Policy Research Working Paper 9992 Fathoming Shipping Costs An Exploration of Recent Literature, Data, and Patterns Adina Ardelean Volodymyr Lugovskyy Alexandre Skiba David Terner Transport Global Practice May 2022 Policy Research Working Paper 9992 Abstract Recent academic research on maritime shipping has been different periods and comparison of different functional affected by two factors: (i) changes in shipping technol- forms of transportation costs: per twenty-foot equivalent, ogy and market structure and (ii) the explosion of new per kilo, and ad-valorem. The paper also presents several micro data sets in shipping and international trade. This novel empirical exercises to further the insights from the paper describes this research, focusing on containerized existing literature. It shows that the effect of sea distance on and dry-bulk shipping and emphasizing recent trends and trade costs slowly “dies” over time, but its effect on seaborne determinants of freight rates. The findings are presented trade volumes is both significant and persistent over time. in a systematic set of stylized facts. Some of the important Moreover, the paper documents the presence of substantial and perhaps underappreciated insights include discussion country- and product-level heterogeneity of the impact of of the scale effects of utilizing fuel based on vessel size over sea distance on international maritime freight markets. This paper is a product of the Transport Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at atardelean@scu. edu, vlugovsk@indiana.edu, askiba@uwyo.edu, and dmterner@iu.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Fathoming Shipping Costs: An Exploration of Recent Literature, Data, and Patterns* Adina Ardelean† Volodymyr Lugovskyy‡ Alexandre Skiba§ David Terner¶ Keywords: Maritime Shipping, Freight Rate, Ports JEL Classification: L9, F1, R4 * This paper was prepared as a background paper for the Global Transport Practice Flagship report Shrinking Economic Distance, led by Matias Herrera Dappe. We thank Matias Herrera Dappe, Jan Hoffman, David Hummels, Ahmad Lashkaripour, Mathilde Lebrand, Alejandro Molnar, Aiga Stokenberga, and participants of the World Bank 2022 Workshop for the Transport Flagship Report Shrinking Economic Distance for helpful comments. † Department of Economics, Santa Clara University; e-mail: atardelean@scu.edu ‡ Department of Economics, Indiana University, Wylie Hall Rm 301, 100 S. Woodlawn, Bloomington, IN 47405-7104; e-mail: vlugovsk@indiana.edu § Department of Economics and Finance, University of Wyoming, Dept 3985, 1000 E University Ave, Laramie, WY 82071, e-mail: askiba@uwyo.edu ¶ Department of Economics, Indiana University, Wylie Hall Rm 101, 100 S. Woodlawn, Bloomington, IN 47405-7104; e-mail: dmterner@iu.edu 1 Introduction Maritime shipping is indispensable for international trade: ships transport 70% of the world’s trade value and 80% of the world’s trade weight (UNCTAD, 2015, 2017). Maritime shipping dates back to at least 5,000 years ago, and its history is well documented, including multiple waves in its devel- opment facilitated by technological advancements, wars, trade with colonies, etc. (Stopford, 2009). One of the more recent waves occurred in the 1950s when containerized shipping revolutionized maritime shipping. This technological breakthrough had a pronounced impact on trade volumes, patterns, and infrastructure (Hummels, 2007; Levinson, 2008; Blonigen and Wilson, 2018). The rate of change has further intensified in the last two decades with rapid fleet upgrades and expansions, im- provements in infrastructure (port upgrading, expansions of Panama and Suez Canals), and changes in regulation and market structure. In the last two decades (i.e., 2000-2020), the availability of qualitatively new and much more dis- aggregated data sets prompted a surge in empirical research of maritime shipping. This paper sum- marizes this research and complements it with novel data analysis and stylized facts. Understanding of the complexity underlying factors of the maritime shipping is even more important in light of the recent disruptions and high freight rates in maritime shipping caused by the COVID-19 pandemic. While we do not directly explore the most recent events, our review might be instrumental for this research as well.1 Our focus is on container and dry-bulk maritime shipping.2 Containerized mar- itime shipping is the