in
port visits is driven by both the increase in visit frequency by the same ships, as
well as the increase in the introduction of new ships.
Variation of Ship Sizes Next, we examine how the sizes of containerships
have evolved over this period (Figure 3.4). We first define the size of a ship by its
container capacity, the maximum number of twenty-foot equivalent unit containers
each vessel can load (total TEU). Figure 3.4a shows that the average capacity of
individual containerships has increased almost eight fold across our sample. This
steady increase continues even though there was a decline in the total number of
containership visit between 2010 and 2017 (Figure 3.2). The utilization of larger
vessels at top US ports can explain the stagnating number of port visits since
these larger vessels do not need to visit as frequently. Figure 3.4b shows that the
total TEU volume of containership mass continues to increase over our period,
with a more recent decline in total capacity around the pandemic period. As the
building of newer and larger vessels require better technology, we can trace the
introduction of each of these new ship classes in our data period (Figure 3.4c).
We also show that this increasing trend in ship capacity is robust when restricted
to unique containerships (Figure 3.4d). This allows us to summarize our second
stylized fact:
Stylized Fact 2. The containerships visiting US ports are getting much larger.
Average capacity has increased by almost eight times over the past 47 years.
While both stylized facts highlight a broad acceleration in the evolution of ves-
sel technology from the mid-2000’s, we can delve deeper into these insights through
the observed vessel characteristics in our data. Given our knowledge of each ves-
sels’ capacity, length and build year, we assign classes to individual containerships
based on the classifications outlined in Rodrigue et al. (2017). The smallest class of
vessels, “Micro and Early Builds”, maintained a prominent share of the US market
until May 1988 (Figure 3.5). From 1988 to 2015, “Panamax” vessels functioned
as the primary means of transportation for containerized goods. By September
2016, “post-Panamax” containership technology had quickly ascended to support
the majority of US shipping. It has only been in recent years that “Neo-Panamax”
vessels and, on rare occassions, some of the largest vessels on the global fleet, have
begun servicing US ports.
66
Container Ports
Figure 3.4: Variation in Containership Capacity
(a) Average Ship Capacity (b) Total Ship Capacity
m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 17 2 7 2 7 2 m m
1 m1 m1 m1 m1
97 98 98 99 99 00 00
7 12 17 22 77 82 87 92 97 02 07 12 17 22
1 1 1 1 1 2 2 20 20 20 19 19 19 19 19 20 20 20 20 20
(c) Maximum Ship Capacity (d) Average Capacity of Unique Ships
1 1 1 1 1 1 1 1 1 1 m1 m1 m1 m1 m1 m1 m1 m1 m1 1
7q q q q q q q q q q 7 2 7 2 7 2 7 2 7 2
m
97 98
2 87 92 97 2 7 2 7 2 7 8 8 9 9 0 0 19 9 9 00 00 01 01 2 9 9 9 9 9 0 0 0 0
1 02
1 1 1 1 1 2 2 2 2 20 1 1 1 1 1 2 2 2 2 2
Notes: In panel (c) values at the tails of the distribution are windsorized by year to less extreme
values (0.05, 99.95) of TEU capacities (red). We do so to moderate the influence of once-off large
vessel visits, normally conducted to promote a port expansion event.
The top five container ports across the US have grown their service of con-
tainership mass ten fold by the peak year of 2017. Despite this impressive growth,
particularly in the cases of Norfolk and Savannah, this group of ports’ total market
share of containership services has waned since heights of 60% between 2002–2008.
Top five shares have since decline steadily to a range 51–54% in 2023, which sug-
gests a less concentrated market with respect to cargo handling services.
As containerized transport became increasingly reliant on larger vessels, ship-
ping distances and transit durations have grown, contributing towards a downward
trend in the number of vessel visits occurring per containership servicing the US
(Figure 3.7). This evidence is strikingly similar to Far East–North Europe patterns
67
Twenty-Foot Equivalent Units (TEU) Twenty-Foot Equivalent Units (TEU)
0 4000 8000 12000 16000 20000 0 2000 4000 6000 8000
Twenty-Foot Equivalent Units (TEU) TEU, Millions
1000 2000 3000 4000 5000 6000 0 1600 3200 4800 6400 8000
Container Ports
Figure 3.5: Changes in Container Capacity Shares
78 80 82 84 6 8 0 2 4 6 8 0 2 4 6 8 0 2 4 6 8 0 2 4
19 19 19 19 19
8 98 99 99 99 99 99 0 0 0 0 0 1 1 1 1 1 2 2 21 1 1 1 1 1 20 20 20 20 20 20 20 20 20 20 20 20 20
I - Micro and Early Builds III - Post Panamax V - Very Large Containership (VLCS)
II - Panamax IV - Neo-Panamax (NPX) VI - Ultra Large Containership (ULCS)
Notes: This figure presents the trends in TEU capacity shares of different vessel-types over the
47 year period.
Figure 3.6: Port Volumes of Vessel TEU Capacity by Year-Month
7m
1 m12 7m
1 1
2m 7m
1 1 1 1
2m 7m 2m 7m
1 m1
97 98
2
1 1 19
8 99 99 00 00 1 11 1 2 2 20 20 20
2
NWSA (Seattle/Tacoma) LA & Long Beach New York
Norfolk Savannah
outlined in Ducruet and Notteboom (2012) in which the number of port calls per
round trip decreased from five in 1989 to three in 2009.
Time Spent at Ports As the number of port visits has increased, along with
the sizes of ships making these visits, we are interested in how both these factors
contribute to the amount of time ships spend at ports. We calculate ship dwell
68
Share of TEU Capacity (%)
Container Units ('000) 0.2 0.4 0.6 0.8 1.0
0 500 1000 1500
Container Ports
Figure 3.7: Average Visits per Containership by Year
7 2 7 2 7 2 7 2 7 2
19
7
19
8 981 19
9
19
9 00 0 1 12 20 20 20 20
2
times as the time it takes a ship to enter and clear a port. Perhaps surprisingly,
average dwell times of containerships have stayed relatively stable at a 2.25 day
average, even in the presence of larger ship sizes and increased visits (Figure 3.8).
There are a number of deviations from this relatively stable number, like the
accumulated levels of traffic during COVID-19 post 202 where average dwell times
increase to more than three days. As we highlight in Figure 3.9, these challenges
were particularly pressing at the ports of Los Angeles and Long Beach, but persist
in other key areas to a lesser extent. Similarly to patterns in concentration of
container mass across ports, we find that the top five ports’ share of dwell times
has lessened from a 60% peak in 2002–2007 down to a modest 48% share in 2023.
This leads to our third stylized fact:
Stylized Fact 3. The average dwell times of containerships at ports have remained
relatively flat over time at an average of 2.25 days per visit, despite the increase
in number of port visits along with the sizes of ships making these visits.
With each period comes added reductions in per-unit handling costs of con-
tainerized goods as the global fleet further intensifies the use of improved vessel
technology. Vessel mass continues to accumulate at ports limited in space and
resources. These details invite two questions; (i) How were existing shipping ser-
vices’ access to port resources affected by these increasingly larger entrants to the
containerized market, and (ii) What are the implications of increased vessel mass
for the handling efficiency of container units at port? We explore these items in
ongoing work.
69
Port Visits
7 14 21 28
Container Ports
Figure 3.8: Average Dwell Times of Containership Visits
m1 m1 m1 m1 m1 m17 2 7 2 7 2 m
1 m1 1 1
7 8 8 9 9 0 07 12 17
m 2m
19 19 19 19 19 20 20 20 20 0
2
2
Figure 3.9: Average Dwell Times of Containership Visits by Port
m1 m1 m1 m1 m1 m1 m1 m1 m1 m17 2 7 2 7 2 7 2 7 2
19
7
19
8 981 19
9 99 00 0 1 1 21 2 20 20 20 20
NWSA (Seattle/Tacoma) LA & Long Beach New York
Norfolk Savannah All Ports
3.4 Empirical Strategy
In this section, we examine patterns in dwell time durations across individual
vessels visiting US ports. We capture average differences in dwell times across the
ship class categories to examine how a given ships’ time-variant mass influences
expected dwell times. Additionally, we consider the effects of existing port traffic,
its vessel type composition and the concentration of larger vessels on vessel dwell
times. Our findings suggest improved vessel technology can be a burden on the
existing global fleet servicing US trade.
70
Days Days
0 5 10 15 1 2 3 4
Container Ports
3.4.1 Dwell Time on Vessel Size
In this subsection, we examine differences in average vessel dwell times across
specific containership categories, each of which is associated with increasingly large
attributes in terms of vessel length, gross tonnage and maximum TEU capacity.
We estimate the following regression:
∑C
ln Ship Dwell Timespdmyr = βc Ship Categorys + δdmy + αpym + γr + εspdmyr,
c=2
(3.1)
where ln Ship Dwell Timespdmyr is the number of calendar days ship s spent at port
p, along coastline r, on day of the week d, month m, and year y. Ship Categorys
denotes the ship-specific size classification of vessel s, based on a series of guidelines
detailed in Rodrigue et al. (2017). Should a vessel be built at a particularly length,
width, maximum TEU capacity and gross tonnage to match these traits, we assign
the vessel a time-invariant categorization. δdmy is day-month-year fixed effects,
αpym is port-year-month fixed effects and γr is a regional fixed effect based on which
coastline vessel s is servicing. We use the smallest category – micro, early build
and Panamax vessels – as our reference group. As a result, βc coefficients should be
interpreted as relative differences in dwell times, compared to the smallest category
of containerships.
The day-month-year fixed effects control for aggregate events that impacts all
ships. The port-year-month fixed effects control for fixed and time-varying charac-
teristics at the port level. Time varying port characteristics account for potential
technology changes over time that ports can undertake that might affect ship dwell
times, for example technology upgrades at ports over time to accommodate larger
ships. Fixed regional characteristics include time-invariant comparative advantage
differences for coastlines that result in larger ships being received at these areas
which mechanically take longer time to unload, for example coastlines with deeper
natural harbors or regions adjacent to larger transport network hubs. We vary the
fixed effects in our model to also address fixed port characteristics like its geogra-
phy, and seasonality broadly applicable across US waterborne commerce, and find
relatively similar results.
We find that vessel size categories are highly predictive of port dwell times
across vessels. The larger a vessel is, the longer the duration of time it will spent
at port filling a berth-zone area while goods are unloaded and loaded. Relative
71
Container Ports
differences in Neo-Panamax and Very Large & Ultra Large Container ships exhibit
dwell times almost 50% larger in magnitude. Despite larger vessels introducing a
greater burden on port resources, in total volume, there may be economies of scale
to benefit from with respect to servicing one large vessel rather than two vessels
that, when combined in max TEU capacity, represent the same level of container
handling volume.
Table 3.2: OLS Vessel Dwell Time by Ship Category
Dependent Variable: Vessel Dwell Time (Days)
(1) (2) (3) (4)
Post Panamax 0.1981∗∗ 0.2215∗∗∗ 0.2214∗∗∗ 0.1989∗∗
(0.0782) (0.0764) (0.0764) (0.0784)
Neo-Panamax (NPX) 0.4308∗∗∗ 0.4755∗∗∗ 0.4757∗∗∗ 0.4315∗∗∗
(0.0967) (0.1084) (0.1081) (0.0956)
Very & Ultra Large Containerships 0.4760∗∗∗ 0.5951∗∗∗ 0.5938∗∗∗ 0.4715∗∗∗
(0.0819) (0.1270) (0.1264) (0.0834)
Vessel Age -0.0017 -0.0016 -0.0016 -0.0018
(0.0066) (0.0061) (0.0061) (0.0066)
Day–Month-Year FE ✓ ✓ ✓ ✓
Port–Year FE ✓
Port FE ✓
Port–Month FE ✓
Port–Year–Month FE ✓
Observations 613,385 613,385 613,385 613,385
R2 0.39 0.35 0.35 0.42
Notes: Clustered port standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. Examines variation
in the average dwell time of individual vessels relative to our reference group of early-build/micro vessels and
Panamax category containerships.
To address economies of scale, we use ratios of dwell time to container capacity
of each ship and repeat the regression specified in Equation 3.1. Assuming vessels
of varying size exchange similar proportions of container volume at port, relative
to their respective capacities, this regression details how efficient container han-
dling is across vessel size categories. We find that economies of scale are detectable
under this specification (Table 3.3). Efficiency improves across larger vessel cate-
gories. At the extreme, modern vessels of very large and ultra large’ proportions
are associated with 65–80% greater container handling efficiency relative to our
reference group of smaller ships. Controlling for size categories, we also find that a
one-percentage increase in vessel age is associated with a 0.13% higher dwell time
per container handled.
One important caveat of this exercise is the fact that the denominator of our
measure is time-variant and relies upon relative similarities across vessel classes
72
Container Ports
with respect to the proportion of handled container units relative to capacity. To
address this concern, we utilize Panjiva data which details individual vessel records
of container offloading. Matching Panjiva records of vessel activity to our existing
data, we are able to display similar results in which greater vessel size is associated
with higher container handling efficiency. These results are available in Appendix
A.3.
Table 3.3: OLS Vessel Efficiency by Ship Category
Dependent Variable: Dwell Time per Container Unit (Days per TEU)
(1) (2) (3) (4)
Post Panamax -0.5649∗∗∗ -0.5187∗∗∗ -0.5188∗∗∗ -0.5645∗∗∗
(0.0932) (0.0917) (0.0918) (0.0935)
Neo-Panamax (NPX) -0.6814∗∗∗ -0.5844∗∗∗ -0.5844∗∗∗ -0.6806∗∗∗
(0.1270) (0.1361) (0.1359) (0.1263)
Very & Ultra Large Containerships -0.8085∗∗∗ -0.6510∗∗∗ -0.6525∗∗∗ -0.8141∗∗∗
(0.1169) (0.1572) (0.1564) (0.1192)
Vessel Age 0.1271∗∗∗ 0.1277∗∗∗ 0.1277∗∗∗ 0.1268∗∗∗
(0.0137) (0.0129) (0.0129) (0.0136)
Day–Month-Year FE ✓ ✓ ✓ ✓
Port–Year FE ✓
Port FE ✓
Port–Month FE ✓
Port–Year–Month FE ✓
Observations 613,385 613,385 613,385 613,385
R2 0.53 0.50 0.50 0.54
Notes: Clustered port standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. Examines variation
container handling efficiency of individual vessels relative to our reference group of early-build/micro vessels and
Panamax category containerships.
3.4.2 Instrumental Variable Approach
A caveat of regressing vessel times on port traffic is the issue of simultaneity
bias. In a given period t, vessel i could choose observed port p due to mitigating
congestion issues elsewhere. As a result, the traffic conditions at port p partly
contribute to the observed duration of vessel i’s port dwell event.
To address this concerns about reverse causality, we intend to implement an
instrumental variable which projects exogenous variation in port traffic using vari-
ation in each port’s measure of trade exposure. This demand-shifter of port traffic
is uncorrelated with unobserved ship dwell times determinants ϵspdmyr. Our cur-
rent IV strategy uses aggregate changes for trade (imports and exports) at the
national level and our top port shares of historical containerized commerce, by
weight, to predict demand for traffic at each port:
73
Container Ports
Table 3.4: OLS Elasticity of Vessel Efficiency with respect to Port Traffic
Dependent Variable: Dwell Time per Container Unit (Days per TEU)
(1) (2) (3) (4)
Port Gross Tonnage -0.0132 -0.0337 -0.0334 -0.0039
(0.0123) (0.0349) (0.0350) (0.0151)
Port Vessel Count 0.1709∗∗∗ 0.0895∗ 0.0899∗ 0.1124∗∗∗
(0.0272) (0.0438) (0.0440) (0.0240)
HHI -0.0415∗∗∗ -0.0824∗∗∗ -0.0826∗∗∗ -0.0207∗∗∗
(0.0044) (0.0097) (0.0098) (0.0040)
Vessel Age -0.0017 -0.0065 -0.0065 -0.0019
(0.0094) (0.0085) (0.0084) (0.0093)
Large Vessel Share (%) 0.0855∗∗ 0.0682 0.0687 0.0705∗
(0.0342) (0.0611) (0.0612) (0.0382)
Ship FE ✓ ✓ ✓ ✓
Day–Month-Year FE ✓ ✓ ✓ ✓
Port–Year FE ✓
Port FE ✓
Port–Month FE ✓
Port–Year–Month FE ✓
Observations 613,385 613,385 613,385 613,385
R2 0.84 0.83 0.83 0.84
Notes: Clustered port standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. All variables are in
logs excluding ‘large vessel share’ which reports the percentage of visiting vessels in the 4th and 5th quintiles of
vessel size across ports and time. ‘Efficiency’ is dwell time per TEU, where the denominator uses each vessel’s
maximum container capacity. ‘Port Tonnage’ reports the weekly tonnage of containership vessels present at port
p, excluding ship i’s contribution. ‘Port Vessel Count’ reports the number of containership vessels present at port
for a given week. ‘HHI’ reports a Herfindahl–Hirschman index of vessel tonnage concentration. The higher the
value, the more densely concentrated total weekly mass is across visiting containerships at a given port p. ‘Vessel
Age’ reports the difference in year and build year of ship i. New entrant vessels are those built in the same year
they are actively servicing US ports. We apply an age of 0.5 to new entrants to preserve them in the sample.
These vessels are potentially key drivers of spillover variation in surrounding vessel dwell times.
For port p year y ∑
Port Trade Exposure py ≡ ωpy0 Ξpym
P
where the lagged weighted sum of total cargo tonnage into each port from year y0
during the historic period of 1959–1976.5 The shift is the nominal value of current-
year-month total trade across the US, non-seasonally adjusted, where trade flows
date back to January 1987.6 In future work, we will leverage the parameters gen-
erate from this exercise, combined with a micro-model of containership port dwell
events, to evaluate counterfactuals related to the introduce of additional berth
space and improved vessel technology. This latter goal of our seeks to enhance
policy guidance on port infrastructure investment following turbulent congestion
and maritime bottlenecks oscillating between the US East and West Coasts.
5Source: USACE Waterborne Commerce in the United States, Parts 1–5.
6Going forward, we will use undisclosed historical commerce flows, which vary by port–year–
commodity, and introduce improved granularity to our analysis.
74
Container Ports
3.5 Conclusion
This paper examines the evolution of containership technology and how it has
influenced both containership dwell times and port spillover effects for the sur-
rounding vessels. Using this evidence, we are able to characterize the length of
port dwell events as being subject to two key competing aspects; (i) each vessel’s
size influencing their own length and efficiency of port visits, and (ii) congestion
externalities associated with the introduction of improved vessel technology.
Upon accounting for size differences across vessel classes we find that efficiency
gains outscale the longer durations that larger vessels are subject to, suggesting
that vessels may be subject to economies of scale while processing goods at port.
This evidence is corroborated by insights from Panjiva Bill of Lading data, which
provide an added caution that US ports may not yet be capable of handling some
of the largest vessels in the global container fleet in an efficient manner.
To examine traffic effects, we consider how amassing volumes of vessel pres-
ence, in a given port–week, influences prevailing dwell times of individual vessels
accessing these same port resources. Both our OLS and instrumental variable
specifications suggest that increased vessel mass extends individual dwell times,
which suggests that the broader evolution of vessel technology may have adversely
influenced the performance of the existing container fleet on the intensive mar-
gin (congestion effect). However, one must also consider the fact that ports are
now far more concentrated in mass as a result of these innovations. Using a
Herfindahl–Hirschman Index of vessel mass concentration, find that as port re-
source use becomes more saturated by individual vessels, resulting dwell times of
vessels present at port diminishes (efficiency effect).
One aspect of an increased concentration of vessels in US transport services
that we do not address is the potential trade-offs between greater economies of scale
in container handling and mitigated gains as a result of weaker market competi-
tion. Hummels et al. (2009) suggests that price discrimination intensifies as vessel
counts narrow on a given lane of transport services. Extending this logic to trans-
port operators’ respective shares of leased port storage facilities, the cost reducing
elements of the efficiency gains we document may be somewhat undermined by
greater market power among choice mega-ships. We encourage further research in
this direction to evaluate how ownership compositions of arriving vessels influence
on-site container handling costs and pass-through to US importers.
75
Container Ports
In future work, we aim to quantitatively explore these competing effects of
improvements in technology both diminishing existing fleet performance on the
intensive margin while enhancing it on the extensive margin. This would allow
us assess the impact of pandemic port congestion, adjustments to port infrastruc-
ture and continued improvements in containership sizes. Our future pursuits will
develop a tractable general equilibrium framework to be able to quantify welfare
effects across counterfactual scenarios of this nature.
Bridge
While these essays up to this point have shed light on how logistical practices and
vessel characteristics influence the ebb-and-flow of maritime trade outcomes, one
big question arises from both of these forms of activity. What influence do vessels
have on environmental outcomes? While goals of vessel and port decarbonizations
have remained objectives of the IMO and US Department of Transportation, re-
spectively, the footloose nature of vessel traffic makes the detection of associated
environmental externalities difficult to apply pinpoint analysis towards. As I will
highlight in the subsequent chapter, even a reliance on air quality monitoring de-
vices adjacent to port areas can lead to a compounding of issues that limit this
for policy analysis. To mitigate these issues, I use vessel position data, allowing
me to dissect individual voyages towards US ports into their various stages and
evaluate the effect of the new San Pedro Bay vessel queuing system on both local
and global emissions.