dominant mode of international transportation: it accounts for two-thirds of total world trade. On the other hand, dry bulk shipping is used for commodities, and it accounts for about half of the total seaborne trade by weight. We start by documenting several vital trends in maritime shipping. First, motivated by recent papers on the back-haul effects in maritime shipping (Wong, 2020; Brancaccio et al., 2020), we show that maritime trade is highly imbalanced across regions and that the magnitudes of the imbalances 1 Please see the UNCTAD discussions for more information on the recent crisis at https://unctad.org/meeting/ad- hoc-expert-meeting-maritime-supply-chain-crisis. 2 Other types of maritime shipping include various types of Tankers, Ro-Ro (Roll-on/roll-off), general cargo, and reefer vessels. have increased over time. Second, we document the rapid increase in the number of vessels and average vessel size between 1996 and 2021. Specifically, containerized fleet capacity has increased 8 times, and the dry-bulk fleet capacity has increased 3.8 times during this period. In the presence of economies of scale that characterize the shipping industry, these changes should have reduced the cost of shipping. But, how pronounced are these economies of scale? We show that in terms of fuel consumption in containerized shipping they became significant only in the last decade. In terms of the new-build vessel prices, they are the most pronounced when comparing the smallest and the mid-size vessels. Finally, turning to the freight rates, we show that they decreased over time in the dry-bulk shipping, but not in the containerized shipping. Our understanding of maritime shipping heavily relies on data availability and quality. In Section 3, we discuss how the measures of transport costs evolved with the arrival of new datasets. By way of motivating example, researchers in the past often had to rely on distance as the only available proxy of transport costs. We show that direct distance measures are a poor proxy for transport cost: the distance elasticity of freight rate elasticity is economically small and, further, decreasing over time. Likewise, in many other datasets, the units of goods were typically unavailable; rather, the only way to measure transport costs was in ad-valorem terms as a ratio of the c.i.f. and f.o.b. measures of trade values. We discuss the advantages and disadvantages of this measure. As the data with the units of traded goods became available, freight rates have been shown to have a pronounced specific (i.e., non-advalorem) component, which shed a new light on our un- derstanding of quality specialization and gains from trade (Hummels and Skiba, 2004; Wong, 2021). Lately, in addition to the per-unit measure of transport costs, we observe per volume and per con- tainer freight rates. We provide the summary and the discussion of these rates and their trends. We also discuss datasets which allow to measure port quality and maritime transport policy restrictions. Our section 4 focuses on the freight rate determinants in containerized liner shipping. The ex- isting literature shows that price formation in these markets is more complex than in most trade models. It crucially depends on the network organization and market structure. Furthermore, liner shipping is split into two markets: the (long-term) contract market and the spot market. Long-term 2 contracts are strictly confidential, and thus we are limited to indirect inference about price forma- tion; an analysis in patterns of the overall distribution of freight rates sheds light on this market’s otherwise opaque price formation. Presumably, the contract market contributes to the within-route freight rate heterogeneity with larger trading firms facing lower freight rates (Ardelean and Lugov- skyy, 2021). Unlike for the contract market, the data on the spot markets are available. Using these data, Wong (2020) evaluated the magnitude of the back-haul effects, while Ganapati et al. (2020) demonstrated the effect of network organization on the distribution of freight rates across routes. Other factors that affect liner freight rates are product types and variety characteristics, distance traveled, and port infrastructure. We discuss all factors in detail. Product variety characteristics are a first-order determinant of freight rates. Failing to account for variety-specific characteristics leads to debilitating omitted variables bias. To this end, we demonstrate that the distance elasticity of the freight rate is heterogeneous across countries and products. Ports provide essential infrastructure for maritime shipping. Section 4.4 describes the literature evaluating and ranking port efficiency across the globe. We discuss and compare two approaches: (i) based on regressing transport cost on port characteristics or fixed effects, and (ii) the production function approach to estimation port efficiency based on the observable inputs. We conclude by discussing the 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. 