Bibliography
Blonigen, B. and W. Wilson (2008). Port Efficiency and Trade Flows. Review of
International Economics 16 (1), 21–36.
Carballo, J., A. Graziano, G. Schaur, and C. Volpe Martincus (2023). Import
Processing and Trade Costs. IDB Publications (Working Papers) 12716, Inter-
American Development Bank.
Clark, X., D. Dollar, and A. Micco (2004). Port efficiency, maritime transport
costs, and bilateral trade. Journal of Development Economics 75 (2), 417–450.
Cullinane, K. and M. Khanna (2000). Economies of scale in large containerships:
76
Container Ports
optimal size and geographical implications. Journal of transport geography 8 (3),
181–195.
Ducruet, C., R. Juhász, D. Nagy, and C. Steinwender (2020). All aboard: The
effects of port development. NBER Working Papers 28148, National Bureau of
Economic Research, Inc.
Ducruet, C. and T. E. Notteboom (2012). Developing Liner Service Networks in
Container Shipping.
Ganapati, S., W. F. Wong, and O. Ziv (2021). Entrepôt: Hubs, Scale, and Trade
Costs. NBER Working Papers 29015, National Bureau of Economic Research,
Inc.
Hummels, D., V. Lugovskyy, and A. Skiba (2009). The Trade Reducing Ef-
fects of Market Power in International Shipping. Journal of Development Eco-
nomics 89 (1), 84–97.
Imai, A., E. Nishimura, S. Papadimitriou, and M. Liu (2006). The Economic
Viability of Container Mega-Ships. Transportation Research Part E: Logistics
and Transportation Review 42 (1), 21–41.
Parola, F., M. Risitano, M. Ferretti, and E. Panetti (2017, January). The Drivers
of Port Competitiveness: A Critical Review. Transport Reviews 37 (1), 116–138.
Rodrigue, J., C. Comtois, and B. Slack (2017). The Geography of Transport Sys-
tems. Routledge.
Tran, N. K. and H.-D. Haasis (2015). An empirical study of fleet expansion and
growth of ship size in container liner shipping. International Journal of Produc-
tion Economics 159 (C), 241–253.
U.S. Army Corps of Engineers (2021). Ports and port statistical areas. data re-
trieved from U.S. Army Corps of Engineers Geospatial, https://geospatial-usace.
opendata.arcgis.com/datasets/b7fd6cec8d8c43e4a141d24170e6d82f 0/explore.
77
Chapter 4
Cargo Ships & Coastal Smog: A
Case Study of San Pedro Bay
78
Cargo Ships & Coastal Smog:
A Case Study of San Pedro Bay
Philip Economides∗
June 2, 2024
Abstract
This paper examines the effect of a new port queuing system at San
Pedro Bay, California, on local and global maritime emissions using
data from the Environmental Protection Agency, MarineCadastre and
MarineTraffic. I find evidence which suggests that an estimated time of
arrival based queue port queue system encourages a slowdown of vessel
voyages. Furthermore, this added travel certainty lessens idling time
in the local port vicinity – despite deeper water waiting experiences
contributing positively to overall fuel consumption. Taking the voyage
and queuing legs of 2,111 vessel transits into account across over 2,000
distinct journeys, I associate the policy with a 30.2% reduction in these
vessels’ CO2 emissions.
JEL Codes: D62, F18, F64, L51, Q52, Q53, R41
Keywords: local environmental policy, transport externalities, local air pollution,
port logistics, vessel queues, congestion, maritime trade.
∗Department of Economics, University of Oregon.
I am indebted to Woan Foong Wong, Bruce Blonigen, Anca Cristea, and Nicole Ngo for their
advice and guidance. I would like to thank Mark Colas, Grant McDermott, Ed Rubin, and
seminar participants at the University of Oregon for suggestions and comments. All remaining
errors of this initial draft are my own.
Cargo Ships & Coastal Smog
4.1 Introduction
According to the World Bank, international trade in 2022 represented 62 percent
of global GDP – a remarkable ascension from a 24 percent share in 1970. With
goods more commonly sourced from abroad, the average transportation distance of
intermediate and final goods has also grown. From 1965 to 2020, the cargo weight
of short-distance trade rose by 45 percent, while longer-distance voyages more than
doubled their cargo volumes (Ganapati and Wong, 2023). Given these trends, the
maritime shipping industry has developed a significant carbon footprint.1
This form of activity – in which 70 percent of maritime emissions occur within
400 km of coasts (Corbett et al., 2007) – has been attributed to harmful health
outcomes for nearby populations (Capaldo et al., 1999; Liu et al., 2016; Gillingham
and Huang, 2021; Zhang et al., 2021). Reductions to the shipping industry’s
global emissions footprint face two distinct challenges; (i) vessel ownership is well
distributed across countries, highlighting a need for concerted global governance
efforts, and (ii) much of the available tools for reducing emissions are indirect in
nature, which allows for strategic responses by key stakeholders.
Both the International Maritime Organization (IMO) and US Department of
Transportation (DOT) have pursued policy seeking to decarbonize ports and the
maritime transport sector. For example, the IMO has member states targeting
reductions in total greenhouse gas (GHG) emissions of maritime transport by 20–
30 percent as of 2030 and 70–80 percent by 2040. Their main goal is net-zero GHG
emissions by 2050. Furthermore, a recent preview of the “US National Blueprint
for Transportation Decarbonization” by the DOT outlines how the US government
seeks to accelerate the transition to clean options in maritime shipping through
zero-emission fuels, technologies, energies, and new vessels.2
1Maritime shipping represents 2.89 percent of global CO2 emissions (IMO, 2021).
2Available at https://www.transportation.gov/priorities/climate-and-sustainability/us-
national-blueprint-transportation-decarbonization. Last accessed on May 30th 2024.
80
Cargo Ships & Coastal Smog
This paper examines how a logistical intervention by port authorities con-
tributes to decarbonization in the maritime transportation sector. I consider a
case study of the ports of San Pedro Bay, which sought to mitigate a large in-
crease in vessel emissions through the introduction of a new port queuing system.
Under the prior system, queue positions at the ports of LA and Long Beach were
issued on a first-come first-serve basis upon entering within 25 nautical miles of the
San Pedro Bay port complex. Vessel operators’ choice of voyage speed influenced
their assigned queue position. Should a port be at full capacity, vessels could ei-
ther drop anchor in close proximity to the cities of Los Angeles and Long Beach
or drift nearby, and begin contributing to local port emissions.
The US economic recovery from the COVID-19 epidemic led to a dramatic
increase in import demand, reflected by record numbers of commodity-laden con-
tainers entering the US between October 2020 and November 2021.3 This increase
in maritime trade activity caused historic levels of congestion across US ports,
leading to prolonged vessel queues and abnormally high GHG emissions. For ex-
ample, recent findings from the Los Angeles Port Authority’s Inventory of Air
Emissions (2021) suggest a 69 percent annual increase in port emissions of PM10
and PM25 between 2020 and 2021, with the rate as high as 143 percent specifically
for ocean-going vessels. Contributions through the anchorage and queuing stages
of vessel operations represented the majority of these emissions (Figure 4.1) and
were driven primarily by containerships (Figure 4.2). This period of congestion
led to numerous cases of environmental damage. In October 2021, just a month
prior to policy introduction, a crude oil spill occurred. This was later attributed to
anchor damage caused to a nearby pipeline by two containerships, which resulted
in losses and clean-up costs worth an estimated 160 million USD.4
32021 represented an 15.4 percent annual increase in container traffic, as opposed to a long
run average of 3.5 percent across the prior decade of activity.
4See National Transportation Safety Board, Project Summary: Marine Investigation - 130
Docket Items - DCA22FM001 and Report MIR-24-01, “Anchor Strike of Underwater Pipeline
and Eventual Crude Oil Release”. Last Accessed 03/09/2024.
81
Cargo Ships & Coastal Smog
Figure 4.1: 2005-2021 POLA OGV NOx Emissions (tons), by mode
Source: Los Angeles Port Authority’s Inventory of Air Emissions (2021)
Figure 4.2: 2020 & 2021 POLA OGV NOx Emissions (tons), by mode
Source: Los Angeles Port Authority’s Inventory of Air Emissions (2021)
To address the ports’ escalating levels of pollution and prevailing environmen-
tal damage, the vessel queuing system was updated. On November 11th 2021,
the ports of LA and Long Beach jointly announced a new procedure in which ves-
sels would be assigned queue positions by Pacific Maritime Management Services.
Rather than the prior “first-come first-serve” process, a queue position would in-
stead be determined by an estimated time of arrival (ETA) from a prior port of
departure, assuming a constant travel speed. Furthermore, arriving ships are re-
82
Cargo Ships & Coastal Smog
quired to wait outside the ‘Safety and Air Quality Area’, 150 nautical miles out
at sea, rather than within the anchorage zone. Although the policy may reduce
travel speeds through guaranteed queue positions, the added requirement to re-
main adrift further out at sea introduces an ambiguous global vessel emissions
effect. Should it be the case that vessels hardly changed their speeds and fuel con-
sumption while adrift off the continental shelf is particularly taxing on emission
levels, this policy could be considered a form of environmental NIMBY-ism, sim-
ilarly to cases featured in Morehouse and Rubin (2021) and Zou (2021).5 In this
case, ports would be reaping the commercial benefits of their property rights while
exporting the cost of any resulting air pollution. Alternatively, if speed reductions
are considerable or vessels spend less time idling due to improved certainty on
port admittance, the greater fuel expended during the queuing process may be
negligible in comparison. Under such circumstances, this logistical practice may
offer oversight bodies, such as the International Maritime Organization or the US
Department of Transportation (DOT), an additional policy tool through which to
meet long-term goals of a decarbonized maritime transport sector.
I quantify the local and global effect of this policy on vessel-related carbon
emissions. First, I use a difference-in-difference approach to estimate the new
system’s effect on areas in proximity to the ports, using the EPA’s local air quality
measures, where relevant controls consist of (i) port areas of Oakland, Seattle
and Tacoma, and (ii) nearby inland regions.6 While Figure 4.3 suggests a rather
dramatic reduction in local emissions, given the subsequent decline in ships in
anchorage, I find evidence to suggest that countervailing effects offset any potential
improvement in local air quality. I also ask whether this decision simply shifted
emissions elsewhere or functioned as a tool to reduce global vessel emissions. I
5“Not In My BackYard”, a term coined by Mitchell and Carson (1986). Often associated with
cases in which a property owner wants a beneficial economic activity to occur on her property
while expelling the negative externalities of production elsewhere.
6This data has previously been used to highlight how local vessel emissions contribute neg-
atively to health outcomes (Gillingham and Huang, 2021; Hansen-Lewis and Marcus, 2022).
83
Cargo Ships & Coastal Smog
resolve this ambiguity by weighing changes in both voyage & queuing behaviour
and estimate a 30.2% decline in the global emissions of containerships inbound for
San Pedro Bay, due to the new vessel queuing system.
Figure 4.3: POLA Number of Container Ships at Anchorage, Daily
Source: Marine Exchange of Southern California
This paper contributes to several strands of literature. Firstly, it adds to a
recent selection of studies leveraging use of relatively modern transportation data.
I rely on automatic identification system (AIS) vessel traffic to observe the exact
geo-spatial positions of vessels at a one-minute resolution through big data queries
(Heiland et al., 2019; Prochazka et al., 2019; Tumbarello et al., 2019; Brancac-
cio et al., 2020; Wong, 2022; Wong and Fuchs, 2022). As Klotz and Berazneva
(2022) highlights, the precision of AIS data allows for granular analysis into coastal
movements and berthing at local ports, which is essential for estimating associ-
ated emissions levels. I consider vessel–specific, within–route emissions across two
stages of vessel transit; (i) voyage – transit between a prior port of departure and
US waters, and (ii) queuing – the anchoring and drifting of vessels awaiting port
entrance along the US west coast. This study offers a new methodology by which
AIS vessel positions can be used to map drifting and anchoring behaviour within
port queues and infer individual emission levels across vessels.
I also contribute to topics of transports policy, congestion and environmental
conservation (Kinney et al., 2011; Cristea et al., 2013; Ngo et al., 2015; Shapiro,
84
Cargo Ships & Coastal Smog
2016; Rivera, 2021). Gillingham and Huang (2021) finds that emissions from
vessels at-berth in US ports contributed to reduced child birth weights. Similarly,
Hansen-Lewis and Marcus (2022) concludes that the introduction of US maritime
emissions control areas significantly decreased fine particulate matter, low birth
weight, and infant mortality. However, strategic responses of transport operators
limited these gains to only half of forecasted reductions. I quantify local vessel
emissions through queuing activities and quantify global emissions associated with
inbound voyages & out-of-port congestion. By holding service lanes fixed across
my sample – I mitigate the influence of various strategic responses that transport
operators may be availing of. I also document that queuing behaviour, while
limited to narrow geographic regions, represents at least 20% of transit time for
inbound containerships along the US West Coast.
Third and lastly, this paper advances our knowledge of viable tools for environ-
mental regulatory intervention. Given that an ETA-based policy resulted in global
emissions falling and vessels spending less time idling outside of harbor regions,
this may not only limit negative externalities associated with the industry but also
boost its productivity. My findings suggest that queuing activity has acted has the
main driver of emission reductions. Small– to medium–sized ports, less subject
to excessive queuing behaviour, may be less likely to capture the environmental
benefits of this logistical practice.
The remainder of the paper proceeds as follows. Section 2 provides background
and a description of the data before shifting into the means by which I estimate
at-sea vessel emissions both at the voyage and queuing stages. Section 3 details
the empirical strategy for identifying the local & global policy effects of the new
queuing system and discusses results. Section 4 concludes.
85
Cargo Ships & Coastal Smog
4.2 Data & Inference
In this section, I describe the combination of several publicly available datasets and
a privately purchased port call data, which allows for the construction of individual
vessel emission records between November 2019 to November 2022. The data fall
into three broad categories: air quality, vessel, and geographic data. I then outline
how I infer associated fuel consumption and emission levels using observed vessel–
voyage events. I elaborate on data preparation in Appendix A3, Section I.
4.2.1 Data
Air Quality: To address local emissions levels, I use daily Air Quality Index mea-
sures provided by the Environmental Protection Agency (AQI). This data relies
on outdoor monitors across the United States. States determine monitor place-
ments based on areas of relatively high population. According to the EPA, each
state is responsible for developing its own monitoring plan, which is then reviewed
and revised every five years. To include additional monitors, these instruments
must follow “stringent siting and quality assurance criteria and are maintained by
state and local agencies.” Each monitor’s report details ozone, particulate matter,
carbon monoxide, sulfur dioxide and nitrogen dioxide levels.7 Each state-county
in the dataset contains multiple site locations.
I form separate sets of inland and coastal sites across Los Angeles and Long
Beach to compare control and treatment group outcomes, respectively. For ro-
bustness, I also use comparable zones in proximity to port regions along the West
Coast, untreated by the new ETA–based queuing system.
7The reported value identifies the highest concentration of a given pollutant among all of the
monitors within each reporting area. It will select two breakpoints that contain the concentration.
The index value is calculated as I = IHi−ILop BP −BP (Cp − BPLo) + ILo, where Cp is the truncatedHi Lo
concentration of pollutant p, {BPHi,BPLo} are the breakpoints associated with the truncated
value, and {IHi, ILo} are their associated AQI values.
86
Cargo Ships & Coastal Smog
Vessels: I have purchased port call data from MarineTraffic (MT), which pro-
vides a record of port entrances and exits internationally. I focus on a subset of
visits across the US west coast – Seattle, Tacoma, Oakland, Los Angeles and Long
Beach – dated between November 2019 and November 2022. MT data details each
vessel’s prior port of departure and departure time, and port of arrival and arrival
time. Each vessel is also uniquely identified by their assigned International Mar-
itime Organization (IMO) codes, which remain invariant to changes in ownership,
flag and vessel name. Upon cleaning, this dataset consists of 1,061 uniquely iden-
tified vessels servicing 10,035 port visits, where vessels’ container capacities range
from 200–23,000 twenty-foot equivalent units.
While port call data is helpful in establishing transit routes, distances and du-
rations, this alone would not allow one to distinguish between voyage and queuing
stages of transit. To separately these separate stages, I combine MT with records
of individual vessel movements in US waters. These US Coast Guard records use
vessels’ transceivers that signal the IMO code, status, speed, geo-location, and
name of a given ship. I use a publicly available provision of this signal data from
MarineCadastre (MC) for containership vessel movements at a minute resolution
between January 2019 to December 2022. Lastly, I use time-invariant vessel char-
acteristics from VesselTracking (VT) to assign each vessel their gross size and
container capacity. These metrics are key in interpreting how vessel speed maps
into fuel consumption.
Using unique vessel IMO codes and associated transit timelines, I combine
the coordinates and timestamps of vessels’ MT port departures with MC reap-
pearances in US waters. This allows for the assignment of exact travel distances
and speed to each individual vessel voyage stage. Transit between US waters ar-
rival and eventual admittance into a US port represents the queuing stage of each
voyage. Combining these two events represents the full transit experience of an
87
Cargo Ships & Coastal Smog
individual vessel. By joining the two datasets, I can discern which transits in US
waters represent international journeys (Figure 4.4).
Figure 4.4: Match Status by Origin and Group
control treatment
100
75
50
25
0
2020 2021 2022 2023 2020 2021 2022 2023
Year-Month
Domestic - Missing Position Domestic - Observed Position
International - Missing Position International - Observed Position
Upon dropping extreme outliers – determined based on average travel speeds
of individual voyages – I use a sample of 10,035 vessel voyages serviced by 1,061
distinct containerships.8 Table 4.1 characterize each included port by period.
Table 4.1: Summary Statistics of Vessel Voyages
Port Period Visits Vessel Age Max TEU Dwell Time Voyage Time Voyage Speed
Long Beach pre- 1708 11.45 7705.59 93.97 237.38 17.22
Long Beach post- 751 12.74 7466.68 100.84 339.94 15.12
Los Angeles pre- 1890 10.94 7491.60 119.64 259.50 18.22
Los Angeles post- 800 11.27 7310.79 132.11 385.55 14.62
Oakland pre- 1906 12.84 7145.26 48.77 36.09 4.01
Oakland post- 729 16.06 5370.96 67.75 124.45 6.81
Seattle pre- 706 12.68 6341.05 56.10 75.57 5.60
Seattle post- 375 13.28 5541.22 52.30 196.03 9.29
Tacoma pre- 936 18.35 4716.93 49.22 150.26 15.30
Tacoma post- 412 20.64 4721.33 64.44 172.31 14.72
Columns 4–8 report averages across vessels for their respective port and time period. “Pre-” is defined as prior
to November 11th 2021, when the new San Pedro Bay queuing system was announced for LA and Long Beach.
‘Vessel Age’ is reported in years. ‘Max TEU’ reports the container capacity of vessels, ‘Dwell Time’ details how
many hours vessels spend handling goods at port. ‘Voyage Time’ details hours between a departure time and the
point at which a vessel reappears in US waters. ‘Voyage Speed’ is reported in nautical miles per hour.
8I follow Davies and Jeppesen (2015), which labels extreme outliers as values outside a specific
closed range. The lower bound is the 25th percentile of average travel speeds less three times
the interquartile range (75th percentile value less the 25th percentile value). The upper bound
is the 75th percentile of average travel speeds plus three times the interquartile range.
88
% Share
Cargo Ships & Coastal Smog
As these statistics highlight, there are a considerable number of journeys taking
place within proximity to the North American continent. Many cases involve
vessels completing subsequent visits at nearby ports upon arriving at the US West
Coast. For my empirical approach, I first consider the new queuing system’s
effect on broader vessel activity, and then limit my evaluations to consider only
international journeys.
There is also the potential that transport operators may strategically respond
to the policy by switching destination ports from abroad. For example, if I wished
to preserve my ability to endogenously influence queue positions, I could elect to
switch my initial US entry point to a non-treated port. Upon originating from the
non-treated port, my subsequent journey to a treated port has now been reduced
dramatically and a far lower percentage of overall transit services is exposed to the
new queuing system. To avoid this issue of partial selection out of treatment, I
focus my third set of results on the subsample of vessels that maintained services to
their original pre–policy destinations after the implementation of the policy. Any
vessels that no longer serviced the West Coast, as well as newly introduced vessels
appearing only after policy implementation are also excluded to avoid concerns of
strategic responses that rotate vessels in and out of the West Coast trade network.
“Switcher”, “Entrant” and “Exit” vessels are excluded from this third empirical
approach. The remaining balanced panel of vessel activity occurring consistently
prior– and post–policy includes 2,354 port visits serviced by 301 distinct vessels.
Geographic: I use Port Statistical Area data, made available through the
United States Army Corp of Engineers (USACE). After some modifications, this
allows me to represent polygon regions of key port areas. Using these boundaries,
I identify the exact minute at which port entrances and departures occur across
US ports, which spells the end of the queuing stage for a given vessel.
89
Cargo Ships & Coastal Smog
4.2.2 Fuel Consumption
To study variation in vessel emissions, I require a measure of fuel consumption. I
infer these levels by inputting vessel’s travel speeds and ship sizes (TEU capacity)
into fuel consumption functions featured in Figure 4.5 from Rodrigue (2020).9 I
interpolate across these TEU-specific functions and generate a bi-variate, polyno-
mial function through OLS.10
Figure 4.5: Fuel Consumption by Speed and TEU Capacity
Source: Rodrigue (2020). The Geography of Transport Systems, Chapter 4, Transportation and Energy.