2 Major Trends in Maritime Shipping 2.1 Containerized Exports, Imports, and Imbalances by Region Maritime shipping is primarily determined by the developments in the world economy and trade, with the growth of international trade being among the most critical factors in determining the de- mand for transportation services (Stopford, 2009). At the same time, containerization has been a 3 key driver of globalization since the 1960s as it allows for (i) faster transportation, loading, and unloading, (ii) reduced cargo damage and theft, and (iii) refrigeration or temperature-controlled en- vironments for perishable goods (Hummels, 2007; Levinson, 2008; Bernhofen et al., 2016). Moreover, the availability of container shipping has changed the geography of world trade and global sup- ply chains (Notteboom and Rodrigue, 2007). These days, 95% of trade in manufactured goods is transported by containerships. To get a better understanding of the trade trends, we leveraged Clarksons Shipping Intelligence Network (SIN)3 to calculate maritime containerized trade and trade imbalances for years 2002-2020 for eight regions: Africa, Europe, Far East (including South East Asia), Indian Sub-Continent, Latin America, Middle East, North America, and Oceania.4 The imports and exports trends are presented in Figure 1 with imports and exports for each region normalized to 100 in the year 2002. We summa- rize it as follows. Stylized Fact 1. Between 2002 and 2020, Far East, Indian Sub-Continent, and Middle East faced the most pronounced export and import growth—up to 300%. Africa, Indian Sub-Continent, and Oceania faced much faster imports than exports growth. 3 Seesection 3.3 for details on Clarksons SIN database. 4 Naturally,even within the same region, the country-specific trade and trade imbalances may differ in terms of mag- nitudes and even directions. Furthermore, the dry-bulk trade imbalances are different from the containerized trade imbalances. 4 Figure 1: Containerized Trade Growth Figure 2: Containerized Trade Imbalances In Figure 2, we present trade imbalances, calculated as both absolute and relative differences of TEUs (twenty-foot equivalent unit), and summarize them below.5 Analyzing these trends is impor- tant because trade imbalances contributed to pronounced asymmetries in freight rates. For example, 5 In the left panel of Figure 2, the summation of all imbalances in a given year adds up to zero. 5 Wong (2020) documented pronounced backhaul effects in the spot freight rates reported by Drewry, especially for the US-China routes, where it costs three times more to ship a container from China to the US than from the US to China. Stylized Fact 2. Between 2002 and 2020, maritime trade was highly imbalanced. The containerized imports exceeded exports for 7 of 8 regions except for the Far East, for which the net exports increased from 10 to 30 million TEU. The largest relative imbalances are observed in the Middle East and Africa. 2.2 Industry Growth, Technological Changes, and Economies of Scale The shipping industry has responded to the growing demand for transportation by building larger vessels that presumably improve efficiency through economies of scale. For example, containerships have increased in size from 700-1,000 TEU in the 1960s to more than 23,000 TEU (Fan et al., 2018). Using the Clarksons SIN data, we document that the growth of the containerized maritime fleet has been remarkable in the last 25 years.6 We plotted the trends in Figures 3 and 4 and summarize them as following. Stylized Fact 3. The total containerized fleet capacity has increased eight times between 1996 and 2021, with the average vessel size increasing 2.7 times. The total capacity share of the largest ships, of over 8,000 TEU, has increased from 0% in 1996 to 55% in 2021. Indeed, while the containerized fleet capacity, measured in twenty-foot equivalent units (TEU), has increased eight times, the number of vessels has increased only three times. From Figure 4, we learn that smaller ships, with the capacity of under 3,000 TEU, were rapidly replaced with the larger ones. Specifically, the capacity share of the smaller vessels has decreased from 65% in 1996 to 18% in 2021. Instead, as pointed out above, the capacity of the largest ships, of over 8,000 TEU, has increased from 0% in 1996 to 55% in 2021. The largest ships in the industry these days exceed 23,000 TEUs.7 6 Clarksons SIN provides a comprehensive list of container and dry-bulk vessel types and their total capacity for each year starting 1996. See section 3.3 for information about SIN. 7 For example, the Korean carrier, HMM owns a dozen of 23,000+ TEUs containerships (HMM Vessel Fleet). 6 Figure 3: Evolution of world fleet by market segment Notes: The left panel depicts a time series of fleet capacity. Containership fleet capacity is measured in TEU (twenty-foot equivalent) units; global TEU capacity in 1996 was approximately 2.91 million TEU. Dry Bulk capacity is measured in DWT (deadweight tonnage); global DWT capacity in 1996 was approximately 62.9 million DWT. The right panel depicts a time series of the total vessel count by market segment. The total number of Containerships and Dry Bulk vessels in 1996 was 1,911 and 2,087, respectively. Source: Clarksons Shipping Intelligence Network, authors’ calculations. For the dry-bulkers, the magnitudes of these trends are smaller. As demonstrated by Figures 3 and 5, the capacity has increased 3.8 times while the number of vessels has increased 2.3 times, with the smaller Handysize ships being gradually replaced with the larger Handymax, Panamax, and Capesize vessels. Stylized Fact 4. The total dry-bulk fleet capacity has increased 3.8 times between 1996 and 2021, with the average vessel size increasing 1.7 times. The total capacity share of the largest ships, Panamax and Capesize, has increased from 51% in 1996 to 65% in 2021. Given such a pronounced increase in the average vessel size, do we also observe economies of scale at the vessel level? To answer this question, we first examined the pricing of the newly-built ships in 2019. The results presented in Figure 6 demonstrate pronounced economies of scale. Stylized Fact 5. For both containerships and dry-bulkers, the capacity-normalized pricing of new ships built decreases in their size, especially when comparing smaller and medium-size ships. 7 Figure 4: Containership Fleet Evolution Notes: Source: Clarksons Shipping Intelligence Network, authors’ calculations. Next, we turn to the fuel consumption aspect. Greater fuel efficiency can be achieved by choos- ing a smaller speed of travelling. Dry-bulkers have a greater flexibility in this respect as they do not operate on the schedule and their cargos are typically much less sensitive than that of the container- ized. Since we do not observe the actual speed by dry-bulkers, we cannot provide a careful analysis of their fuel efficiency. Instead, we focus only on the containerships as they have to operate within a given schedule and thus their speed can be perceived as more homogeneous across various routes. Cullinane and Khanna (1999) provide a simple equation to model vessel v’s daily fuel consumption, FO (assuming an 80 percent power utilization to achieve designed service speed), namely: FOv = Installed BHPv × Specific Fuel Oil Consumptionv × 80% × 24/(106 ), where BHP stands for the input brake horsepower of an engine. We use Clarksons World Fleet Registry to compute FOv for all containerized ships built between 1969 and 2020,8 and then estimate 8 Note that Installed bhp is available for 99% of all containerships. Likewise, specific fuel oil consumption (SFOC) data is available for roughly 73% of all ships; missing SOFC data are replaced with average SFOC value per decade. 