For static berth activity, I recommend the use of a scaling factor proposed by
Hulskotte and Denier van der Gon (2010) which receive frequent use in the mar-
itime logistics literature (Jalkanen et al., 2012; Ju and Hargreaves, 2021; Schwarzkopf
et al., 2021). This is a linear function of the gross tonnage of vessels times the
hours spent stationary (Table 4.2). For the purposes of this study, assessing voy-
9Since I do not observe engine types across vessels, I do not use an alternative naval engineer-
ing formula featured in Corbett et al. (2009) and Lugovskyy et al. (2023), which combines speed
and engine types across main and auxiliary units. Ship size is recognized as a strong predictor
of emissions and has sufficed in applied studies of vessel emissions (Walsh and Bows, 2012).
10Assessing the model fit of this exercise, the function appears to best perform at standard
cruising speeds of 10–25 nautical miles (knots) per hour. However, at slower speeds, inferred
functions of fuel consumption levels do not behave reliably. I introduce a conditional estimation
of fuel consumption in which levels below cruise speed (10 knots) receive additional adjustments
based on prevailing studies of vessel emissions at slow movement stages. See Appendix A3 (II.)
for further details.
90
Cargo Ships & Coastal Smog
age and queuing stages of transit, I do not include berthing activity in broader
analysis.
Table 4.2: Estimate of Fuel Consumption by GT-Dwell Hours
Type of ship Fuel Consumption Rate Average hotelling time
(kg fuel/1000 GT h) at berth (hours)
Oil Tankers 19.3 28
Chemical and other tankers 17.5 24
Bulk Carriers 2.4 52
Containers 5.0 21
General Cargo 5.4 25
Ferries and RoRo 6.9 24
Reefers 24.6 31
Other 9.2 46
Source: Hulskotte and Denier van der Gon (2010). Fuel Consumption and Associated Emissions from Seagoing
Ships at Berth Derived from Onboard Survey, Atmospheric Environment, Vol. 44(9), pp 1229–1236.
Upon calculating fuel consumption for each stage of transit, I use assumptions
detailed in the proceeding subsection to generate associated emissions levels for
each vessel’s various stages of transit.
4.2.3 Vessel Emissions
I refer conversion rates provided by Czermanski et al. (2021), which converts tons
of fuel to CO2, SOX , NOX and PM2.5 kilogram emissions. I assume vessels are
in accordance with IMO 2020, which mandates an upper limit of 0.5% sulfur
content in vessel fuel sources and therefore exclusively use the first row of featured
conversion rates (Table 4.3).
Table 4.3: Emissivity Indices for Selected Marine Fuels
[kg/t of Fuel]
Fuel CO2 SOX NOX PM2.5
MDO 0.5% 3206.00 10.50 50.50 2.30
HFO 1.5% 3114.00 31.50 51.00 3.40
HFO 2.0% 3114.00 42.00 51.00 3.40
HFO 3.5% 3114.00 71.50 51.00 3.40
LSHFO 0.5% 3151.00 10.50 51.00 2.30
LSMGO 0.1% 3151.00 2.10 50.50 2.30
LNG 2750.00 ¡0.02 8.40 0.02
Methanol 1375.00 0.00 26.10 0.02
HFO + SCRUBBER + SCR 3176.00 0.84 7.65 0.51
Source: Czermanski et al. (2021), based on the assumptions of the Med Atlantic Ecobonus (MAE) Project, MAE
External Cost Calculator Tool.
91
Cargo Ships & Coastal Smog
For a full set of details regarding how vessel emissions are calculated for slow
travel speeds in the local port area, see Appendix A3, Section III.
4.3 Empirical Strategy & Results
Variation in local emissions can be represented through two types of measures;
(i) observed general measures of air quality from the Air Quality Index, and (ii)
estimated vessel-specific emission levels implied by the fuel consumption of contain-
erships reporting AIS signals in the San Pedro Bay area. Using global movement
data, I also examine whether this decision shifted the emissions problem elsewhere
or reduced the global emissions of vessels. This involves carefully dissecting in-
dividual vessel voyages across voyage, queuing and berthing stages of transit. In
the proceeding section, I first detail my empirical strategy and then yield policy
effects for both local and global vessel emissions.
4.3.1 Empirical Strategy
Vessel position data reports the speed and status codes of each vessel in US waters,
allowing me to identify whether vessels are underway (U), maneuvering (M), at
anchor (A) or at berth unloading their goods ((B). U)sing local emissions efficiency
eL
– represented as an emissions per hour rate iptL – summed across these fourhipt
local transit stages,
eL U M A Bipt eipt + eipt + eipt + eipt
= , (4.1)
hL hU + hM + hA Bipt ipt ipt ipt + hipt
may analysis is capable of focusing on each individual stages of local activity. For
simplicity, I combine activity across underway, maneuvering and anchoring activity
as the “Queuing Stage” (Q) – which excludes berthing.
92
Cargo Ships & Coastal Smog
To measure global emissions, I take these measures jointly with voyage (V)
emissions of each ship from their respective ports of departure, eVipt. The fuel
efficiency of each global voyage is expressed as,
eG eV Qipt ipt + eipt
= (4.2)
hGipt hV
Q
ipt + hipt
To exploit the quasi-experimental nature of this policy change – announced
and introduced within a 5-day window – I consider the following specification:
Y Siodt = +δ
ST S S S S1 d + δ2Dt + δ3 (Td ×Dt) + ϕio + ϕt + µirpt, (4.3)
where i identifies the individual vessel, io indicates the vessel–voyage route serviced
repeatedly over time, and t represents the associated time period. Dummies Td and
Dt act as treatment and post-period dummies, respectively. Control group ports
include Oakland, Seattle and Tacoma which are also exposed to shocks common
across the US West Coast, while not treated by this particular policy. The fuel
efficiency outcome variable, Y Lipt, is emissions-per-hour on a given transit stage,
S ∈ {V,Q,G} and can therefore be decoupled for any particular stage of service
within a given vessel-voyage, iodt. The coefficient of interest, δG3 , indicates the
global emissions effect of the policy, where δ̂G3 < 0 would suggest that this new
queue system is a tool that could assist in efforts to decarbonize vessels servicing
US ports. Key to this identification approach is the assumption of parallel trends.
In Figure 4.6, a clear parallel trend exists in the absence of treatment between
treated and control port groups.
I suggest that the guaranteed queuing position issued in the post-period portion
of the sample may have encouraged a relatively higher percentage of vessels to
slow-steam towards their West Coast destination, lowering average emission levels
across the voyage length of a given transport service. Referring to Eq. (4.2),
93
Cargo Ships & Coastal Smog
Figure 4.6: Parallel Trends in Average Voyage Speeds
20
16
12
2020 2021 2022
Control Treatment
eV
this would suggest that iptV should decline. Furthermore, if a queue position ishipt
known then transport operators would be able to better time their arrivals and
minimize unproductive idling near the port area. Counteracting this effect, the
increased extent to which vessels must remain idle outside of the treated ports’
respective anchorage zones would contribute to greater degrees of drifting and
repositioning (distance travelled) relative to being able to lay anchor, which may
elevate emission per hour rates among visiting vessels. Examining Eq. (4.2), and
supposing we treat drifting as an alternative form of deep-sea anchorage, this would
eQ
suggest that iptQ could rise. This leads to an ambiguity in the values of {δG Qh 3 , δ3 }ipt
which may be of empirical interest to port authorities that wish to implement
environmentally-friendly logistical changes to port management. Should δ̂G3 < 0,
this would suggest that a simple queuing system change could contribute to existing
efforts by the IMO to reduce the carbon footprint of the maritime transport sector.
If δ̂G3 >= 0, this would implies that the policy is either ineffective in achieving air
quality improvements and may possibly be a negative externality generating form
of environmental NIMBY-ism.
94
Speed, Average (NM per Hour)
Cargo Ships & Coastal Smog
I also employ use of a difference-in-difference estimator to examine the policy
effect on local air emissions in treated and non-treated US West Coast port areas.
I estimate the policy effect on local emissions using the following specification:
Yipt = α + γ1Tp + γ2Dt + γ3 (Tp ×Dt) + εipt, (4.4)
where i identifies the individual location (or vessel), p indexes a given port region,
t represents the associated date. Dummies Tp and Dt act as treatment and post-
period dummies, respectively. Tp takes a value of 1 for any given coastal locations
along Los Angeles and Long Beach, and is 0 otherwise.11 Dt takes a value of 1
for any time period from November 11th 2021 onward. The dependent variable
includes the following forms of emissions from AQI data; carbon monoxide (CO),
sulfur oxides (SOX), nitrogen oxide (NOX), fine particulate matter (PM2.5), and
coarse particular matter (PM10).
4.3.2 Local Emission Effects
In this subsection, I outline the results of Equation 4.4. In Figure 4.7, I display
Zone I – ‘port-adjacent’ monitors within 25 miles of the centroid of highlighted port
areas – and Zone II – monitor groups between 25-50 miles from their respective
port centroid. Miles of distance between points are calculated using Vincenty
distance, which accounts for oblate spheroid curvature of the earth.
I follow two specifications which alternate the key control group. For the
initial set of results, the treatment (control) group consists of monitors within 25
(between 25-50) miles of the San Pedro Bay port complex. In the second set of
results, the treatment (control) group consists of monitors within 25 miles of the
11To address concerns about vessel pollutants spreading across nearby coastal and inland
regions of San Pedro Bay in a similar manner, an alternative treatment indicator compares sites
associated with treated ports against sites of non-treated port cities along the West Coast (e.g.
Oakland, Seattle, Tacoma).
95
Cargo Ships & Coastal Smog
San Pedro Bay port complex (ports of Seattle or Oakland). Despite a considerable
reduction in port traffic in the anchorage and drift zones adjacent to the ports of
Los Angeles and Long Beach, a review of variation in local conditions suggests no
notable reduction in associated air pollutant concentrations.
Figure 4.7: AQI Monitor Local Sites
Note: Each polygon depicts a port area. Red sites represent zone I monitors. Blue sites represent zone II monitors.
Upon reviewing the on-site effect, which compares zones I and II within San
Pedro Bay, I find a significant increase in local emissions (Table 4.4). I find that
the new system contributed positively to port-adjacent emissions through a; (i)
10.9% increase in carbon monoxide, (ii) 12.1% rise in nitrogen dioxide, (iii) 11.2%
rise in oxides of nitrogen, and (iv) 3.9% elevation in ozone levels.12
Using particulate matter (PM) measures of pollutant concentrates, I find sim-
ilar evidence of broad increases in emissions across monitors adjacent to the port
area, relative to the outer zone of monitors (Table 4.5). PM10 emissions rose by a
23% and PM2.5 rose by 5.4%. Sulfur emissions showed a slight decline, although
not to a statistically significant extent.
In the second specification of this assessment, I contrast outcomes across port-
adjacent monitors near the ports of Los Angeles and Long Beach to other port-
adjacent monitors positioned along the West Coast (Table 4.6). This may address
concerns that particular vessel emissions may struggle to permeate regions beyond
12These upward shifts are relative to adjacent inner-city monitor zones distanced between
25-50 miles from the San Pedro port area.
96
Cargo Ships & Coastal Smog
Table 4.4: Difference-in-Difference, Control: San Pedro Bay, Zone-II
Dep. Variable: CO NO2 NOx Ozone
(1) (2) (3) (4)
Post–Period 0.0976 0.1382 0.2977 -0.2403
(0.2539) (0.1670) (0.2551) (0.1635)
Treated 0.0189 -0.0392∗∗ 0.0819∗∗∗ -0.1824∗∗∗
(0.0031) (0.0014) (0.0012) (0.0001)
DiD 0.1089∗∗ 0.1213∗∗∗ 0.1121∗∗∗ 0.0386∗∗∗
(0.0019) (0.0015) (0.0013) (4.14× 10−5)
Day FE ✓ ✓ ✓ ✓
Month-Year FE ✓ ✓ ✓ ✓
Observations 17,804 21,127 21,141 18,294
R2 0.30 0.33 0.32 0.49
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering by monitor zone. Each observation
is a distinct day-monitor-port emission type reading. ‘Post-Period’ is equal to 1 for dates November 11th 2021
to October 30th 2022. ‘Treatment’ is equal to 1 for air pollutant concentration monitors within a 25-mile radius
(Zone I) of the centroid of the San Pedro Bay port complex. The relevant control group consists of monitors
within 25-50 miles of the same reference point (Zone II).
Table 4.5: Difference-in-Difference, Control: San Pedro Bay, Zone-II
Dep. Variable: PM10 PM10−2.5 PM2.5 SO2
(1) (2) (3) (4)
Post–Period -0.1898∗ 0.2919 -0.3486 0.3764∗
(0.0160) (0.0582) (0.2103) (0.0311)
Treated -0.3151∗∗∗ -0.5804∗∗∗ 0.0032 -0.4552∗
(0.0018) (0.0035) (0.0014) (0.0370)
DiD 0.2302∗∗ 0.1582∗∗ 0.0541∗ -0.0966
(0.0052) (0.0070) (0.0045) (0.0349)
Day FE ✓ ✓ ✓ ✓
Year–Month FE ✓ ✓ ✓ ✓
Observations 6,949 1,303 14,471 3,853
R2 0.22 0.40 0.25 0.19
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering by monitor zone. Each observation
is a distinct day-monitor-port emission type reading. ‘Post-Period’ is equal to 1 for dates November 11th 2021
to October 30th 2022. ‘Treatment’ is equal to 1 for air pollutant concentration monitors within a 25-mile radius
(Zone I) of the centroid of the San Pedro Bay port complex. The relevant control group consists of monitors
within 25-50 miles of the same reference point (Zone II).
25 miles of the US coastline. Results are broadly supportive of findings in Table
4.4, both in terms of direction and relative magnitudes across emission types.
Referring to particulate matter emissions in Table 4.7, port-to-port difference-
in-difference comparisons yield further evidence of a statistically significant rise
in emissions upon the introduction of the new port queuing system. Notable
deviations from the zone-based results listed in Table 4.5 include a near doubling
of particulate matter between 2.5 and 10 micrometers in diameter. Additionally,
97
Cargo Ships & Coastal Smog
Table 4.6: Difference-in-Difference, Control: Seattle/Oakland, Zone-I
Dep. Variable: CO NO2 NOx Ozone
(1) (2) (3) (4)
Post–Period 0.3445∗∗ 0.3377∗∗ 0.5386∗∗ -0.4985∗∗
(0.0639) (0.0639) (0.1071) (0.0712)
Treated -0.2141∗ 0.3427∗∗ 0.3074 0.2007∗∗∗
(0.0502) (0.0579) (0.1054) (0.0078)
DiD 0.1577∗∗∗ 0.1085∗∗ 0.1864∗∗∗ 0.0402∗∗
(0.0078) (0.0145) (0.0019) (0.0073)
Day FE ✓ ✓ ✓ ✓
Year–Month FE ✓ ✓ ✓ ✓
Observations 17,788 21,738 20,732 17,385
R2 0.29 0.41 0.36 0.41
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering by monitor zone. Each observation
is a distinct day-monitor-port emission type reading. ‘Post-Period’ is equal to 1 for dates November 11th 2021
to October 30th 2022. ‘Treatment’ is equal to 1 for air pollutant concentration monitors within a 25-mile radius
(Zone I) of the centroid of the San Pedro Bay port complex. The relevant control group consists of monitors
within 25 miles of the Port of Seattle and the Port of Oakland.
a somewhat significant decline in sulfur dioxide is detected.
Table 4.7: Difference-in-Difference, Control: Seattle/Oakland, Zone-I
Dep. Variable: PM10 PM10−2.5 PM2.5 SO2
(1) (2) (3) (4)
Post–Period 0.0164 -0.3332 0.4302 0.1943
(0.2726) (0.2562) (0.3694) (0.1447)
Treated 0.5940 1.426∗∗∗ 0.5539∗∗ -0.2624∗∗
(0.2523) (0.0051) (0.1093) (0.0510)
DiD 0.2863∗∗∗ 0.9336∗∗∗ 0.0107 -0.3240∗
(0.0264) (0.0040) (0.0113) (0.1001)
Day FE ✓ ✓ ✓ ✓
Year–Month FE ✓ ✓ ✓ ✓
Observations 3,330 925 27,184 8,802
R2 0.38 0.71 0.22 0.07
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering by monitor zone. Each observation
is a distinct day-monitor-port emission type reading. ‘Post-Period’ is equal to 1 for dates November 11th 2021
to October 30th 2022. ‘Treatment’ is equal to 1 for air pollutant concentration monitors within a 25-mile radius
(Zone I) of the centroid of the San Pedro Bay port complex. The relevant control group consists of monitors
within 25 miles of the Port of Seattle and the Port of Oakland.
Reviewing these local emissions effects jointly, both control groups appear to
corroborate the same story; the new port queuing system is associated with ele-
vated emissions. Whether contrasting local emissions to adjacent regions within
Southern California or similar coastal port regions along the US West Coast, the
system contributed to elevated CO, NO2, NOx, Ozone and particulate matter pol-
lutant concentrations. Reviewing the policy in question, there are less vessels now
98
Cargo Ships & Coastal Smog
using anchorage zones and drifting in the nearby bay area. However, given that
these vessels were beyond 20 nautical miles out at sea, it is possible that these
emissions did not strongly permeate the US coastline. Additionally, other factors
may have offset any negative contributions the policy potentially made via the
vessel congestion channel. For example, the en masse departure of these vessels
– almost reached 100 containerships at the peak of the congestion crisis – poten-
tially rekindled public interest in visiting coastal areas such as Santa Monica Pier
and Venice Beach. Furthermore, the added certainty of vessel arrival times likeli-
hood enhanced the appeal of connective modal services via truck and rail, further
intensifying emissions in the region.
For reasons such as these, broad reviews of emissions can be far too subject
to confounding factor biases. In the following subsection of this paper, I use
vessel movement data to isolate the local and global effects of the policy on vessel
emissions and mitigate any local confounding factors concerns.
4.3.3 Voyage Emission Effects
Table 4.8 details the effect of the new queue system on vessel voyage speeds.
Column 3 excludes individual vessels that may have switched ports as a result of
the implementation of this policy and acts as the main set of results. This restricts
strategic responses across vessel operators that otherwise potentially biases results
in columns 1 and 2 (Klotz and Berazneva, 2022). I find that the introduction of
this new queue system lowered travel speeds by 16.97%.13
Using each vessel’s voyage speed and their respective mass, I generate a set
of emissions across individual vessel–voyages servicing the US west coast. As
displayed in Table 4.9, my preferred subsample of international voyages yields a
10.47% reduction in total emissions following the introduction of the new queue
13For a log-level regression, I take the exponent of the coefficient and subtract 1.
99
Cargo Ships & Coastal Smog
Table 4.8: Difference-in-Difference Estimates – Voyage Speed
Nautical Miles Per Hour
(1) (2) (3)
Post–Policy 0.2206∗∗∗ 0.1828∗∗ 0.2400∗∗∗
(0.0595) (0.0776) (0.0909)
Treatment 0.0796∗∗∗ 0.0414∗ 0.1080
(0.0256) (0.0244) (0.1013)
DiD -0.2129∗∗∗ -0.1540∗∗∗ -0.1860∗∗∗
(0.0321) (0.0373) (0.0394)
Vessel–Voyage FE ✓ ✓ ✓
Year–Month FE ✓ ✓ ✓
Observations 7,055 5,665 2,354
Average Speed, Pre-Policy 18.98 19.14 19.65
R2 0.82 0.71 0.63
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering within vessel–voyage lanes of transport
service. Each observation is a distinct voyage arriving on the US west coast between Nov 2019 and Nov 2022.
Column 1 reports the broad diff-in-diff treatment effect on vessels. Column 2 excludes transshipping activity –
short subsequent journeys between US ports after their initial arrival from a foreign port of origin. Column 3 only
includes Column 2 vessel voyages that maintained the same international trade routes pre- and post- policy. Each
regression uses a logged dependent variable. To limit extreme outlier distortions, I exclude any voyages with travel
speeds less than the 25th percentile minus three times the interquartile range (75th percentile - 25th percentile)
or higher than the 75th percentile plus three times the interquartile range (Davies and Jeppesen, 2015). “Average
Speed, Pre-Policy” refers average vessel voyage travel speed of treated US ports prior to 11/11/2021.
system. Furthermore, per hour and per nautical mile emissions fell by 25.12 and
9.83 percent, respectively.14 Although I refer to CO2 emissions in my findings, the
linear relationship prescribed by Czermanski et al. (2021) between fuel use and
emissivity implies that log changes will be identical across all emission types.
I use an event study design to examine whether these policy effects are perma-
nent or transitive in nature and subset for international journeys that continued
maintaining the same service lines following the introduction of the policy. Rela-
tive to October 2021, vessels temporarily slow their transit speeds across voyages.
As displayed in Figure 4.8, the two-way fixed effects findings suggests that reduced
travel speed adjustments eased off by September 2023.
Similarly, emissions across voyages do see a level reduction that shows mean-
reversion behaviour after July 2022. These short–lived effects suggest that an
ETA-based queuing system may not yield substantial boons to the port’s efforts
14Relaxing the sample to consider all 5,628 international vessel visits yields similar results.