8 Figure 5: Dry Bulk Fleet Evolution Notes: Handysize vessels have a capacity up to 35,000 DWT, Handymax vessels have a capacity between 35,000 and 65,000 DWT, Panamax vessels have a capacity between 65,000 and 80,000 DWT, and Capesize vessels have a capacity of at least 80,000 DWT. These capacity thresholds should be viewed as approximate. Source: Clarksons Shipping Intelligence Network, authors’ calculations. the simple OLS model distinguishing between 3 time periods:9 −2009 −2009 log( FOv ) = −3.10 + 0.93 log( TEUv ) − 0.23 12000 v + 0.04 12000 v × log( TEUv ) (0.07) (0.01) (0.09) (0.01) (1) −2020 −2020 + 0.68 12010 v − 0.11 12010 v × log( TEUv ) + ε v (0.08) (0.01) where 12000 v −2009 is 1 if vessel v is built during 2000-2009 and 0 otherwise; 12010−2020 is 1 if vessel v is v built during 2010-2020 and 0 otherwise; and TEUv is the ship’s TEU capacity. Equation (1) allows us to explore two aspects of the fuel consumption in containerized shipping: (i) the scale economies and (ii) changes in the fuel consumption and scale economies over time. For example, the magnitude of the log( TEU ) coefficient, 0.93, indicates that the ship-level scale economies are rather weak for ships built between 1969 and 1999: a 10% greater TEU capacity increases the expected fuel consumption by 9.3%. A positive coefficient of 0.04 of the interaction term of log( TEU ) and 12000 v −2009 tells us that the 9 Thenumber of observations per year is significantly lower for the earlier periods. That is why the first period is the longest, accounting for 31 years between 1969 and 1999. The total number of observations is 5,734. All coefficients reported in eq. 1 are statistically significant at 1% level; R2 = 0.92. 9 (a) Containtership (b) Dry Bulker Figure 6: Prices of the Newbuild Vessels Normalized by Vessel Size (2019) Source: Clarksons Shipping Intelligence Network. Newbuild vessel prices are normalized by the vessel size: TEU for the containerships and DWT for the dry-bulk vessels. Both figures demonstrate that normalized prices decrease in the vessel size. scale economies were 4 percentage points weaker for the ships built between 2000 and 2009. Finally, between 2010 and 2020, the scale economies are the strongest: the corresponding size elasticity of fuel consumption is 0.82. In addition to the period-specific elasticities, equation 1 also accounts for the time-specific multipliers. We use this information to calculate the predicted fuel consumption based on the vessel size in each of the three time periods. Figure 7 reports the results. Based on these results, we make the following statement. Figure 7: Fuel Economies in Containerized Shipping 10 Stylized Fact 6. Between 1969 and 2009, the fuel consumption per container was roughly the same for all ship sizes. The fuel consumption economies of scale became more visible only starting in 2010. The overall fuel efficiency for all ship sizes has also visibly improved starting in 2010. An increase in fuel efficiency in the 2010s has coincided with a rapid increase in the number and capacity share of the eco-ships with electronic engines. As a result, their capacity share in the total liner fleet has increased from 0% in 2004 to 56% in 2020 (Lugovskyy and Terner, 2021). The data on the share of the fuel costs in the operational and total costs is sparse. Based on the data from Delhaye et al. (2010), highlighted by Bernacki (2021), we report the following. Stylized Fact 7. In 2010, the fuel costs in the total costs constituted 54% for vessel sizes between 500 and 2,000 TEU and 42% to 46% for vessel sizes between 5,000 and 12,000 TEU. The calculations are based on the speed of 14 knots for the first (smaller) type of vessels and on the speed of 18 knots for the second type. 2.3 Freight Rates The research on international trade pays special attention to transportation costs as they represent a sizeable part of the overall trade costs. Transport costs affect the trade volume and determine trade patterns, including the organization of just-in-time production and global supply chains (Hummels, 2007; Behar and Venables, 2011). Over time, the contribution of shipping costs to trade costs increased in both developed and developing countries (Clark et al., 2004; Hummels, 2007; Hummels et al., 2009). The magnitudes and trends differ for the dry-bulkers and containerships, which is not surprising given how different the industries are. Dry-bulkers are used for commodities with lower per tonne value and time sensitivity than manufactured goods transported by containerships. Dry-bulk ship- ping markets are organized as decentralized spot markets where trading firms hire a ship at the spot market for a specific route—similar to hiring a taxi. The market is characterized by high dispersion 11 in freight rates across destinations. It stems from the fact that trade in commodities is highly imbal- anced. As a result, countries with large net imports face much higher ‘incoming’ than ‘outcoming’ freight rates, with many ‘outcoming’ vessels traveling without a cargo, i.e., ballast traveling (Bran- caccio et al., 2021).10 Leveraging the Clarksons SIN spot freight rates, we demonstrate this fact with Table 3 (Column 5) and summarize it as follows. Stylized Fact 8. There is a pronounced dispersion in the normalized dry-bulk freight rates across des- tinations. Using Clarksons Research data from 2010 to 2016, Brancaccio et al. (2020) demonstrated that the asymmetry and endogeneity in dry-bulk freight rates have important implications for international trade and beyond. First, it weakens the strength of comparative advantage and reallocates produc- tion from exporters to importers. Second, there is a network effect which can be demonstrated as follows. If one of the large importers experiences a slow-down in its imports, some ships reallocate to other countries and regions, thus reducing their freight rates and increasing trade in those regions. Over time, the freight rates are affected by the demand volatility and uncertainty and by the sluggish response in supply due to time to build and convex operating costs (Kalouptsidi, 2014).11 To construct the trend of dry-bulk freight rates, we utilize Jacks and Stuermer (2021) data and estimates, who draw on a bevy of historical sources to construct an impressive 170 year-long time series (see Appendix A in their paper for specifics). We present the resulting trend in Figure 8. When explaining the long-run trend in real dry-bulk freight rates, the authors appeal to Harley (1988), Mohammed and Williamson (2004), and Tenold (2019) who suggested the long-run productivity gains as the major contributing factors.12 Instead, for the short-run variation, Jacks and Stuermer (2021) found the demand shocks to account for approximately 50% of short-run boom/bust historical freight rate 10 The magnitude of the effect is very large. According to Brancaccio et al. (2021), a staggering 42% of dry-bulk vessels are travelling without cargo. 11 Kalouptsidi (2014) described pronounced “echo effects” in this market: entry spikes are likely to repeat with intervals equal to the ship’s lifetime. Government interventions potentially may smooth the supply and the volatility in this market. 12 For example, a significant drop in the freight rates between 1850 and 1914 can be attributed to switching from wind sailing to steam engine shipping and the opening of the Suez Canal in 1869 (Pascali, 2017). 12 variation. Importantly, these innovations resulted in a significant increase in trade, GDP, and welfare of countries involved in maritime trade (Pascali, 2017; Fajgelbaum and Redding, 2021). Stylized Fact 9. Between 1850 and 2020, the dry-bulk freight rate decreased by 71%. However, the trend was not monotonic over time and is characterized by high volatility. Figure 8: Real Dry Bulk Freight Rates in the Long Run Source: Jacks and Stuermer (2021)’s data/estimates. The red line corresponds to a dry bulk real freight index (1850=100), whereas the blue line denotes the long-run trend derived from the Christiano-Fitzgerlad band pass filter (assuming a cyclical component of 70 years). Containerized shipping is critical for trade in manufactured goods. We describe its market struc- ture and the determinants of the freight rates in Section 4. In this section, we will briefly describe the relevant trends. Hummels (2007) reported that the nominal liner freight rates were increasing at the annual rate of 10-15% in the 1970s. Even adjusted for the prices of traded goods, the liner price index has increased 2.4 times between the late 1950s and 1985. It was dropping afterward, but in 2004 it was still 1.7 times greater than in the late 1950s (Hummels, 2007, Figure 4). Notteboom and Vernimmen (2009) demonstrate that the economies of scale at the vessel level did not translate into lower prices for maritime shipping between 2000-2009 because of the increased investment in these larger ships over the period. For the later period, using Chilean imports data, we also do not detect a clear trend in the containerized freight rates (see Figure 10).13 Why did containerization not decrease freight rates? Hummels (2007) attributes the non-declining 13 For the most recent trends in the containerized and dry-bulk freight rates, see also UNCTAD (2021a). 