100
Cargo Ships & Coastal Smog
Table 4.9: Difference-in-Difference Estimates – Voyage Emissions
CO2 Emissions Em. per Hour Em. per Knot
(1) (2) (3) (4) (5)
Post–Policy 0.2929 -0.0503 -0.0895 0.1153 -0.1246
(0.2917) (0.0797) (0.0996) (0.1422) (0.0867)
Treatment 1.929∗∗∗ 0.0918∗∗ 0.1193 0.1984 0.0900
(0.3091) (0.0363) (0.0823) (0.1760) (0.0762)
DiD -1.159∗∗∗ -0.1053∗∗ -0.1107∗∗ -0.2894∗∗∗ -0.1035∗∗
(0.1411) (0.0430) (0.0536) (0.0777) (0.0457)
Vessel–Voyage FE Yes Yes Yes Yes Yes
Year–Month FE Yes Yes Yes Yes Yes
Observations 10,004 5,628 2,346 2,346 2,346
R2 0.91 0.95 0.91 0.75 0.83
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering within vessel–voyage lanes of transport
service. Each observation is a distinct voyage arriving on the US west coast between Nov 2019 and Nov 2022.
Column 1 reports the broad diff-in-diff treatment effect on vessels. Column 2 excludes transshipping activity –
short subsequent journeys between US ports after their initial arrival from a foreign port of origin. Column 3
only includes Column 2 vessel voyages that maintained the same international trade routes pre- and post- policy.
Columns 4 and 5 shift focus to per hour and per nautical mile measures of emissivity, respectively. Each regression
uses a logged dependent variable. To limit extreme outlier distortions, I exclude any voyages with hourly emissions
less than the 25th percentile minus three times the interquartile range (75th percentile - 25th percentile) or higher
than the 75th percentile plus three times the interquartile range (Davies and Jeppesen, 2015).
Figure 4.8: Event Study (TWFE) - Voyage Speed
−25 −20 −15 −10 −5 0 5 10
Time to treatment
to decarbonize long-distance maritime shipping. However, in the next section I
shift focus to the queuing leg of transit and demonstrate a substantial reduction
in local vessel emissions. When combined with the queuing stage effect, global
emission display a broad-based reduction as a result of this new system.
101
Estimate and 95% Conf. Int.
−0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4
Cargo Ships & Coastal Smog
Figure 4.9: Event Study (TWFE) - Voyage Emissions
−25 −20 −15 −10 −5 0 5 10
Time to treatment
4.3.4 Queuing Emission Effects
In this subsection, I examine changes in queuing emissions, following the introduc-
tion of the new vessel queuing system. To measure queuing emissions, I take the
subsample of vessels of international origin that maintained consistent service of
the same routes pre– and post– policy. For each vessel IMO, I identify four key
identifying features to filter for the relevant vessel movement data; IMO, MMSI,
timestamp of arrival in US waters and timestamp of arrival at a west coast US
port. I then filter the vessel position data for these key dates and unique vessel
identifying codes, arranging observations by timestamp. This allows me to map
the path of each vessel travelling through US waters. For each observation, I ob-
serve the speed (SOG), geospatial location (LAT, LON), and timestamp of the
vessel. I calculate the distance, time elapsed and rate of speed for each pinged sig-
nal of a given vessel. Using the speed and container capacity, I also determine the
associated emissions of each step in the vessels local journey to port in kilograms
(CO2, SOX , NOX , and PM2.5).
102
Estimate and 95% Conf. Int.
−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6
Cargo Ships & Coastal Smog
To represent the emissions of each arriving vessels’ queuing experience, I sum
across emission contributions for a given vessel and the calculate three indicators;
(i) total emissions, (ii) per nautical mile rate of emissions and (iii) per hour rate
of emissions. In Figure 4.10, I depict changes in emissions by treatment group
and period. Additionally, I track the duration, average movement speed and dis-
tance coverage of each individual vessel’s queuing stage. While there is a distinct
reduction in queued vessel emissions among the treated group, the distribution of
emissions among control vessels exhibits no directional drift (Figure 4.10a). The
increased variance of the control group may be attributed to the particular drop
in sample size for the period, as highlighted in Figure 4.10b.
I feature these adjustments in a difference-in-difference specification (Table
4.10) – sampling for vessels operating inbound from foreign origin ports and sub-
ject to no lane switching following the queue system’s introduction. In Column 1,
my findings suggest a 30.21% decline in vessel emissions during the queuing stage
of transit. This evidence is supportive of the position that increased certainty of
admittance timing, prior to arrival, encourages less loitering around the local port
area. Delving deeper into the effect, policy has contributed to a substantial reduc-
tion in time spent locally prior to port admittance – supportive of my proposed
mechanism of greater certainty for arriving vessels enabling less time spent wait-
ing. Since vessels are no longer idling as much and continuing to emit pollutants
into the local area, emissions per nautical mile (knot) travelled decline (Column 3)
and relatively more active vessels exhibit higher average travel speeds (Column 4).
Columns (5) and (6) address potentially higher emissions resulting from increased
exposure to the deep sea, and consequentially greater fuel consumption due to
more frequent repositioning while waiting adrift offshore. These relatively less sta-
tistically significant findings highlight possible evidence that vessels were travelling
slightly greater distances during their queuing stage, likely due to more frequent
103
Cargo Ships & Coastal Smog
Figure 4.10: CO2 emissions by hour across queued vessels
(a) Distribution of Emissions by Subsample
Control Treatment
0.6
0.4
0.2
0.0
8 10 12 14 8 10 12 14
Logged Kilograms
post- pre-
(b) Number of Vessel-Voyages by Subsample
Control Treatment
75
50
25
0
8 10 12 14 8 10 12 14
Logged Kilograms
post- pre-
repositioning. Controlling for the shorter wait times, vessels’ emissions per hour
while queuing also rose by 11%, suggesting a greater degree of exertion following
the blanket requirement to wait 150 nautical miles out at sea. Despite these de-
mands though, it appears that improved certainty of queue positions outweights
the slightly worse waiting experience that vessels were exposed to.
104
Vessel Count Sub-Sample Share (%)
Cargo Ships & Coastal Smog
Table 4.10: Difference-in-Difference Estimates – Queuing Emissions
CO2 Emissions Duration Speed Em. per Knot Distance Em. per Hour
(1) (2) (3) (4) (5) (6)
Post–Period -0.3114 -0.3302 0.0325 0.1116 -0.4231∗ 0.0187
(0.2384) (0.2663) (0.1623) (0.1685) (0.2418) (0.1068)
Treatment 1.299∗∗∗ 1.470∗∗∗ -0.6342∗∗∗ -0.4931∗∗ 1.792∗∗∗ -0.1708
(0.2067) (0.2370) (0.1051) (0.2505) (0.3774) (0.1252)
DiD -0.3021∗∗∗ -0.4119∗∗∗ 0.4806∗∗∗ -0.4508∗∗∗ 0.1487 0.1098∗
(0.0820) (0.1183) (0.0940) (0.0871) (0.1028) (0.0658)
Vessel–Voyage FE ✓ ✓ ✓ ✓ ✓ ✓
Year–Month FE ✓ ✓ ✓ ✓ ✓ ✓
Observations 2,111 2,111 2,111 2,111 2,111 2,111
R2 0.59 0.57 0.59 0.52 0.45 0.68
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering within vessel–voyage lanes of transport
service. Each observation is a distinct queuing experience of a vessel arriving on the US west coast between Nov
2019 and Nov 2022. I filter only for vessel voyages that maintained the same international trade routes pre-
and post- policy and drop any observations that lack matching vessel movement data. To limit extreme outlier
distortions, I exclude any voyages with emissions less than the 25th percentile minus three times the interquartile
range (75th percentile - 25th percentile) or higher than the 75th percentile plus three times the interquartile range
(Davies and Jeppesen, 2015).
Given these estimates, I next weigh the contributions of reduced voyage and
queuing emissions together and calculate the policy’s overall global effect.
4.3.5 Global Emission Effects
In this section, I separately identify the voyage and queuing legs of a given vessel
transit. As detailed in Figure 4.11, queuing represents a substantial share of the
distance, duration, and associated emissions of international shipping.
Figure 4.11: Queuing Share of Global Containership Transit for US West Coast
Control Treatment
60
40
20
2020 2021 2022 2020 2021 2022
Year-Month
Total Emissions Transit Distance Transit Time
105
Share (%)
Cargo Ships & Coastal Smog
Prior to the introduction of the new queuing system, both port groups experi-
enced a doubling of vessel transit time allocated to the queuing experience across
a two year time span. The introduction of ETA-based queue positions among
treated ports – issued upon departure from a prior port of laden – was followed by
a severe decline in transport operators’ time allocated towards queuing. Unlike the
control group, upward trends in the share of travel distance attributed to the queu-
ing stage halted. Given the shorter time spells that vessels remain within the port
area, the queuing share of global emission contributions has also declined. Mean-
while, the control ports of Seattle, Tacoma and Oakland have largely maintained
prior trends.
Referring to Equation 4.2, I assess the broad impact of the policy on global
emissions by combining each vessels’ voyage and queuing stages of transit. I find
that the introduction of the new queuing system contributed to an 11.84% reduc-
tion in global vessel emissions servicing the US west coast.
Table 4.11: Difference-in-Difference Estimates – Global Emissions
CO2 Emissions Em. per Hour Em. per Knot
(1) (2) (3)
Post–Period -0.2081∗∗ 0.0891 -0.0954
(0.0902) (0.1103) (0.0909)
Treatment 0.2084∗∗∗ -0.1481 -0.1223
(0.0744) (0.1607) (0.0865)
DiD -0.1184∗∗∗ -0.0372 -0.1560∗∗∗
(0.0447) (0.0682) (0.0481)
Vessel–Voyage FE ✓ ✓ ✓
Year–Month FE ✓ ✓ ✓
Observations 2,111 2,111 2,111
R2 0.90 0.80 0.72
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering within vessel–voyage lanes of transport
service. Each observation is a distinct global transit experience of a vessel arriving on the US west coast between
Nov 2019 and Nov 2022. I filter only for vessel voyages that maintained the same international trade routes pre-
and post- policy and drop any observations that lack matching vessel movement data. To limit extreme outlier
distortions, I exclude any voyages with emissions less than the 25th percentile minus three times the interquartile
range (75th percentile - 25th percentile) or higher than the 75th percentile plus three times the interquartile range
(Davies and Jeppesen, 2015).
Furthermore, emissions per nautical mile travelled have declined by 15.6%. I do
not detect a significant decline in emissions per hour due to increases during queu-
106
Cargo Ships & Coastal Smog
ing offsetting reductions associated with voyage. While vessels have temporarily
slowed down while at sea, the opposing effect of being more active during the queu-
ing experience and potentially having to readjust position more frequently while
waiting have added a partial upward effect to any broad adjustments in emissions
per hour. In other words, vessels have repurposed available time that would have
otherwise been allocated to the idling experience towards their voyages.
4.4 Discussion & Conclusion
This study focuses on the medium run environmental effects of an ETA-based
vessel queuing system. My results show that the emissivity of containership ves-
sels destined for San Pedro Bay declined, when controlling for prevailing trends
along the other major container ports of the US west coast. These new conditions
allowed transport operators to learn their queue position prior to arriving in the
local port area. As a result, they could adjust speeds and minimize the amount of
idling they would perform in the local port vicinity. This contributed to a 41.2%
reduction in the duration of queuing times at port and a 30.2% reduction in total
local emissions. However, local monitors adjacent to the port areas show signs of
a broader elevation in emissions among post-period treated ports. These patterns
are possibly a symptom of increased commercial activity in the port areas, follow-
ing the policy’s success in relieving the local logjam in container movements. This
and the added certainty of vessel arrivals would likely bolster multimodal trans-
port activity. Additionally, the increased appeal of the coastal area, following the
removal of unsightly containership queues, may have also boosted local emissions
in the area. In future work, I intend to extend this analysis to incorporate endoge-
nous queuing activity along the US west coast, using weather-related instrumental
variables to project against containership traffic within the San Pedro Bay area.
107
Cargo Ships & Coastal Smog
Additionally, my findings suggest that the new queuing system encouraged
a slowdown in vessels bound for the ports of Los Angeles and Long Beach. In
total, emissions were reduced by approximately 10.48% and travel speeds fell by
16.97%. Coupled with the event study design of this analysis, I posit that these
adjustments are transitive in nature and that relative slowdowns in across vessels
became statistically insignificant in less than a years time.
To document the relative importance of these two patterns, I combine the
international and local effects of the new queuing system to assess global implica-
tions. Firstly, I document that despite the small geographic region that queuing
is contained within, this activity represents a substantial share of the lifespan of
a given vessel–voyage. In duration, it can range between a 20–60% year–month
average share of total international transit times between ports. In emissions and
distance, year–month averages range from 5–40%. The queuing stage is therefore a
very important aspect of containership voyages that carries great weight in efforts
to mitigate emissions. Accounting for both legs of transit, I find that the new
queuing system of San Pedro Bay is attributable to an 11.8% reduction in total
vessel emissions. Vessels in fact speed up during the queuing stage, due to less
idling. This combined with the temporary slowdown in voyage speeds leads to a
muted effect on average emissions per hour. Despite this fact, a substantial reduc-
tion in wait times occurs, which has led to emissions per nautical mile traversed
in service of San Pedro Bay falling by 15.6%.
Coupled together, these findings suggest that an ETA-based queuing system
offers a substantial decarbonization boon to ports exposed to lengthy wait time
and queuing processes. For small or mid-tier ports less likely to exhibit lengthy
idling times, further study is required to establish whether the logistical costs of
maintaining an ETA-based queuing system would be environmentally beneficial.
108
Cargo Ships & Coastal Smog
Bibliography
Brancaccio, G., M. Kalouptsidi, and T. Papageorgiou (2020). Geography, trans-
portation, and endogenous trade costs. Econometrica 88 (2), 657–691.
Capaldo, K., J. J. Corbett, P. Kasibhatla, P. Fischbeck, and S. N. Pandis (1999).
Effects of ship emissions on sulphur cycling and radiative climate forcing over
the ocean. Nature 400 (6746), 743–746.
Corbett, J. J., H. Wang, and J. J. Winebrake (2009). The effectiveness and costs
of speed reductions on emissions from international shipping. Transportation
Research Part D: Transport and Environment 14 (8), 593–598.
Corbett, J. J., J. J. Winebrake, E. H. Green, P. Kasibhatla, V. Eyring, and
A. Lauer (2007). Mortality from Ship Emissions: A Global Assessment. Envi-
ronmental Science & Technology 41 (24), 8512–8518. PMID: 18200887.
Cristea, A., D. Hummels, L. Puzzello, and M. Avetisyan (2013). Trade and the
greenhouse gas emissions from international freight transport. Journal of Envi-
ronmental Economics and Management 65 (1), 153–173.
Czermanski, E., G. T. Cirella, A. Oniszczuk-Jastrzabek, B. Pawlowska, and
T. Notteboom (2021). An Energy Consumption Approach to Estimate Air
Emission Reductions in Container Shipping. Energies 14 (2).
Davies, R. and T. Jeppesen (2015). Export mode, firm heterogeneity, and
source country characteristics. Review of World Economics (Weltwirtschaftliches
Archiv) 151 (2), 169–195.
Ganapati, S. and W. F. Wong (2023). How Far Goods Travel: Global Transport
and Supply Chains from 1965–2020. Journal of Economic Perspectives 37 (3),
3–30.
109
Cargo Ships & Coastal Smog
Gillingham, K. and P. Huang (2021). Racial Disparities in the Health Effects from
Air Pollution: Evidence from Ports. NBER Working Papers 29108, National
Bureau of Economic Research, Inc.
Hansen-Lewis, J. and M. M. Marcus (2022). Uncharted waters: Effects of maritime
emission regulation. Technical report, National Bureau of Economic Research.
Heiland, I., A. Moxnes, K. H. Ulltveit-Moe, and Y. Zi (2019). Trade From Space:
Shipping Networks and The Global Implications of Local Shocks. CEPR Dis-
cussion Papers 14193, C.E.P.R. Discussion Papers.
Hulskotte, J. and H. Denier van der Gon (2010). Fuel consumption and associ-
ated emissions from seagoing ships at berth derived from an on-board survey.
Atmospheric Environment 44 (9), 1229–1236.
IMO (2021). Fourth Greenhouse Gas Study 2020. Full report, International Mar-
itime Organization.
Jalkanen, J.-P., L. Johansson, J. Kukkonen, A. Brink, J. Kalli, and T. Stipa (2012).
Extension of an assessment model of ship traffic exhaust emissions for particulate
matter and carbon monoxide. Atmospheric Chemistry and Physics 12 (5), 2641–
2659.
Ju, Y. and C. A. Hargreaves (2021). The impact of shipping co2 emissions from
marine traffic in western singapore straits during covid-19. Science of The Total
Environment 789, 148063.
Kinney, P. L., M. G. Gichuru, N. Volavka-Close, N. Ngo, P. K. Ndiba, A. Law,
A. Gachanja, S. M. Gaita, S. N. Chillrud, and E. Sclar (2011). Traffic impacts
on PM2.5 air quality in Nairobi, Kenya. Environmental Science & Policy 14 (4),
369–378.
110
Cargo Ships & Coastal Smog
Klotz, R. and J. Berazneva (2022). Local Standards, Behavioral Adjustments, and
Welfare: Evaluating California’s Ocean-Going Vessel Fuel Rule. Journal of the
Association of Environmental and Resource Economists 9 (3), 383 – 424.
Liu, H., M. Fu, X. Jin, Y. Shang, D. Shindell, G. Faluvegi, C. Shindell, and K. He
(2016, November). Health and climate impacts of ocean-going vessels in East
Asia. Nature Climate Change 6 (11), 1037–1041.
Lugovskyy, V., A. Skiba, and D. Terner (2023). Unintended Consequences of En-
vironmental Regulation of Maritime Shipping: Carbon Leakage to Air Shipping.
In SSRN Electronic Journal.
Mitchell, R. C. and R. Carson (1986). Property Rights, Protest, and the Siting of
Hazardous Waste Facilities. American Economic Review 76 (2), 285–90.
Morehouse, J. M. and E. A. Rubin (2021). Downwind and Out: The Strategic
Dispersion of Power Plants and their Pollution. In SSRN Electronic Journal.
Ngo, N. S., M. Gatari, B. Yan, S. N. Chillrud, K. Bouhamam, and P. L. Kinney
(2015). Occupational exposure to roadway emissions and inside informal set-
tlements in sub-Saharan Africa: A pilot study in Nairobi, Kenya. Atmospheric
Environment 111, 179–184.
Prochazka, V., R. Adland, and F.-C. Wolff (2019). Contracting decisions in the
crude oil transportation market: Evidence from fixtures matched with AIS data.
Transportation Research Part A: Policy and Practice 130 (C), 37–53.
Rivera, N. (2021). Air quality warnings and temporary driving bans: Evidence
from air pollution, car trips, and mass-transit ridership in Santiago. Journal of
Environmental Economics and Management 108 (C), S0095069621000371.
Rodrigue, J.-P. (2020, May). The Geography of Transport Systems. Fifth edition.
— Abingdon, Oxon ; New York, NY : Routledge, 2020.: Routledge.
111
Cargo Ships & Coastal Smog
Schwarzkopf, D. A., R. Petrik, V. Matthias, M. Quante, E. Majamäki, and J.-P.
Jalkanen (2021). A ship emission modeling system with scenario capabilities.
Atmospheric Environment: X 12, 100132.
Shapiro, J. (2016). Trade Costs, CO2, and the Environment. American Economic
Journal: Economic Policy 8 (4), 220–54.
Tumbarello, P., S. Arslanalp, and M. Marini (2019). Big Data on Vessel Traf-
fic: Nowcasting Trade Flows in Real Time. IMF Working Papers 2019/275,
International Monetary Fund.
Walsh, C. and A. Bows (2012). Size matters: Exploring the importance of vessel
characteristics to inform estimates of shipping emissions. Applied Energy 98 (C),
128–137.
Wong, W. F. (2022). The Round Trip Effect: Endogenous Transport Costs and
International Trade. American Economic Journal: Applied Economics 14 (4),
127–66.
Wong, W. F. and S. Fuchs (2022). Multimodal Transport Networks. FRB Atlanta
Working Paper 2022-13, Federal Reserve Bank of Atlanta.
Zhang, Y., S. D. Eastham, A. K. Lau, J. C. Fung, and N. E. Selin (2021, aug).
Global air quality and health impacts of domestic and international shipping.
Environmental Research Letters 16 (8), 084055.
Zou, E. Y. (2021). Unwatched Pollution: The Effect of Intermittent Monitoring
on Air Quality. American Economic Review 111 (7), 2101–26.
112
Chapter 5
Concluding Remarks
113
Cargo Ships & Coastal Smog
In this dissertation, I have demonstrated the importance of accounting for the
logistical practices of transport intermediaries in representing endogenous trans-
port costs, used historical port call data to study the evolution containership tech-
nology, and evaluated the environmental implications of vessel queuing systems.
In Chapter 2, I focused on the empty container repositioning practice faced
by the United States and its respective trade partners. To pursue this time, I
used novel port-level US container traffic data for January 2012 to December 2021
and combined the data with US Census records of port–level monthly container-
ized imports and exports. I first documented that the US maintains a nationally
balanced exchange of transport equipment with the rest of the world, well–timed
to support monthly variation in the asymmetries of the US trade deficit. I also
established a micro-founded model of transport equipment repositioning, which
involved embedding logistical decision making across intermediary container ship-
ping firms into a roundtrip model of transportation and trade. Lastly, I conduct a
counterfactual analysis of policy implications for the recently passed Ocean Ship-
ping Reform Act of 2022, which seeks to limit empty container outflows in favor of
stimulating US exports. I find that restricting container repositioning stifles the
vessel capacity of a given round trip and reduces bilateral trade activity, leading to
lower gains from trade. The policy also behaves in a particularly targeted manner,
with East Asian trade parterns – most reliant on empty container returns – facing
the greatest degrees of loss in exports to the US.
In Chapter 3, we examined how adjustments in the composition of vessels ser-
vicing US containerized trade contributed container handling outcomes at the port
level. Our depiction of container shipping relies on four key data sets, which al-
lowed us to observe US activity between January 1977 and December 2023. We
find that transport operators initially met the increasing volume of demand of
trade by simply throwing more vessels at the problem. We documented a five-fold
114
Cargo Ships & Coastal Smog
increase in the average size of vessels for 1977–2023, while in comparison the num-
ber of visiting vessels increased three-fold. Upon saturating US ports with vessels
in 2012, we posit that tastes for shipbuilding increasingly shifted towards the up-
grading of vessel technology and expansion of vast hull sizes. These increasingly
larger vessels are associated with longer dwell times, but upon accounting for the
scale of containers they are capable of carrying, we identified efficiencies of scale in
their use in handling transport equipment. To examine how accumulating traffic
and existing congestion influences the outcomes of incumbent visiting vessels at
the port, we developed an instrumental variable that acts as a demand-shifter for
port traffic. We found that increases in port traffic contribute to negative spillovers
for surrounding vessels. We identified these two countervailing effects on container
handling efficiency as having contributed to dwell times across ports stabilizing
at a 2.4 day average duration across the past 50 years of containerized transport,
despite the novel innovations made in individual shipping capacities.
In Chapter 4, I studied the environmental effects of a new containership queu-
ing system introduced in San Pedro Bay that issues vessel queue positions based
on an estimated time of arrival (ETA). The prior system had required vessels to
join these ports’ queues by entering within 20–25 nautical miles of the port vicin-
ity. This effectively encouraged a system of “rat races”, in which vessels were
collectively incentivized to speed against one another to endogenously secure ear-
lier queue positions. I find that while this slowdown lowered vessel emissions at
the voyage stage of transit, the effect was transitive in nature. The substantial
reduction in global containership emissions that I document was driven instead
by adjustments in behaviour at the queuing stage of transit. By guaranteeing a
queue position upon departure from a prior port of origin, transport operators
were able to minimize idling time and contribute to a 30.2% reduction in total
global emissions.
115
Cargo Ships & Coastal Smog
The results from my dissertation emphasize the importance of maintaining a
multidisciplinary approach, which has allows for the dovetailing of studies in mar-
itime transport, trade and environmental economics. In each of my substantive
chapters, I leverage use of novel data sets in order to answer key questions of the
prevailing literature. For example, in a recent review of academic research of mar-
itime shipping Ardelean et al. (2022) identifies “...gaps on the policy questions that
need further research, such as regulation of liner shipping, a better understanding
of the determinants of port efficiency, and the effects of environmental regulation
on the transportation sector.” My array of essays makes a serendipitous contri-
bution to each one of these three respective topics of interest by evaluating the
consequences of physical transport equipment in liner shipping, providing a histor-
ical decomposition of port congestion via vessel technology and port efficiency, and
developing an event study of how queuing regulations impact the environmental
outcomes of maritime transport. To continue benefiting from global trade, we must
better understand the underlying supply–side factors that enable the increasingly
large volumes of commerce we aspire to exchange. This requires a deep under-
standing of how the state and logistical use of transport equipment contributes to
trade outcomes, whether that be the ship bearing our goods, the containers they
are laden with, or the port infrastructure receiving both items jointly.
116
Unconventional Protectionism
Appendix I
A1 (I.) General Equilibrium with Homogeneous Input Prices
The assumption of common input prices across loaded and empty containers is gen-
eralizing restriction that yields zero freight rates for transport services originating
from net importer countries.Consider equation (3)
max π↔ =Tijlij + Tjilji − cijlij − cjilji − r↔(eij + eji) s.t. lij + eij = lji + eji
lij ,lji,eij ,eji ij ij
I adjust this specification to a more general form which sets all container input
prices equal to a route specific cost term {cij, cji, r↔} = c↔. Consider Case II in
ij ij
which a trade imbalance exists between countries i and j such that lij = lji + eji
and eij = 0. Under these circumstances, imbalance trade and balanced container
flows imply a zero freight rate on route ji.
max π↔ =Tijlij + Tjilji − c↔lij − c↔lji − c↔(eji) s.t. lij = lji + eji
lij ,lji,eij ,eji ij ij ij ij
=Tijlij + Tjilji − c↔(lij + lji + lij − lji)
ij
FOC
∂π↔
ij
=0 =⇒ Tij = 2cij
∂lij
∂π↔
ij
=0 =⇒ Tji = 0
∂lji
Similarly to Behrens and Picard (2011), I find that both bilateral freight rates of a
given round trip route are non-zero only when shipments of loaded containers are
balanced. In practice, incoming loaded containers being converted into an input
for outgoing transport services involve more time, weight and cleaning relative
to incoming empty containers. This suggests higher marginal costs of revenue-
117
Unconventional Protectionism
generating loaded container inputs relative to using inbound empties to service
outbound transport services.
Upon acknowledging these underlying differences in handling costs between
empty and loaded containers through heterogeneous input prices, within route,
the general equilibrium model is capable of generating positive freight rates for
↔
both sides of an imbalanced round trip trade on ij. I use heterogeneous input
prices to generate empty container flows in conjunction with positive bilateral
tariff rates.
A1 (II.) Balanced Trade Scenario
The perfectly competitive transport operator will yield prices where the marginal
benefit of an additional loaded container transport is equal to the marginal cost.
Using the implied lji from equation (4), and setting these quantities equal to one
another, we arrive at a case of two equations and two unknowns for {lij, Tij}.
Setting these equations equal to one another allows for freight rates to be solved.
( )−ϵ ( )−ϵ
ϵ 1 −ϵ ϵ 1
( )−ϵ
(wiτij + Tij) = wjτjic↔ + c↔ − T− ijϵ 1 aij ϵ− 1 aji ij ij
1 1
( (wiτij +)Tij) = ((wjτji)+ 2c↔ − Tij)aij aji ij1 1 1 1
Tij = ( 2c↔) − (wiτij) + (wjτji)aij + aji aji ij aij aji
(aij + aji)Tij = aij 2c↔ − aji (wiτij) + aij (wjτji)
ij
1 1 1
T ∗ij = a (2c↔)− a (wji ij iτij) + a (w τ )1 + ij 1 + 1 + ji j ji
aij aji aij
(5.1)
With freight rates expressed in terms of exogenous variables, solving for p∗ij is
relatively straightforward and simplifies solving for l∗ij.
118
Unconventional Protectionism
p∗ij = wiτ
∗
ij + Tij
1 − 1 1= wiτij + a (2c↔) a (wiτij) + a (w τ )
1 + ji
j ji
ij 1 + ij 1 + ji
aij aji aij
a
1 + ij1 − 1a 1
= (2c ) + jia ↔ (w τ ) + (w τji ( a i ij a j ji)1 + ij 1 + ij ) 1 + jiaij aji aij
p∗
1
ij = a 2c↔ + wiτij + wjτji (5.2)1 + ji ij
aij
To solve for l∗ij, plug T
∗
ij into equation (4).
( )−ϵ
l∗
ϵ 1 ( )−ϵ
ij = ( ) (w τ + T
∗
i ij
ϵ− 1 a ijij )
−ϵ −ϵ
ϵ 1 1 − 1 1= (wiτij + a (2c↔) a (wiτij) + a (wjτji)(ϵ− 1 aij) (1 +
ji ij 1 + ij 1 + ji
aij aji) aij
−ϵ −ϵ
ϵ 1 1 )
=
− a
2c↔ + wiτji ij + wjτji (5.3)ϵ 1 aij 1 + ijaij
The equilibrium value of trade is simply price times quantity:
( ) ( ( ))−ϵ 1−ϵ
∗ ϵ 1 1Xij = a 2c↔ + wiτij + wjτji (5.4)ϵ− 1 aij 1 + ji ijaij
A1 (III.) Container Traffic Sample
In Table 5.1, each row reports a given year’s number of contributing ports, the
total number of loaded and empty container units handled by the set of contribut-
ing ports, the total number of loaded and empty container units handled at the
national level, and the sample’s share of national throughput.
119
Unconventional Protectionism
Table 5.1: Sample Representation - US Total Container Throughput
Year Number of Ports Sample TEU National TEU % of National
2003 8 21,150,609 32,689,484 64.70
2004 8 23,357,414 34,901,628 66.92
2005 8 25,826,230 38,497,839 67.08
2006 8 27,661,831 40,896,742 67.64
2007 8 27,797,684 44,839,390 61.99
2008 9 26,652,498 42,411,770 62.84
2009 10 23,169,814 37,353,575 62.03
2010 10 27,122,000 42,031,000 64.53
2011 11 29,181,883 42,550,784 68.58
2012 12 35,350,843 43,538,254 81.19
2013 12 35,937,976 44,340,866 81.05
2014 12 37,548,916 46,233,010 81.22
2015 13 40,501,360 47,886,446 84.58
2016 13 41,021,434 48,436,472 84.69
2017 13 44,209,298 52,132,844 84.80
2018 13 46,619,407 54,776,341 85.11
2019 13 47,064,791 55,518,878 84.77
2020 13 46,555,563 54,963,689 84.70
2021 13 53,748,362 62,044,503 86.63
Source: National thruflows use ‘Container port throughput, annual’ from UNCTAD.
A1 (IV.) Unilateral and Port-Specific Results
In this section of the Appendix, I address alternative specifications which mirror
those proposed in the main body of this study. Figure 5.2 depicts the co-movement
between empty container units and trade flows travelling in the same direction for
a given year-month, between the US and RoW. I find no distinct relationship,
suggesting that only opposite leg variation in trade flows stimulate systematic
adjustments to empty container repositioning. This opposite-leg relationship be-
tween trade flows and corresponding empty container unit adjustments is reflected
in Table 3.2.
Tables 5.3 & 5.4 and Figure 5.1 mirror national regressions featured in the main
120
Unconventional Protectionism
Table 5.2: Empty Container Elasticity with Respect to Trade Flows (kg)
Dependent Variable: Empty Container Flows (TEU)
ln(Outbound) ln(Inbound)
Model: (1) (2) (3) (4)
ln(Inbound Trade) 1.582∗∗∗ -0.0881
(0.1152) (0.2576)
ln(Outbound Trade) 0.0033 0.6352∗∗∗
(0.1292) (0.1770)
n-obs 120 120 120 120
Within R2 0.65 2.89× 10−6 0.002 0.13
Clustered (month) standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. US empty container
flows are regressed on US containerized trade flows, expressed in terms of kilograms. For example, a one percent
increase in the weight of ‘Inbound Trade’ is associated with a 1.58% rise in outbound empty container flows. I
use month and year fixed effects to control for influences of the US business cycle and seasonality.
body of the paper. Generally these findings are weaker, which is partly due to ports
not individually maintaining balanced container flows. Only in conjunction with
other ports does the US maintain nationally balanced container flows and response
relationships between prevailing trade flows and opposite-end empty container
movements.
Table 5.3: (Ports) Trade Flow Ratio & Empty Shares
Dependent Variable: Empty Container Share of Total Flows
Outbound Inbound
Model: (1) (2) (3) (4)
Export/Import (USD) -0.0847∗
(0.0412)
Export/Import (kg) -0.0582∗
(0.0278)
Import/Export (USD) -0.0063∗
(0.0033)
Import/Export (kg) -0.0124∗∗∗
(0.0027)
Mean Dep. Var 34.6% 15.27%
Mean Regressor 0.496 0.901 2.865 1.499
n-obs 1,440 1,440 1,440 1,440
Within R2 0.03 0.06 0.01 0.02
Clustered (port) standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. Examines variation empty
containers as a share of total container outflows, given variation in the skewedness of the trade balance. I use
month and year fixed effects to control for influences of the US business cycle and seasonality.
121
Unconventional Protectionism
Table 5.4: (Ports) Empty Container Elasticity w.r.t. Opposite-Direction Trade
Flows
Dependent Variable: Empty Container Flows (TEU)
Outbound Inbound
Model: (1) (2) (3) (4)
ln(Imports, USD) 0.6218∗∗∗
(0.1256)
ln(Imports, kg) 0.3348∗∗
(0.1339)
ln(Exports, USD) 0.4949∗
(0.2278)
ln(Exports, kg) 0.3210∗
(0.1464)
n-obs 1,440 1,440 1,440 1,440
Within R2 0.064 0.044 0.01 0.005
Clustered (port) standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. Each variable is log-
transformed. The regression results portray the elasticity of total US empty container flows with respect to
opposite-direction US containerized trade flows expressed in terms of deflated USD (value) and by total weight
(kilograms). All models include port-year, port-month and year-month fixed effects.
Figure 5.1: Balanced Port Container Flows
1.2
1.0
0.8
0.6
0.4
0.2
Time Interval
Clustered (port) standard-errors in parentheses. Both the dependent variable and regressor are log-transformed.
Total inbound containers are reported across a balanced panel of 12 US ports and represent both loaded and
empty containers, is regressed on total outbound containers for these same set of ports. Sums are taken across
windows of varying lengths of time, ranging from bilateral exchanges within a single month to exchanges across
12 month backward sums. All models include port-year, port-month and year-month fixed effects.
122
Change in Total Inbound Container Units (%)
1 month
2 months
3 months
4 months
5 months
6 months
7 months
8 months
9 months
10 months
11 months
12 months
Unconventional Protectionism
A1 (V.) US-RoW Model Results
Solution Method and Model Calibration
To establish a baseline set of exogenous parameters, I first calibrate a select
subset of exogenous parameters and then estimate the remaining set of unknown
model primitives. For a given ij round trip con(tainerized shipping route, th)e set
of unknown exogenous parameters ρ is equal to aij, aji, wi, wj, τij, τji, c↔, r↔ and
ij ij
the elasticity of substitution measure is represented by ϵ.
The wage-tariff product wiτij is a component of tradeable good prices featured
in Section 3. I use an OECD index of monthly manufacturing income growth
rates, the International Labor Organization (ILO) annual measure of monthly
manufacturing income levels, and UNCTAD Trade Analysis Information System
(TRAINS) database on effective manufacturing goods’ tariff rates, all of which
are reported across a subset of key US trade partners. 1 I deflate these measures
using the Bureau of Labor Statistics Consumer Price Index for all urban consumers,
which considers all final good items less food and energy, averaged across major
US cities.2 I focus primarily on statistics associated with manufacturing due to its
high share of overall containerized goods flows. For more of an elaboration on the
calibration of wiτij, see Appendix IV. Lastly, I use an estimate of price elasticity
of demand provided by Wong (2022) and specific to containerized trade, where
ϵ̂ = 20.95 is assumed to be common across individual trade routes.
Given calibrated estimates of real wage levels, tariff rates(and the price e)lasticity
of demand, the remaining four unknown parameters, ρ̃ = aij, aji, c↔, r↔ can be
ij ij
identified via a Generalized Method of Moments (GMM) approach. I minimize
1Upon establishing a login for http://wits.worldbank.org/, select ‘Advanced Query’ and then
the ’Tariff and Trade Analysis’ subsection. I use the SITC 4 product group labelled ‘manufac-
tures’ and the
2U.S. Bureau of Labor Statistics, Consumer Price Index for All Urban Consumers: All Items
Less Food and Energy in U.S. City Average [CPILFESL], retrieved from FRED, Federal Reserve
Bank of St. Louis; https://fred.stlouisfed.org/series/CPILFESL, November 1, 2022.
123
Unconventional Protectionism
the object function,
R = dist′ × W̄ × dist, (5.5)
where dist represents the log difference in vectors of observed and model-guess
trade outcomes, log(Y data) − log(Y G) and W̄ is a weight matrix that assists in
speeding the identification of ρ̃.3 I use observables from 2017 to estimate these
parameters of underlying long-run primitives of containerized trade. This decision
allows me to avoid any complications or concerns that the use of data from the
proceeding China-US trade war, COVID-19 pandemic and port congestion saga
could introduce.
Table 5.5: Key Parameters, 2017
aij aji c↔ r↔
ij ij
65,972 32,978 20,770 8,929
I provide four means of assessing model fit for this baseline scenario of the
counterfactual exercise; (1) referring to Table 5.5, the difference in preference
parameters attributes greater demand towards US imports relative to US exports,
which is reflective of the existing import-export ratio for 2017; (2) using marginal
costs of handling loaded, (c↔), and empty container flows, r↔, the implied freight
ij ij
rates suggested these costs are greater for the portion of US round trips that feature
a full set of loaded containers, which is reflective of freight rate asymmetries under
imbalanced trade (Hummels et al., 2009); (3) the empty container redistribution
share of container fleet managing costs is 11%, which places it relatively close to
15% reported by Notteboom et al. (2022); (4) baseline scenario empty container
3For each US trade partner, a vector of four observables are used Y data = (lij , lji, Xij , Xji).
From left to right, these variables represent loaded container inflows, loaded container outflows,
containerized imports, and containerized exports between the US and that respective trade part-
ner.
124
Unconventional Protectionism
outflows each year of 2012 to 2021 are 99% correlated with untargeted observed
empty outflows.
Counterfactual Scenarios
I consider two cases of restrictions to transport equipment use by the US poli-
cymaker, where the expressed goal is to discourage empty container redistribution
in favor of stimulating US exports. In each case, restrictions are implemented
through a per-unit tax on empty outflows from the US, which increases marginal
costs to (1 + γ) r↔. The tax rate, γ, is configured to target a specific ECO quota,
ij
represented by Ēji, the maximum share of empties as a percentage of total con-
tainer outflows. I establish two scenarios which demonstrate how sensitive trade
outcomes are to variation in the availability of empty container equipment.
1. In the case of a moderate policy response, the US policymaker set a tax
on empty input costs of γmod, which targets the historical average of empty
container share of container net outflows, Ēji = 0.3. This scenario represents
a case in which policymakers are content with the former status quo of the
empty container redistribution problem.
2. In this second scenario, I consider a case in which policymakers set a suffi-
ciently high tax of γext, which eliminates empty container outflows from the
US by establishing an extreme quota of Ēji = 0. This second case allows me
to quantify the contribution the empty container redistribution problem to
variety of US trade outcome variables.
In the next section I outline how this unconventional form of trade policy
backfires in each of these exercises, relative to the baseline scenario of γ = 0, via
the round trip effect.
Results
125
Unconventional Protectionism
The targeting of ECO quotas, achieved through per-unit taxes on empty con-
tainer unit outflows, reduces the scale of the empty redistribution problem and
lowers overall round trip service capacity. Reduced transport capacity yields de-
↔
bilitating effects on the opposite leg of a given ij trade route.
As displayed in column 2 of Table 5.6, a moderate ECO quota contributes
to a one-third decline in the volume empty container redistribution problem. If
focusing only on this outbound leg of US round trip transport, the changes ap-
pear positive from the policymaker perspective. Relative to the baseline scenario,
US containerized exports increase by 12.5% in real value as transport operators
substitute away from relocating empties and towards servicing additional loaded
container units. The US containerized trade deficit, represented by the import-
export ratio, also declines by 21.5%. While the combination of these two findings
would likely signal a positive outlook for similar policies of transport equipment
restrictions, this outflow perspective alone would ignore malaise effects observed
on the opposite leg of a given round trip.
On the opposite leg, US trade partners now face a freight rate which includes
a higher cost of redistributing empties back for round trip transport service pro-
visions. The equilibrium quantity of container units declines, which represents a
reduction in the transport capacity for containerized transport services for the US.
As a result of government intervention on export routes, the opposite leg of trade
exhibits the round trip effect where available capacity declines by 6.1% and import
prices rise by 0.3%. When combined, this contributes to a 5.8% reduction in the
real value of US imports. The gross values of total imports and exports combined
declines by 2% relative to the baseline scenario, suggesting an overall reduction in
trade activity. ( )
In the extreme quota case Ēji = 0 , the backfiring of this policy has far more
dramatic effects on bilateral trade flows. The value of imports fall by 16% while
126
Unconventional Protectionism
Table 5.6: National Counterfactual Outcomes
US Measures (2017) %∆YĒ %∆Yji=0.3 Ēji=0
Import Value with RoW -5.82% -15.96%
Export Value with RoW 12.46% 40.96%
Loaded Container Inflows from RoW -6.11% -16.68%
Loaded Container Outflows to RoW 13.13% 43.41%
Import-Export Ratio -21.48% -40.38%
Empty Share of US container outflows -16.26% -100.00%
Empty Container Outflows from US -32.78% -100.00%
export activity grows by 41%. The US trade deficit narrows, reflected by a 40 per
cent drop in the import-export ratio. Despite empty containers no longer featuring
on round trip routes, the US still maintains net importer status with imports being
2 fold that of exports (see Figure 5.3). The gross value of trade flows declines by
3% under these circumstances and the ocean-borne capacity of round trip trade
servicing the US declines by 16.7%.
Figure 5.2: Counterfactual Outcomes by Empty Outflow Tax, 2017
Empty Share of US Outflows US Import−Export Ratio
40
3.0
30
20
2.5
10
0 2.0
0 1 2 3 0 1 2 3
Empty Outflow Tax, % Empty Outflow Tax, %
Note: The required rates of tax for moderate and extreme quota outcomes are 1.1 and 3.1 percent rates, re-
spectively. The empty share of US container outflows declines concavely with respect to an empties tax. The
Import-Export ratio, although more than 3.5 in the baseline scenario, declines in moderate and extreme counter-
factual cases to ratios of 2.8 and 2, respectively.
127
%
Unconventional Protectionism
These results highlight that if policymakers focus only on the immediate goal of
stimulating exports, without acknowledging the market response this would have
on round trip trade patterns, they may underestimate the costs these policies are
likely to have for the general public. Specifically, lower levels of imports at more
expensive rates would need to also be taken into account. The combination of
the exports increases and import declines, due to the round trip effect, worsens a
country’s overall level of trade participation, which limits the gains to trade.
A1 (VI.) Loading Factor Estimates & Container Flow Diag-
nostics
While allowing commodity-specific loading factors to vary by directional flow is
one decision worth considering, I have also included aggregations of particularly
low volume commodity types to observe how costly a lowering of regressors is
to the accuracy of my methodology. As displayed in Table 5.10, I compare the
national container predicted by these varying specifications relative to a time series
of observed loadedc container flows, both items being aggregated to total container
inflows (In) and outflows (Out), respectively. I find that estimating loading factors
for specific commodities by direction (separately) across panel data sets of export
and import activity yields the most accurate set of results. Additionally, the ‘Full’
and ‘Union’ sets of regressors perform best, of which more details are provided for
in the notes section of the table. For the purposes of this paper, I use the ‘Full –
Separately’ approach to generate country-specific container flows.
Alternative specifications for regressors have been evaluated with respect to
loading factors that vary across spatial– and income–based groupings. Although
neither of these specifications are used for the main results of this paper, their
associated results are available upon request. In the following section, I detail the
128
Unconventional Protectionism
Table 5.7: Jointly Estimated Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.145 0.199 19 62.361 85.625 26.39 none
0.078 0.108 21 62.208 85.414 29.17 port
0.125 0.171 21 61.769 84.812 29.17 year
0.126 0.172 22 60.240 82.712 30.56 mon
0.077 0.106 22 60.553 83.142 30.56 port+year
0.077 0.105 23 59.150 81.216 31.94 port+mon
0.126 0.173 21 61.769 84.812 29.17 year+mon
0.071 0.098 18 63.910 87.751 25.00 port-year
0.127 0.174 22 59.969 82.340 30.56 year-mon
0.078 0.107 23 60.485 83.049 31.94 port-mon
0.067 0.091 20 61.062 83.842 27.78 port-year + mon
0.074 0.102 21 60.600 83.207 29.17 year-mon + port
0.076 0.105 23 58.985 80.989 31.94 port-mon + year
0.057 0.078 16 64.163 88.099 22.22 portˆyearˆmon
0.075 0.103 23 60.330 82.836 31.94 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
performance of these measures, which generally appear to under–perform relative
to the directional loading factors used.
Geographic Loading Factors: Loading factors are estimated both jointly –
pooling import and export data together – and separately, where commodity-
specific loading factors vary based on whether they are an import or export.
Groups include Asia, Australia & Oceania, Europe, the Middle East & Africa,
and Southern & Central America.
Income Loading Factors: Loading factors are estimated both jointly – pooling
import and export data together – and separately, where commodity-specific load-
ing factors vary based on whether they are an import or export. Groups include
quartiles of countries, divided by World Bank measures of GDP per capita.
129
Unconventional Protectionism
Table 5.8: Import-Specific Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.199 0.229 18 71.492 82.449 25.00 none
0.119 0.137 3 86.318 99.546 4.17 port
0.152 0.175 19 70.990 81.869 26.39 year
0.150 0.173 19 71.276 82.199 26.39 mon
0.114 0.132 2 86.410 99.653 2.78 port+year
0.120 0.139 3 86.318 99.546 4.17 port+mon
0.152 0.175 19 70.990 81.869 26.39 year+mon
0.114 0.131 2 86.139 99.340 2.78 port-year
0.153 0.176 20 70.976 81.854 27.78 year-mon
0.119 0.137 4 83.897 96.754 5.56 port-mon
0.113 0.131 2 86.477 99.730 2.78 port-year + mon
0.115 0.132 2 86.410 99.653 2.78 year-mon + port
0.114 0.131 4 82.490 95.132 5.56 port-mon + year
0.115 0.133 2 86.410 99.653 2.78 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
Table 5.9: Export-Specific Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.080 0.150 18 45.637 85.852 25.00 none
0.071 0.133 4 48.449 91.142 5.56 port
0.064 0.121 13 48.464 91.169 18.06 year
0.064 0.121 13 48.464 91.169 18.06 mon
0.072 0.136 4 48.449 91.142 5.56 port+year
0.069 0.129 4 48.449 91.142 5.56 port+mon
0.064 0.121 13 48.464 91.169 18.06 year+mon
0.062 0.117 0 53.158 100.000 0.00 port-year
0.065 0.123 10 48.685 91.584 13.89 year-mon
0.068 0.129 4 48.449 91.142 5.56 port-mon
0.059 0.111 0 53.158 100.000 0.00 port-year + mon
0.070 0.133 5 48.442 91.127 6.94 year-mon + port
0.071 0.134 5 48.423 91.093 6.94 port-mon + year
0.071 0.133 4 48.449 91.142 5.56 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
Container Flow Diagnostics
130
Unconventional Protectionism
Table 5.10: Performance Diagnostics by Methodology
Method In-RMSE In-Corr Out-RMSE Out-Corr
Full — Jointly 56,638.14 0.980 39,092.72 0.775
Full — Separately 31,520.21 0.993 17,796.20 0.958
Intersect — Jointly 76,182.46 0.973 66,964.02 0.397
Intersect — Separately 34,837.47 0.992 19,368.11 0.951
Union — Jointly 60,875.81 0.979 48,363.68 0.658
Union — Separately 30,748.43 0.994 17,887.69 0.957
Note: The method list indicates which set of commodities were used as regressors in the estimation of commodity-
specific loading factors. ‘Full’ uses the entire set of HS2 product types. ‘Intersect’ uses a subset of HS2 products
that represent the top 50 highest commodity-specific shares of total export weight and total import weight. The
resulting commodity set is the intersection of common commodities between these two shortlists. ‘Union’ uses
the full set of top 50 commodities, rather than their intersection. RMSE columns denote root mean square error
and Corr columns list the correlation of each measure, relative to observed total container inflows and outflows.
Table 5.11: Jointly Estimated Geographic Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.139 0.192 31 65.051 89.550 11.31 none
0.090 0.124 31 65.818 90.606 11.31 port
0.111 0.153 36 61.362 84.472 13.14 year
0.112 0.154 36 62.023 85.382 13.14 mon
0.088 0.122 30 66.755 91.896 10.95 port+year
0.088 0.121 32 64.542 88.850 11.68 port+mon
0.110 0.152 36 62.137 85.538 13.14 year+mon
0.082 0.113 29 67.714 93.216 10.58 port-year
0.110 0.151 38 62.013 85.368 13.87 year-mon
0.091 0.126 31 66.112 91.010 11.31 port-mon
0.078 0.108 26 68.834 94.757 9.49 port-year + mon
0.085 0.117 28 67.002 92.236 10.22 year-mon + port
0.090 0.123 30 65.002 89.482 10.95 port-mon + year
0.065 0.090 29 65.783 90.558 10.58 portˆyearˆmon
0.086 0.118 28 65.829 90.620 10.22 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
As highlighted in Tables 5.17 and 5.18, models which include port and year
fixed effects yield the lowest root-mean-square error (RMSE) scores. These scores
compare predicted and observed US – East Asian and US – European container
flows, where the measure of interest is the ratio of bilateral loaded container unit
131
Unconventional Protectionism
Table 5.12: Import-Specific Geographic Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.175 0.203 21 75.554 87.264 8.4 none
0.119 0.137 7 86.090 99.433 2.8 port
0.123 0.142 21 74.528 86.079 8.4 year
0.120 0.138 22 77.493 89.503 8.8 mon
0.115 0.133 7 86.251 99.619 2.8 port+year
0.120 0.139 7 86.004 99.334 2.8 port+mon
0.120 0.138 24 77.213 89.180 9.6 year+mon
0.111 0.129 6 84.942 98.107 2.4 port-year
0.124 0.143 28 75.290 86.959 11.2 year-mon
0.119 0.137 9 86.131 99.480 3.6 port-mon
0.111 0.129 7 84.820 97.966 2.8 port-year + mon
0.117 0.135 5 86.157 99.510 2.0 year-mon + port
0.116 0.133 8 86.162 99.516 3.2 port-mon + year
0.116 0.134 6 86.177 99.533 2.4 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
Table 5.13: Export-Specific Geographic Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.083 0.157 11 52.322 98.630 4.10 none
0.072 0.136 11 52.446 98.864 4.10 port
0.071 0.133 19 51.811 97.666 7.09 year
0.073 0.137 19 51.135 96.393 7.09 mon
0.071 0.134 9 52.723 99.385 3.36 port+year
0.072 0.135 9 52.572 99.101 3.36 port+mon
0.072 0.136 19 51.811 97.666 7.09 year+mon
0.060 0.114 7 53.004 99.914 2.61 port-year
0.072 0.136 17 51.966 97.958 6.34 year-mon
0.076 0.143 9 52.711 99.362 3.36 port-mon
0.058 0.110 7 53.004 99.914 2.61 port-year + mon
0.069 0.130 9 51.818 97.680 3.36 year-mon + port
0.075 0.142 10 52.699 99.341 3.73 port-mon + year
0.071 0.134 8 52.726 99.391 2.99 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
flows. For East Asian, geographic country groupings perform similarly to load-
ing factors which vary only by commodity. For Europe, the standard approach
of commodity-specific loading factors with no interference in the loading factor
132
Unconventional Protectionism
Table 5.14: Jointly Estimated Income-based Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.131 0.180 45 61.401 84.459 20.55 none
0.083 0.115 38 63.967 87.988 17.35 port
0.111 0.153 47 57.436 79.004 21.46 year
0.111 0.153 45 59.264 81.518 20.55 mon
0.083 0.114 36 64.128 88.209 16.44 port+year
0.080 0.110 39 62.921 86.549 17.81 port+mon
0.111 0.153 47 58.716 80.765 21.46 year+mon
0.078 0.108 31 65.610 90.247 14.16 port-year
0.111 0.152 44 59.028 81.195 20.09 year-mon
0.080 0.110 39 61.791 84.995 17.81 port-mon
0.072 0.100 33 63.906 87.903 15.07 port-year + mon
0.076 0.105 36 64.100 88.170 16.44 year-mon + port
0.080 0.109 39 61.957 85.223 17.81 port-mon + year
0.057 0.079 27 62.860 86.465 12.33 portˆyearˆmon
0.079 0.109 36 63.977 88.001 16.44 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
Table 5.15: Import-Specific Income-based Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.170 0.196 29 78.826 90.925 13.12 none
0.123 0.142 9 86.246 99.484 4.07 port
0.133 0.153 25 79.163 91.314 11.31 year
0.130 0.150 26 79.066 91.202 11.76 mon
0.123 0.142 6 86.528 99.810 2.71 port+year
0.123 0.142 10 86.235 99.471 4.52 port+mon
0.131 0.151 26 79.066 91.202 11.76 year+mon
0.127 0.146 8 85.387 98.493 3.62 port-year
0.133 0.153 25 81.085 93.531 11.31 year-mon
0.120 0.138 6 86.615 99.910 2.71 port-mon
0.125 0.144 7 85.512 98.637 3.17 port-year + mon
0.122 0.141 6 86.521 99.801 2.71 year-mon + port
0.120 0.138 6 86.615 99.910 2.71 port-mon + year
0.123 0.142 7 86.520 99.800 3.17 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
estimations delivers the most accurate results. Considering both regions jointly, I
proceed with using no arbitrary country groupings for estimated loading factors.
133
Unconventional Protectionism
Table 5.16: Export-Specific Income-based Loading Factors
Weighted Weighted (M) Negative LFs % Trade % Trade (M) % Neg Coeff Fixed Effects
0.085 0.160 8 52.334 98.629 3.86 none
0.073 0.137 7 52.573 99.078 3.38 port
0.075 0.141 9 52.691 99.301 4.35 year
0.074 0.139 13 52.492 98.927 6.28 mon
0.074 0.140 6 52.614 99.157 2.90 port+year
0.071 0.135 8 52.572 99.076 3.86 port+mon
0.076 0.143 11 52.682 99.284 5.31 year+mon
0.062 0.117 5 52.959 99.806 2.42 port-year
0.077 0.144 11 52.612 99.152 5.31 year-mon
0.075 0.142 11 52.567 99.067 5.31 port-mon
0.058 0.110 5 53.033 99.945 2.42 port-year + mon
0.071 0.133 7 52.573 99.078 3.38 year-mon + port
0.077 0.145 11 52.566 99.065 5.31 port-mon + year
0.073 0.138 8 52.368 98.692 3.86 port+year+mon
Note: Column (1) reports trade value weighted average of loading factor coefficients. Column (2) reports the
same measure limited to manufactured goods. Column (3) reports the number of negative manufacture coefficients
estimated. Column (4) reports the non-negative manufacture coefficients’ share of total trade flows. Column (5)
reports the non-negative manufacture coefficients’ share of manufacture trade flows. Column (6) reports the
negative coefficient count as a percentage of manufacture coefficient count. Column (7) lists the associated fixed
effects used.
Table 5.17: RMSE of US - E. Asian Container Flow Ratios
Country Grouping Coef Filter Products none p p+y p+m py pm py+m ym+p pm+y p+y+m
Geographic None Agri+Manu 0.388 0.346 0.200 0.528 0.204 0.707 0.291 0.366 0.511 0.342
No Groups None Agri+Manu 0.058 0.408 0.224 0.574 0.180 0.908 0.240 0.314 0.695 0.359
Geographic Directional Agri+Manu 4.740 0.303 0.271 0.346 2.582 0.342 2.512 0.527 0.674 0.315
Income-based None Agri+Manu 0.240 0.423 0.335 0.584 0.231 0.777 0.361 0.505 0.724 0.487
No Groups Directional Agri+Manu 2.353 1.154 1.022 1.138 1.773 0.324 1.868 0.833 0.301 0.978
No Groups None Manufacturing 3.073 1.812 1.550 1.999 2.469 2.183 2.753 1.551 1.807 1.675
Geographic None Manufacturing 4.523 1.845 1.704 1.954 2.788 1.929 3.033 1.794 1.768 1.793
Income-based None Manufacturing 2.415 2.063 1.905 2.215 2.706 1.994 3.021 2.094 1.842 2.037
Income-based Directional Agri+Manu 3.952 2.718 2.642 2.307 2.598 0.870 3.042 1.808 0.976 2.224
Geographic Directional Manufacturing 8.346 2.877 2.735 2.860 6.314 2.616 6.019 3.231 3.496 2.723
No Groups Directional Manufacturing 5.422 4.087 3.693 4.038 5.552 2.327 5.207 3.083 2.110 3.579
Income-based Directional Manufacturing 8.192 6.129 5.537 6.377 7.118 6.067 8.141 6.212 6.208 5.876
Country Groupings includes (i) No grouping, (ii) Geographic (Asia/Oceania, Europe, South America and
Africa/Middle East, and (iii) Income-based (four quartiles based on each country’s average GDP per capita
between 2012 and 2021). Coef Filter includes (i) None – no corrections to estimated loading factors, and (ii) Di-
rectional – replaces negative loading factors with their opposite-direction counterpart for the same country-group,
iff the opposite-direction coefficient is of a a lower value. Products represents measures generated using either
(i) Agri+Manu – the entire set of commodity weight flows listed in the data set, or (ii) Manufacuturing – the 72
manufactures featured at the HS2 level, as defined on the TRAINS product grouping ‘manufactures’ set.
A1 (VII.) The European Customs Union and Container
Flows
Many of the countries featured in the multi-country baseline scenario of this pa-
per are European. Of those countries, Austria, the Czech Republic, Hungary and
134
Unconventional Protectionism
Table 5.18: RMSE of US-European Container Flow Ratios
Country Grouping Coef Filter Products none p p+y p+m py pm py+m ym+p pm+y p+y+m
No Groups None Manufacturing 2.130 0.056 0.064 0.045 0.110 0.126 0.216 0.070 0.166 0.055
No Groups None Agri+Manu 0.932 0.083 0.071 0.122 0.151 0.070 0.217 0.081 0.046 0.080
No Groups Directional Manufacturing 1.640 0.064 0.111 0.170 0.236 0.089 0.321 0.055 0.090 0.091
No Groups Directional Agri+Manu 1.636 0.208 0.262 0.292 0.447 0.125 0.495 0.191 0.106 0.221
Income-based None Agri+Manu 1.593 0.337 0.401 0.207 0.625 0.183 0.507 0.244 0.218 0.218
Income-based None Manufacturing 2.632 0.268 0.420 0.097 1.063 0.070 0.939 0.219 0.111 0.207
Geographic None Manufacturing 1.605 0.454 0.545 0.355 1.850 0.320 1.787 0.197 0.335 0.432
Geographic Directional Manufacturing 2.236 0.866 0.866 1.021 0.853 0.594 0.427 1.563 0.058 0.309
Geographic Directional Agri+Manu 2.337 0.938 0.911 1.045 0.907 0.818 0.564 1.445 0.237 0.373
Geographic None Agri+Manu 4.920 1.131 0.965 1.024 1.591 1.381 1.491 0.652 1.111 0.831
Income-based Directional Manufacturing 1.984 0.824 0.996 0.772 0.657 0.577 0.623 0.697 0.685 0.875
Income-based Directional Agri+Manu 2.288 1.033 1.168 0.874 0.740 0.397 0.685 0.757 0.498 0.932
Country Groupings includes (i) No grouping, (ii) Geographic (Asia/Oceania, Europe, South America and
Africa/Middle East, and (iii) Income-based (four quartiles based on each country’s average GDP per capita
between 2012 and 2021). Coef Filter includes (i) None – no corrections to estimated loading factors, and (ii) Di-
rectional – replaces negative loading factors with their opposite-direction counterpart for the same country-group,
iff the opposite-direction coefficient is of a a lower value. Products represents measures generated using either
(i) Agri+Manu – the entire set of commodity weight flows listed in the data set, or (ii) Manufacuturing – the 72
manufactures featured at the HS2 level, as defined on the TRAINS product grouping ‘manufactures’ set.
Switzerland represent inland regions which could only be accessed by US container-
ized trade via third party coastal channels such as the ports of the Rotterdam or
Antwerp. Each of these countries is also part of the European Customs Union.
Due to the frictionless nature of trade and apparent interdependence of countries
with respect to port access, I treat the EU Single Market as a single trade partner
entity. Eurostat container flow data suggests that only upon cross-country aggre-
gation does the European Customs Union region function as a balanced container
redistribution system. In contrast, individual European countries which form this
union maintain imbalanced container flow systems at the national level (Figure
5.3). This pattern of local imbalances is strikingly similar to the heterogeneous
roles played by individual US ports which, only when combined, maintain a bal-
anced redistribution system of bilateral container flows.
135
Unconventional Protectionism
Figure 5.3: European Specialization by Net Flow Status (2017)
10
5
0
-5
lan
d
ium ar
k y d d y l y n a n e s
o lg m It
al a i
n inl
an lan an do
m ug rke a
i
e nm g rt u Sp a ed
e anc ndP e e F Ir er in Po T om Sw F
r
her
la
B D G  K R et
nit
ed N
U
Note: The net flow to thruflow ratio uses inflows less outflows of loaded and empty container units divided by
the total flow of loaded & empty container unit traffic. This 2017 data is sourced from “Volume of containers
transported to/from main ports by direction, partner entity, container size and loading status”, extraction ID:
MAR GO QM.
A1 (VIII.) Container Monopsony
This section is motivated by a particular quirk of the cost minimization problem
that firms would face in a round trip setting and the one-for-one transformation of
inputs (inbound loaded and empty containers) into transport services (outbound
loaded containers). Suppose trade is imbalanced and the net importer country
generates a positive amount of outbound empties (eji > 0). In this case the
output of transport services is a function of these two inputs.
lij = f(lji, eji)
Since container flows are assumed to be balanced between countries, this would
imply that transport services from i to j are equal to total container inflows at
136
Net Flow-Thruflow, Ratio
Unconventional Protectionism
port i, or, lij = f(lji, eji) = lji + eji, our usual profit function constraint in a trade
imbalance setting. Taking the ratio of marginal products with respect to these
two inputs:
MPlji ∂f(lji, eji)/∂lji cjiMRTS = = = 1 = = Input Price Ratio
MPe ∂f(lji, eji)/∂eji r↔ji ij
Consider a conventional MRTS in a transport setting, where capital K and
labour L inputs generate a transport service Y . Normally the MRTS varies along
a given isoquant, given different bundles of inputs zj. For example, should the
capital-labor ratio be particularly high, a relatively more capital-intense input
bundle that generates the same of output, Ȳ , requires significantly more units of
capital compared to labor-intense input bundle. The input price ratio between
capital and labor is fixed across all possible consumption bundles. A cost mini-
mizing firm selects an input bundle where MRTS is tangent to a constant price
ratio.
In the container redistribution case, the MRTS is instead fixed to a value of
1 across all consumption bundles, which under constant input price ratios implies
corner solutions where a firm will only utilize the cheapest input. To introduce a
unique solution on the net importer side which features positive container outflows
in both empty and loaded units, I use a loaded container input price that increases
in the level loaded container inputs.4 This yields variation in the input price ratio
rather than the MRTS, given variation in input bundles. Tangency occurs at the
level of loaded containers lji necessary to set cji(lji) = r↔, where c
′
ji(lji) > 0.ij
4Intuition: Additional loaded containers on a net importer route would imply a longer du-
ration with respect to unloading and cleaning at the net exporter port before the containers are
ready to be utilized as inputs. Each loaded container takes more time relative to an empty. The
shipping service cannot commence until the last arriving loaded unit is processed and emptied.
Since the first “processed” loaded container input is not usable until the last loaded container
input is prepared, I represent this accumulating time challenge with a rising input price per
loaded container input.
137
Unconventional Protectionism
The resulting profit maximization problem can be expressed as follows.
π↔ = Tijlij + Tjilji − cij(lij)lij − cji(lji)lji − r↔(eij + eji) (5.6)
ij ij
s.t. lij + eij = lji + eji,
There are a number of ways of introducing this increasing input cost parameter.
I resort to using the simplest possible expressions, where loaded container input
prices increase linearly with respective quantities.
As displayed in Figure 1, the inclusion of rising input prices for one particu-
lar input eliminates the possibility of corner solutions, as arbitrage opportunities
across input prices are eliminated by a perfectly competitive market. The higher
slope of cji(lji) implies there is a greater cost or more rapid elevating trade-off
associated with loading containers at the net importer country compared to the
net exporter country. Upon intersection with the input price of empty containers,
the loaded container quantity is identified.
The relative differences in slopes establish the capacity max{lji, lij}, empty
container load |lij − lji| and associated input prices of providing a shipping ser-
vice. These differences should be representative of exogenous supply and demand
factors. For example, should relative demand for lij increase due to an exogenous
preference shock, the slope of cji(lji) should increase and the slope of cij(lij) should
decrease, causing the trade imbalance displayed above to widen, shipping capacity
to increase and empty, eji = lij − lji container flows to rise.
I display two variations and solve for both balanced and imbalanced trade.
1. This first form of input price rises as loaded container inputs rise on route ji.
The producer will continue to stack loaded containers onto the ‘backhaul’
route until the input price is equal to the constant input price of an empty
138
Unconventional Protectionism
Figure 5.4: Input Price by Loaded Container
Input Price
cji(lji)
cij(lij)
rij
0
l∗ ∗ji lij Containers (lji, eji)
container, eji. Form: cji(lji) = θijlji
2. The inclusion of an added loaded container input, lji, yields the correspond-
ing freight rate, Tji but comes at the cost of a percentage θij of a completed
‘full’ haul trip’s from i to j, Tij. The percentage scales as the loaded in-
put rises. This adjustment captures hows the increased velocity that round
trips can complete laps at in cases where the ‘backhaul’ features a relatively
greater level of empties per container input. Form: cji(lji) = θijTijlji
Case I: Balanced Trade
The production function for transport services appears as lij = f(lji), where the
marginal product of the input (MPLij ) is equal to 1 since lij = lji. Plugging this
updated production constraint into the profit maximization problem of equation
(11), the problem becomes analogous with Section 1.2.1 and Wong (2022). Using
139
Unconventional Protectionism
the first increasing input price function, the transport operator problem becomes:
max π↔ = Tijlij + Tjilij − (θijlij)lij − (θjilij)lij
{l ijij}
∂π
FOC: = 0 =⇒ Tij + Tji = 2θijlij + 2θjilij
∂lij
Tij + Tji
lij = lji = (5.7)
2(θij + θji)
Consider the inverse demand function implied by equation (4).
ϵ− 1 − 1
Tij = a ϵijl
ϵ ij
− wiτij
Substituting out freight rates in equation (12),
ϵ−1 −
1
ϵ−1 −
1 − 1
a ϵ ϵijlij − wiτij + ajilji − w τ ϵ−1(a + a )l ϵ − w τ − w τ
l = l = ϵ ϵ
j ji ϵ ij ji ij i ij j ji
ij ji =
2(θij + θji) 2(θij + θji)
Appears to be a non-linear solution. Below I detail a case in which the wiτij terms
1
do not feature. In this scenario, I divide by (l −ij) ϵ to solve for l
∗
ij,
( ) ϵ
∗ 1+ 1 ϵ− 1 aij + aji ⇒ ∗ ϵ− 1 aij + a
1+ϵ
ji
(l εij) = = l =ϵ 2(θij + θ
ij
ji) ϵ 2(θij + θji)
Substituting this expression into the inverse demand function, the equilibrium
140
Unconventional Protectionism
freight rates are;
(( ) )− 1ϵ
∗ ϵ− 1 ϵ−
ϵ
1 a 1+ϵij + aji
Tij = aijϵ ( ϵ 2(θij + θj)i)
− −
1
ϵ 1 ϵ− 1 aij + a 1+ϵji
=( aijϵ ) ϵ 2((θij + θji)
− 1−
1 ) 1
ϵ 1 1+ϵ 2(θij + θji) 1+ϵ
=( ) a(ijϵ aij + ajiϵ ) 1
ϵ− 1 1+ϵ 2(θ + θ ) 1+ϵij ji
= aij
ϵ aij + aji
Shifting to the increasing input price function based on opportunity cost and
round trip velocity, solving the model involves the following steps.
max π↔ = Tijlij + Tjilij − (θjilijTji)lij − (θjilijTij)lij
{l ijij}
∂π
FOC: = 0 =⇒ Tij + Tji = 2θijTijlij + 2θjiTjilij
∂lij
∗ ∗ Tij + Tjilij = lji = (5.8)2(θijTij + θjiTji)
A similar non-linear solution case is arrived upon.
Case II: Imbalanced Trade
The production function for transport services on the net exporter route is lij =
f(lji, eji), where the marginal product of a loaded input (MP
L
ij ) is equal to the
marginal product of an additional empty input (MPEij ), since lij = lji + eji. In
MPL
this case the marginal rate of technical substitution, ijE , is equal to 1. Using theMPij
first form of the increasing input cost function, the profit maximization problem
can be expressed as:
141
Unconventional Protectionism
max π↔ = Tijlij + Tjilji − (θijlji)lji − (θjilij)lij − r↔(0 + eji) s.t. eji = lij − lji
{l ij ijij ,lji,eji}
max π↔ = Tijlij + Tjilji − (θijlji)lji − (θjilij)lij − r↔(lij − lji)
{l ,l } ij ijij ji
FOC:
∂π
= 0 =⇒ Tij − 2θjilij − r↔ = 0
∂lij ij
∂π
= 0 =⇒ Tji − 2θijlji + r↔ = 0
∂lji ij
Supply and inverse supply of transport services can be expressed as follows, im-
plying an upward sloping supply curve.
Tij + r↔ Tji − r↔
ij
lS = , lS
ij
ij ji = , T
S = 2θ Sjilij + r↔ , T = 2θijlji − r↔
2θji 2θ
ij
ij ij
ji ij
Using equation (4), the demand for these goods are downward sloping in freight
rates, points of intersection can be identified.
( )−ϵ ( ) T ∗ϵ 1 −ϵ ij + r↔D ijlij = wiτ ∗ S− ij + T = = l1 ϵ a ij 2θ ijij ji
D ϵ− 1 ∗− 1Tij = aijl ϵij − wiτij = 2θ l∗ji ij + r↔ = T Sϵ ij ij
In this case, the round trip effect does not present itself. Ships are not setting
maximum capacity due to circumstances pertaining to both i and j. Need an
expression in which these equilibrium outcomes of price and quantity reflect use
of {aij, aji, τij, τji}.
Using instead the increasing function based on opportunity cost of a slower com-
142
Unconventional Protectionism
pletion rate of round trips:
max π↔ = Tijlij + Tjilji − (θijljiTij)lji − (θjilijTji)lij − r↔(0 + eji) s.t. eji = lij − lji
{l ,l ,e } ij ijij ji ji
max π↔ = Tijlij + Tjilji − (θijljiTij)lji − (θjilijTji)lij − r↔(lij − lji)
{l ,l } ij ijij ji
FOC:
∂π
= 0 =⇒ Tij − 2θjiTjilij − r↔ = 0
∂lij ij
∂π
= 0 =⇒ Tji − 2θijTijlji + r↔ = 0
∂lji ij
Tij − r↔ Tji + r↔
S ijl = , lS
ij
ij ji =2θijTji 2θjiTij
Using the inverse demand function implied in equation (4), the solutions for
quantities become:
1 1
ϵ−1( ∗− ∗−a l ϵij ij − wiτij − r↔) ϵ−1a( l ϵji ji − wjτji + r↔ϵ ij ϵ ijl∗ = , l∗ij 1 ji = 1 )
2θ ϵ−1
∗− ∗−
a ϵij jilji − w ϵ−1 ϵϵ jτji 2θji aijlϵ ij − wiτij
143
Unconventional Protectionism
Rearranging l∗ij
ϵ− 1 ∗− 1 1
2θ ( a l ϵ ∗
ϵ− 1 ∗−
ϵ
ij ji
ϵ ji
− wjτji)lij = aijlij − wiτij − r↔ϵ ij
ϵ− 1 ∗− 1 ϵ− 1 ∗− 1
2θij( ajil ϵ
∗
ji )lij = aijl
ϵ
ij − wiτij − r↔ + (w ∗jτji)lϵ ϵ ij ij
ϵ− 1 ∗− 1 ϵ− 1 ∗− 1−1
2θ a ϵ ϵ −1ij jilji = aijlij − (wiτij − r↔)lϵ ϵ ij ij
+ wjτji
∗− 1 1 a ∗− ϵ+1ij 1 ϵ 1 −1 1 ϵ 1l ϵji =( l ϵij − (wiτij − r↔)l + (wjτji)2θij aji 2θij ϵ− 1 a ijji( ij ) 2θij ϵ− 1 aji )−ϵ1 a ∗− ϵ+1∗ ij − 1 ϵ 1 1 ϵ 1lji = l ϵ w τ − r l−1( ij ↔ + (w τ )2θij aji 2θij ϵ− i ij ij − j ji1 aji ij 2θij ϵ 1 aji )−ϵ
1 a ∗− ϵ+1∗ ji − 1 ϵ 1
( ) 1 ϵ 1
l = l ϵij ji wjτ
−1
ji − r↔ l + (wiτij)
2θji aij 2θji ϵ− 1 aij ij ji 2θji ϵ− 1 aij
In this case I have two equations and two unknowns, but the explicit solutions for
{lij, lji} are not clear nor would the associated comparative statics be. Likely need
to reconsider another method of going about solving this model, or else go down a
computational route where the comparative statics can only be assessed through
simulation. The benefit of this approach would be incorporating round trip effects
in an unbalanced trade setting.
144
Container Ports
Appendix II
A2 (I.) Data Appendix
Our dataset of vessel movements and characteristics extends from 1977 to 2023
and requires the use of multiple data sources. We first detail two daily records
of containership visits at US ports. Lastly, we describe our automatic identifi-
cation system (AIS) data of individual vessels’ geo-locations within US waters,
which varies at a minute level. This description includes our means of converting
position records into port call data, which yields a precise minute-level measure
of containership dwell times at port. Lastly, we elaborate on VesselTracker data
which complements all three data sets by allowing us to update each vessels build
year and maximum container capacity.
Port Statistical Areas (USACE)
We identify the top ports of containerized shipping based on Panjiva records
of twenty-foot equivalent unit (TEU) container traffic.5 We then access public
records of port statistical areas, provided by the USACE. There are some ports
missing from this list. We construct polygon areas manually for these ports, taking
guidance from each ports provided facility maps. In cases where facility maps are
not provided, we research to satellite imagery provided by Google Earth. These
port polygons are used to track the entry and exit of individual vessels, using their
geolocations which are updated in minute intervals.
US port callings (Lloyd’s List Intelligence)
Lloyd’s List Intelligence (LLI) maintains a network of agents on the ground
at ports worldwide to validate information and address customized information
requests. Ultimately, the Lloyd’s Agency Network, which is exclusively operated
5See https://www.logisticsmgmt.com/article/top 30 u.s. ports big ports got bigger in 2020
145
Container Ports
by Lloyd’s List Intelligence, is the signature capability that distinguishes the com-
pany’s data collection and validation process from any other company within the
maritime trade data industry. We have partnered with Lloyd’s to secure ports
callings data for 1977–2002. This data is the product of hands-on effort maintain-
ing presences across US ports and frequently referring to on-site paper records of
visits. Given that AIS (vessel tracking) technology not coming into the market
until 2009, our source offers a unique edge over our competitors that do not have
access to this historical data. As a result, LLI is the sole source of reliable human
intelligence capable of providing port callings data from 1977 – 2002. We must
also highlight that while we focused our attention on US data, there are global
historical records of port callings available through LLI.
This data reports the International Maritime Organization code of each con-
tainership offloading or onloading goods at ports across the US. Additionally, we
observe the arrival and departure times of vessel visits and the vessel name of each
ship. For post-2002 data, we refer to public data made available through the US
Army Corps of Engineers (USACE).
Port Entrances and Clearances (USACE)
The publication of data from the CBP form 1300 is done so via the CBP form
1400 (Entrance) and 1401 (Clearance). The CBP publishes the latest 1400 and
1401 from the previous week. The USACE has established a feed for the data from
CBP and compiles records of port entrances and clearance, releases these statistics
via the Waterborne Commerce Statistics Center. The CBP form 1300 dates back to
1965 where an international convention successfully unified the forms necessary for
processing vessels arriving into countries. The agreement was adopted in London,
England on 9 April 1965 and entered into force on 5 March 1967.6 The US used to
have the CBP form 1300 and the CBP form 1301. Both forms were consolidated
6See IMO documentation on the Convention on Facilitation of International Maritime Traffic
(FAL).
146
Container Ports
into the CBP form 1300.
Upon the late 90s and early 2000s, it was a big time for CBP as far as electronic
records and digitizing information. The CBP form 1400 and 1401 used to be
“published” on paper at each Customhouse around the country for the local records
of that particular port, where any individual could walk up and browse through the
records. Now they are published via PDF on CBP.gov, updated weekly each Friday
morning. According to speculation from our contacts with the CBP, the ‘late 90s
and early 2000s’ CBP was in the process of automating this data collection and
publication and it simply took time for all ports to get onboard. Upon a national
security disaster on September 11th 2001, these processes were accelerated by the
then Customs Service. We note sparseness in records of port calls from 1997–2001,
which underreport vessel activity at US ports relative to LLI data. Based on our
understanding of the CBP’s transition into digitalized records, we assume records
are representative of US containerized transport services from 2002 onward.
These records list individual vessel IMO codes, along with each visiting vessel’s
name, the date at which they enter, the port of entrance, the date at which a
vessel leaves and fixed characteristics such as the gross and net tonnage of each
vessel. Gross Tonnage (GT) is the volume of all enclosed spaces on a vessel. This
includes the Engine Room and other non-cargo spaces. Most Maritime Regulations
(SOLAS, MARPOL, etc.) apply to vessels based on their Gross Tonnage. Net
Tonnage (NT), on the other hand, is the volume of only the cargo-carrying spaces
on the vessel. This is the tonnage that determines the earning capability of the
vessel. Most port/anchorage dues apply to vessels based on their Net Tonnage.
Crucially, each record of a visit also yields a vessel’s ICST code, which details the
‘International Classification of Ships’. This allows us to identify which cargo ships
in the AIS data set are containerships (ICST code 310).
Upon combining 2002–2021 records, we correct a subset of IMO codes that
147
Container Ports
appear to have been incorrectly entered by those filing 1400 and 1401 forms. In
cases of non-reported gross tonnages of vessels, we refer to each vessel’s IMO code
and draw upon the single reported measure of gross tonnage. In cases where
multiple tonnages are reported, we use the most recent reported gross tonnage of
a uniquely identified vessel for each one of its port dwell events. This is due to
records, on average, exhibitiing more accuracy in data entry we advance further
in the reported years of the dataset.
Vessel Position, AIS data (MarineCadastre)
For minute-level vessel activity between January 1st 2010 and December 31st
2014, available data is provided in a .zip format, where each issued file represents
vessel movements within a specific Coordinated Universal Time (UTC) section of
the US. These files each feature an entire month’s worth of vessel position data,
in some cases reaching half a GB in size before unzipping. Upon unzipping each
file, the resulting .gdb file contains separate broadcast, vessel and voyage data. We
merge these items together into a single data object and save specific year–month–
zone files across the timezones belonging to ports in our 30 Port Statistical Areas
of interest. Vessels are identifiable only through reported MMSI codes, which have
been scrambled in order to mask vessel identities. We have been provided with a
descrambler by the National Oceanic and Atmospheric Administration (NOAA)
which allows us to revert MMSI codes back to their original state. Upon consul-
tation with the U.S. Coast Guard, we can been provided with the associated IMO
code of each one of these unscrambled MMSI code identifiers for vessels, which
is crucial in pairing fixed characteristics from the Port Entrance and Clearances
(USACE) data.
Upon a series of cross validation checks on raw monthly and daily data of vessel
positions with overlapping USACE records, a narrowing sparseness in AIS-imputed
dwell events up to 2016 was identified. After March 2016, USACE containership
148
Container Ports
visit records mirror the scale of activity reflected by AIS movements. We attribute
the prior pattern of under reported vessel movements to the fact that AIS trans-
port use only became a requirement as of the issuing of a federal mandate on
January 30 2015, entitled “Vessel Requirements for Notices of Arrival and Depar-
ture, and Automatic Identification System”.7 This filing required the fitting of
AIS transponders for all commercial vessels and their persistent use in US waters
by March 2016. We settle on relying on vessel position data from March 2016 to
December 2023 to ensure we are studying port call activity in an environment in
which transponder use is mandatory.
For minute-level vessel activity between January 1st 2015 and December 31st
2023, available data is provided in a .csv format, where each issued file represents
all vessel movements within US waters for a single calendar date. Vessel identities
were not scrambled and each file not only reports MMSI codes, but also each vessels
IMO code and name. The download and processing of AIS data is automated and
takes the following steps.
1. Download raw data daily (March 2016- December 2023) vessel positions.
2. IMO codes for positions between 2016 and 2022 allow us to identify cargo
vessels, which includes – containerships, vehicle carriers, bulk carriers, roll-
on roll-off (ro-ro) cargo ships, reefers, livestock carriers, and general cargo
multi-deck carriers.
3. Filter out any vessels that neither report a cargo-related ICST code or do
not maintain a cargo/vessel type code of 79.
4. Filter out any vessels that do not operate within our port polygon areas of
interest.
7For more details, see 80 FR 5282 Docket No. USCG-2005-21869. Last visited on
April 17 2024 at https://www.federalregister.gov/documents/2015/01/30/2015-01331/vessel-
requirements-for-notices-of-arrival-and-departure-and-automatic-identification-system
149
Container Ports
5. For each month of observations: Apply algorithm to remaining set of vessel
positions which identifies key timestamps in which vessels (i) enter a port
region, (ii) come to a full stop, (iii) begin moving, and (iv) exit the same
port. While we originally paired items (ii) and (iii) with AIS status signals of
“mooring” and “underway”, respectively, subsequent analysis revealed that
vessel operators are not always diligent or prompt in indicating their vessel’s
current status. Each of these four states represents a specific stage of a port
dwell event.
6. The time elapsed between when mooring starts and ends is recorded as the
duration of a dwell period between states (ii) and (iii). This allows us to
capture any dwell events that occur within a month or across the last 20
days of a given month (monthly) and subsequent 20 days of the following
month (month–gap). Dwell experiences that exceed 40 days in length are
not captured.
7. We combine the monthly and month–gap dwell events and retain the subset
of distinct visits to avoid any duplications of containership activity.
8. We apply post-processing filtration to these dwell events, in adherence with
the guidelines detailed by the Bureau of Transport Statistics (BTS). This
includes combining multiple dwell events at a single port together, if the
time elapsed between each dwell event is less than two hours. Furthermore,
any dwell events that take place in two separate ports and imply an average
travel speed of over 25 knots are discarded as such as travel speed is infeasible
for cargo vessels. Lastly, we discard short dwell events of less than 2 hours
to avoid any over–counting of dwell events which can instead be attributed
to crew changes and refueling.
150
Container Ports
Vessel Characteristics Registry Data (VesselTracking)
Through use of the ‘xml2’ R package, we have scraped 233,581 vessels’ fixed
characteristics (name, IMO code, build year, container capacity and length) from
records published on www.vesseltracking.net/ships (VT). This data applies across
the global fleet of existing and decommissioned vessels and offers a near 100%
match rate across the vessels listed in our LLI, USACE and AIS data sets. We
combine the three datasets and further enhance the granularity of the data by
merging in these fixed characteristics.
Any cleaning attributed to the data is very minimal, due to the high quality
of the source material. In the case of ten distinct IMOs, the VT data yields
container capacities of zero. We utilize third-party vessel–spotting & –tracking
portals, which offers details pertaining to relatively old vessels and newly built
containerships. Below we detail our matches for each distinct vessel:
Table 5.19: Manually Entered IMO–TEU Capacity Pairs
Vessel IMO 9849643 9719068 9809904 9809916 9819947 9819959 9869667 9757228 9757230 9757216
TEU Capacity 1268 3600 1148 1148 1148 1148 1268 14220 14220 14000
A2 (II.) Coastal Stylized Facts
We also examine these shifts towards new technology by coastal region in Figure
5.5. The West Coast exhibits a far more smooth adoption of technology, relative
to the East Coast. This is likely driven by many shipbuilding sites and large man-
ufacturing centers being based in East Asia. Upon the expansion of the Panama
Canal, Post Panamax vessels – designed for the greater width of the Canal –
rapidly gained market share on the East Coast. In contrast, the Gulf Coast ap-
pears to utilize such technology far more sparsely. This is potentially attributable
to factors such as older port infrastructure and lower population density.
151
Container Ports
Figure 5.5: Vessel Category Shares of Coastal Containership Activity
West Coast Gulf Coast East Coast
80 90 00 10 20 80 90 00 10 20 80 90 00 0 0
19 19 20 20 20
1 2
19 19 20 20 20 19 19 20 20 20
I - Micro and Early Builds III - Post Panamax V - Very Large Containership (VLCS)
II - Panamax IV - Neo-Panamax (NPX) VI - Ultra Large Containership (ULCS)
Note: The long dashed vertical line indicates the point at which the Panama canal was expanded, June 26 2016.
As displayed in Figure 5.6, growth in the total mass of vessel transit to the
US West Coast has stagnated since 2002 and begun declining from 2018. Limited
port resources are becoming increasingly depended upon along the US East Coast.
While the West Coast was the main host for some of the largest vessels servicing
the US between 1995–2016, the gap has largely narrowed since the expansion of
the Panama canal.
Figure 5.6: Container Capacity of Vessel Visits by Region
(a) Total TEU Capacity (b) Average TEU Capacity
m1 m17 2 7m
1
2m
1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 m1 17 2 7 m m
1 m1
97 98 98 99 99 00 00 01
2 17 22 77 82 87 92 97 02 07 12 17 22
1 1 1 1 1 2 2 2 20 20 19 19 19 19 19 20 20 20 20 20
West Coast East Coast All Ports West Coast East Coast All Ports
152
Container Units ('000) Share of Max TEU Capacity (%)
0 2000 4000 6000 8000 0.2 0.4 0.6 0.8 1.0
Container Units
0 2000 4000 6000 8000 10000
Container Ports
A2 (III.) Panjiva Records
Given the supportive nature of the port of laden records we rely upon for validation
exercises, we encapsulate all data discussion, cleaning methodology and associated
empirical exercises in the following Appendix section. We acquired Panjiva Inc.
data – a division of Standard and Poor’s – which details bill of lading information
for all seaborne US imports from January 2007 to September 2023. Panjiva cleans
this data to standardize the names of the ports, ships, companies, and associated
unloaded container volumes. We subset this data to only consider goods that
arrive on seaborne container ships. To do so, we use the vessels’ IMO codes, which
uniquely identify ships across time. Paired with our knowledge of the population
of IMO’s associated with containership voyages at key US ports, this allows us to
focus on containerized shipping specifically.
The bill of lading data Panjiva offers is an industry-standardized system, which
act as receipts of shipment, recording all information on the shipment and all the
parties involved in the shipping process. The US Customs and Border Patrol
(CBP) agency collects these bills in addition to customs information at all ports
of entry into the US. We apply the following steps in order to further clean and
wrangle the provided data;
1. All IMO entries that do not contain 7 digits are dropped from the sample –
3% sample attrition,
2. Observations with missing IMO codes are isolated and vessel names are
matched to VesselTracking records. In cases where one unique IMO code
is associated with these records, we assign the IMO code to the vessel – 6.5%
of IMOs remain missing,
3. We repeat Step 2 but focus on unique IMO codes associated with distinct
vessel name and port combinations.
153
Container Ports
4. Subset the remaining set of vessel names with missing IMOs and postive TEU
offloadings. Using the set of unique vessel names from VesselTracker, apply
a Jaro Winkler algorithm which yields a numeric value of [0, 1] regarding
the similarity of two string variables (Winkler, 1990) – we use matches of
0.93789 similarity or higher and repeat Steps 3 & 4 – 4.4% of IMO entries
are missing.
5. For multiple observations of a distinct Arrival Date, Port Name, IMO, Vessel
Name offloading event, we sum the number of offloaded containers into one
single entry
6. We subset our combined sample of vessel dwell events to 2007–2023. For each
row of our port call data, we filter the Panjiva sample for an identical IMO–
port match. Among these matches, we calculate the difference in arrival time
by days and isolate the minimium difference. Any port dwell events that are
within a 9 day window of time – 4 days before or after a Panjiva arrival –
are labelled a match.8
We attribute differences in arrival times to administrative differences in the
submission of paperwork and the differing natures of these two items. Dates in
which a portion of containerized goods imports are processed do not always reflect
with the date a given vessel carrying those goods arrives. In scenarios where
individual vessel dwell times elongate abnormally and cause the timing between a
vessel arrival and completion of the offloading process to increase in length, such
as in the COVID-19 period, our matching process weakens. Nevertheless, only
very thin windows of time are necessary to achieve match rates of over 90% across
our sample.
Additionally, Panjiva imputes TEU (twenty-foot equivalent) container volumes
8Match rates are displayed across 0 to 9 day windows of time in Figure 5.7
154
Container Ports
based on container information and other shipment characteristics (Flaaen et al.,
2023). In appraising the accuracy of these estimates, we find that Panjiva sys-
tematically under-reports the number of offloaded container units associated with
imports, when compared aggregated data of port–level container inflows via two
sources. Roughly half of TEU imports are represented by Panjiva data.
These extensive and intensive margins of missingness should be considered
when evaluating our findings, relative to those from Section 4. In Table 5.20,
we revisit differences in efficiency across containership classes of increasing capac-
ity, length and volume. Our measure of efficiency, dwell time per unloaded TEU
container, tends to exhibit higher performance rates across vessels of larger size
categories. However, evidence suggests that the largest containerships in this sub-
sample are handled handled any better than our reference group of micro-build
and Panamax containership classes. Referring to Figure 5.5, this may be due
to US ports not yet being well-optimized for the latest generation of supersized
containerships. Upon incremental adjustments to port infrastructure and market
shares of these vessels elevating beyond 1%, we anticipate that the efficiency of
port service interactions with these vessels may become highly rewarding.
Figure 5.7: Match Rate, Panjiva Vessel Offloading to Port Call Records
100
90
80
70
60
+/- 0 Days +/- 1 Day +/- 2 Days +/- 3 Days +/- 4 Days
Note: Day windows range across port call records and are in reference to individual arrival dates of Panjiva data.
155
Share (%)
2010
2015
2020
Container Ports
Table 5.20: OLS Vessel Efficiency by Ship Category – Panjiva
Dependent Variable: Efficiency
(1) (2) (3) (4)
Post Panamax -0.3528∗∗∗ -0.3633∗∗∗ -0.3635∗∗∗ -0.3518∗∗∗
(0.0604) (0.0585) (0.0584) (0.0609)
Neo-Panamax (NPX) -0.4483∗∗∗ -0.4046∗∗∗ -0.4049∗∗∗ -0.4439∗∗∗
(0.0518) (0.0590) (0.0589) (0.0529)
Very & Ultra Large Containerships -0.3485 -0.2710 -0.2729 -0.3647∗
(0.2107) (0.2106) (0.2101) (0.2082)
Vessel Age 0.1028∗∗∗ 0.0999∗∗∗ 0.1000∗∗∗ 0.1039∗∗∗
(0.0283) (0.0277) (0.0278) (0.0282)
Day-Month-Year Yes Yes Yes Yes
Port-Year Yes
Port Yes
Port-Month Yes
Port-Year-Month Yes
Region Yes
Observations 226,078 226,078 226,078 226,078
R2 0.12 0.10 0.10 0.13
Notes: Clustered port standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. Examines variation
container handling efficiency of individual vessels relative to our reference group of early-build/micro vessels and
Panamax category containerships.
Table 5.21: IV Elasticity of Vessel Efficiency with respect to Port Traffic – Panjiva
Dependent Variables: Efficiency Port Vessel Count Efficiency
OLS First Stage 2SLS
(1) (2) (3) (4) (5) (6)
Port Vessel Count 0.2351∗∗∗ 0.2390 -2.782 0.0834 -1.676
(0.0652) (0.3004) (225,593.7) (0.3062) (62,994.3)
HHI -0.0433∗∗ -0.1612∗∗∗ -0.0427 -0.6765 -0.1130∗ -0.1737
(0.0160) (0.0208) (0.0573) (45,013.1) (0.0636) (5,656.3)
Vessel Age -0.0280 -0.0016 -0.0280 -0.0259 -0.0141 -0.0263
(0.0447) (0.0021) (0.0449) (988.3) (0.0459) (47.72)
Large Share (%) 0.0663 0.1113∗∗∗ 0.0659 -0.0063 0.0910 0.3023
(0.0479) (0.0306) (0.0609) (7,782.0) (0.0590) (7,349.9)
Port Trade Exposure 0.1740∗∗∗
(0.0467)
Ship FE ✓ ✓ ✓ ✓ ✓ ✓
Port–Year FE ✓ ✓ ✓
Port FE ✓
Port–Month FE ✓
Port–Year–Month FE ✓
Coastline FE ✓
Observations 226,078 226,078 226,078 226,078 226,078 226,078
R2 0.23 0.96 0.23 0.11 0.22 0.23
Notes: Clustered port standard-errors in parentheses. Codes: ***: 0.01, **: 0.05, *: 0.1. All variables are in
logs excluding ‘large vessel share’ which reports the percentage of visiting vessels in the 4th and 5th quintiles of
vessel size across ports and time. ‘Efficiency’ is dwell time per offloaded TEU. ‘Port Traffic’ reports the weekly
tonnage of containership vessels present at port p, excluding ship i’s contribution. ‘Port Vessel Count’ reports the
number of containership vessels present at port for a given week. ‘HHI’ reports a Herfindahl–Hirschman index of
vessel tonnage concentration. The higher the value, the more densely concentrated total weekly mass is across
visiting containerships at a given port p. ‘Vessel Age’ reports the difference in year and build year of ship i. New
entrant vessels are those built in the same year they are actively servicing US ports. We apply an age of 0.5 to
new entrants to preserve them in the sample. These vessels are potentially key drivers of spillover variation in
surrounding vessel dwell times.
156
Cargo Ships & Coastal Smog
Appendix III
A3 (I.) Data Appendix
The construction of voyage data yields each individual vessel’s associated emissions
in lieu of arriving at a given port. These measures are constructed through many
stages of cleaning with respect to vessel position data provided by MarineCadastre.
Port Statistical Areas (USACE)
I identify the top ports of containerized shipping based on Panjiva records of
twenty-foot equivalent unit (TEU) container traffic.9 i then access public records
of port statistical areas, provided by the United States Army Corps of Engineers
(USACE). I construct polygon areas manually for any missing ports, taking guid-
ance from each ports provided facility maps. In cases where facility maps are
not provided, I refer to satellite imagery provided by Google Earth. These port
polygons are used to track the entry and exit of individual vessels, using their
geolocations which are updated in minute intervals.
Matching MarineTraffic and MarineCadastre
MT provides records of port departure timestamps across vessels bound for
the US west coast. MC provides the individual micro-movements of uniquely
identified vessels for the same time period in US waters. Using unique identifiers,
the MMSI and IMO codes of each vessel, and range of calendar dates between
a prior departure and US west coast arrival, I capture the initial moment each
vessel enters US waters. For the subset of uniquely identified vessels that are not
detected in US waters between these key dates, I label these vessels as “missing”.
The rate of missingness is relatively low, averaging less than 1 percent of the MT
sample. However, I do detect a notable spike in missingness in 2020 for the months
9See https://www.logisticsmgmt.com/article/top 30 u.s. ports big ports got bigger in 2020
157
Cargo Ships & Coastal Smog
of July, August and September. This appears to be largely concentrated among
the ports of California (Figure 5.8).
Figure 5.8: Matching AIS positions to Port Call Records
9
6
3
0
2020 2021 2022
Year-Month
LONG BEACH LOS ANGELES OAKLAND
20
15
10
5
0
SEATTLE TACOMA 2020 2021 2022
20
15
10
5
0
2020 2021 2022 2020 2021 2022
Year-Month
158
% of journeys with missing AIS data % of journeys with missing AIS data
Cargo Ships & Coastal Smog
A3 (II.) Fuel Consumption Function
Using Figure 4.5 from Rodrigue (2020), I digitalize each function into a set of
coordinates, paired with the midpoint of ship size value based on the ranges dis-
played. I then generate an approximated function to represent each of these five
functions, excluding “10,000+” for representing an undefined midpoint. Taking
1,000 equidistant grid points of speed (knots) between 0 and 25, I feed this range
into each function and generate a dataframe of speeds (X), fuel consumption levels
(Y), and container capacities (Z).
I regress the following specification of interacted and polynominal terms to
estimate fuel consumption (Y), given a combination of speed (X) and TEU capacity
values (Z).
Yi = α + β1X + β2Zi + β3X
2 + β Z2 + β X × Z + β X2i 4 i 5 i i 6 i × Zi + εi (5.9)
Taking the set of estimated parameters {α̂, β̂1, . . . , β̂6}, I use the resulting bi-
variate polynomial function to convert a given vessel’s speed and container capacity
into estimated fuel consumption. The resulting function is convex in speed and
appears to rise non-linearly with increased to vessel mass. In Figure 5.9 below, I
plot these estimates against the defined functions featured in 4.5 for select capacity
ranges to better represent the appropriate fit this specification delivers. It should
be highlighted that the function performs poorly at levels of speed less frequently
observed in practice. In these extreme cases of slow speed, I refer to maritime
transport estimates detailed in the next appendix subsection.
159
Cargo Ships & Coastal Smog
Figure 5.9: Fuel Consumption Function
400
300
200
100
0
0 10 20 30
Speed (Knots)
Port-Specific Average Voyage Speed
Note: Dashed lines represent imputed functions with TEU fixed at the midpoint of ranges associated with solid
line. These solid line functions are mapped using cartesian-based graph readers, which were applied to figures
featured in Rodrigue (2020).
A3 (III.) Queuing Emissions & Additional Find-
ings
In this section, I provide further insight into how queuing emissions are summarised
for individual vessel–voyages. Additionally, I provide supportive evidence of my
main findings in which I broaden the sample of relevant vessels.
160
Fuel Consumption (Tonnes per day)
Cargo Ships & Coastal Smog
Local Queuing Emissions Calculations
I refer to Bai et al. (2020) to estimate fuel consumption associated with queuing
activity – a combination of stationary and active movements, which includes cruis-
ing, maneuvering, and anchored stauses. This study decomposes time and speed
across these various stages of local presence, and identifies associated emissions as
momentum of large vessels slowly accumulates from a stationary position. Figure
5.11 demonstrates how speed adjusts across difference stages of vessel transit. Fig-
ure 5.11 provides a depiction of the corresponding adjustments in emission types
based on these stages and speeds within each stage.
Figure 5.10: Speed by Stage
Source: Bai et al. (2020), Gaseous Emissions from a Seagoing Ship under Different Operating Conditions in the
Coastal Region of China, Atmosphere, Vol. 11(3), pp 305.
For a given series of local movements, I observe the travel speed S and container
capacity C of each vessel within US waters at timestamp t. Timestamps are
updated at a minute-level resolution and the previous observation’s timestamp for
vessel i corresponds to the earlier point of t′. The set of emissivity indices are
represented by (3206, 10.50, 50.50, 2.30), which assumes compliance with the IMO
2020 sulfur content mandate. For each reported position p of vessel i, I take the
following steps.
161
Cargo Ships & Coastal Smog
Figure 5.11: Emissions by Speed-Stage
Source: Bai et al. (2020), Gaseous Emissions from a Seagoing Ship under Different Operating Conditions in the
Coastal Region of China, Atmosphere, Vol. 11(3), pp 305.
If Sipt > 10:
• Generate Φipt (Sipt, Ci) for fuel consumption in tonnes per day
• Measure Φipt (Sipt, Ci) ∗ (Emissivity Indices) as the kilograms rate of associ-
ated emissions per day
• ′Take timestamps t and t−1 to calculate γipt (t, t′) = t−t to scale for the per-24
centage of a day that has elapsed between two consecutive vessel coordinate
pings
• Calculate δipt(Sipt, Ci, t, t′) = γipt (t, t′) Φipt (Sipt, Ci) (Emissivity Indices) to
indicate the total kilogram emissions of vessel i between time t and t′, for
CO2, SOX , NOX and PM2.5, respectively
If Sipt ≤ 10, repeat the above steps with speed held fixed at 10 for fuel consump-
tion, Φipt (10, Ci). Upon multiplying by the appropriate emissivity index value, the
total kilogram emissions contribution between t′ and t is reduced by an emissions
scaling factor, based on slow travel speed pollutant levels mapped in Figure 5.11.10
10At a speed of zero, NOX during the maneuvering stage is measured at a concentration of
450ppm (64.29% of 10 knot speed emissions). Plotting each knots per hour measure relative to
their emission concentrate deviations from 700 (the 10 knot maximum) allows for the mapping of
an increasing, concave emission scaling factor function of speed. This approach assumes that each
emission type accumulates at the same diminishing marginal rate at maneuvering speeds between
0–10 knots per hour. The exact formula I estimate is ESF = 64.28571+7.539686x−0.3968257x2.
162
Cargo Ships & Coastal Smog
DiD – All International Voyages
Rather than keeping vessel voyages on fixed lanes, I relax the sample to allow for
vessels to adjust which origins they stem from following the introduction of the
new queuing system. As displayed in Figure 5.22, I find strikingly similar results
to those featured in Figure 4.10 across each of my indicators of local vessel activity
and emissivity.
Table 5.22: Difference-in-Difference Estimates – Queuing Emissions
Total Emissions Duration Speed Em. per Knot Distance Em. per Hour
(1) (2) (3) (4) (5) (6)
Post–Period -0.4646∗∗ -0.5030∗∗ 0.0733 0.1206 -0.5852∗∗ 0.0384
(0.1969) (0.2179) (0.1507) (0.1542) (0.2306) (0.0920)
Treatment 0.7243∗∗∗ 0.8548∗∗∗ -0.4183∗∗∗ -0.3460∗∗∗ 1.070∗∗∗ -0.1305∗∗
(0.0970) (0.1082) (0.0746) (0.0776) (0.1296) (0.0539)
DiD -0.2125∗∗∗ -0.2865∗∗∗ 0.3617∗∗∗ -0.3787∗∗∗ 0.1662∗ 0.0741
(0.0754) (0.1000) (0.0801) (0.0755) (0.0946) (0.0563)
Vessel–Voyage FE ✓ ✓ ✓ ✓ ✓ ✓
Year–Month FE ✓ ✓ ✓ ✓ ✓ ✓
Observations 5,090 5,090 5,089 5,090 5,090 5,090
R2 0.70 0.70 0.72 0.66 0.63 0.78
Note: ***: 0.01, **: 0.05, *: 0.1. Standard-errors are robust to clustering within vessel–voyage lanes of transport
service. Each observation is a distinct queuing experience of a vessel arriving on the US west coast between Nov
2019 and Nov 2022. I filter only for vessel voyages that maintained international trade routes pre- and post-
policy and drop any observations that lack matching vessel movement data. To limit extreme outlier distortions,
I exclude any voyages with emissions less than the 25th percentile minus three times the interquartile range (75th
percentile - 25th percentile) or higher than the 75th percentile plus three times the interquartile range (Davies
and Jeppesen, 2015).
Appendix References
Bai, C., Y. Li, B. Liu, Z. Zhang, and P. Wu (2020). Gaseous emissions from a
seagoing ship under different operating conditions in the coastal region of china.
Atmosphere 11 (3).
Behrens, K. and P. Picard (2011). Transportation, freight rates, and economic
geography. Journal of International Economics 85 (2), 280–291.
Davies, R. and T. Jeppesen (2015). Export mode, firm heterogeneity, and
source country characteristics. Review of World Economics (Weltwirtschaftliches
Archiv) 151 (2), 169–195.
163
Cargo Ships & Coastal Smog
Flaaen, A., F. Haberkorn, L. Lewis, A. Monken, J. Pierce, R. Rhodes, and M. Yi
(2023). Bill of lading data in international trade research with an application to
the COVID-19 pandemic. Review of International Economics 31 (3), 1146–1172.
Hummels, D., V. Lugovskyy, and A. Skiba (2009). The Trade Reducing Ef-
fects of Market Power in International Shipping. Journal of Development Eco-
nomics 89 (1), 84–97.
Notteboom, T., A. Pallis, and J.-P. Rodrigue (2022, January). Port Economics,
Management and Policy. Routledge.
Rodrigue, J.-P. (2020, May). The Geography of Transport Systems. Fifth edition.
— Abingdon, Oxon ; New York, NY : Routledge, 2020.: Routledge.
Winkler, W. E. (1990). String Comparator Metrics and Enhanced Decision Rules
in the Fellegi-Sunter Model of Record Linkage. In Proceedings of the Section on
Survey Research, pp. 354–359.
Wong, W. F. (2022). The Round Trip Effect: Endogenous Transport Costs and
International Trade. American Economic Journal: Applied Economics 14 (4),
127–66.
164