13 containerized freight rates to the increasing input costs, such as fuel, ship prices, and port costs. From the market structure perspective, the industry has significant entry lags due to the time it takes to build a new ship. Bridgman (2021) attributes the lack of a dramatic fall in freight rates at least in part to the strong labor unions in ports. Due to their strong bargaining power, port-related labor costs did not fall much even though containerization resulted in enormous labor productivity gains. Finally, the demand for international trade and transportation services is constantly increasing (see Fig. 1), which may allow maintaining high freight rates in the highly concentrated industry with the legalized collusion. Stylized Fact 10. Between 1970 and 2004, containerized freight rates did not exhibit a decreasing trend. The non-decreasing trend, despite some technological advances, can be blamed on the increasing input costs, market structure, and constantly increasing demand. 3 Transport Costs Measures and Data Sources We consider three measures of transportation costs: (i) approximation by distance, (ii) the c.i.f. over f.o.b. ratio, and (iii) the direct (either specific or ad-valorem) measures. This section discusses the advantages and disadvantages of each measure in the context. We also describe available datasets and provide their summary in Table 1. 3.1 Transportation Costs Approximated by Distance Understanding the size and nature of transport costs requires detailed data on freight rates. These data are sparse. Researchers have widely used physical distance as a proxy for transport cost to overcome the paucity of transport cost data.14 This tendency in empirical literature has been in part motivated by a robust to many specifications and periods ln( Distance) coefficient of (−1) in the aggregate trade gravity regressions initially introduced by Tinbergen (1962). In these regressions, 14 An example of the widely-used bilateral distances is CEPII trade barriers and geography dataset. 14 distance serves as a proxy of all trade costs, including transportation costs. These papers are nicely summarized in two meta-studies: (i) Disdier and Head (2008) reported that between 1985 and 2000, the elasticity persisted at around −0.90 (it was somewhat lower before that), and (ii) Head and Mayer (2014) report an average elasticity of −0.93 based on 159 papers with 1,835 elasticity estimates. Such a strong and robust partial correlation between the distance and trade flow prompted using distance in many trade and growth papers. Notably, Romer and Frankel (1999) use distance as the exogenous shifter of trade flows to disentangle the endogeneity between trade and growth. Redding and Venables (2004) rely on distance as an exogenous determinant of market access and trade to explain income differences across countries. At the same time, this strong correlation is not necessar- ily an indicator of distance being a good proxy for transportation costs. Many other potential trade barriers are likely to be correlated with distance.15 Indeed, both the previous empirical work and our own analysis in this paper document a rather weak correlation between the distance and mar- itime transport costs. Martínez-Zarzoso and Nowak-Lehmann (2007) demonstrated that distance is a weak predictor of transportation cost. Hummels and Lugovskyy (2009) report the distance elastic- ity of the c.i.f./f.o.b. ad-valorem measure of maritime transportation cost to be between 0.003 and 0.15 depending on the dataset and sample selection. Hummels and Skiba (2004) reported the dis- tance elasticity of the per-unit maritime freight rate in the range between 0.11 and 0.26. Hummels et al. (2009) reported this elasticity to be under 0.18. For comparison, they showed that the average product-level prices from different origins explain up to 94 times more variation in freight rates than distance.16 Stylized Fact 11. Distance elasticity of freight rate is low: across studies, it varies between 0.03 and 0.26. We further extend this analysis by estimating transport costs’ distance transportation using the 15 Disdier and Head (2008) considered the robust distance coefficient as a puzzle, definitely not explainable by trans- portation costs alone. Chaney (2013) stated that previous models were not able to explain the distance coefficient; instead, he explained it with the effect of distance on the costs of creating contacts. 16 For Latin American importers, the 95th percentile f.o.b. price of a given HS6 good demands an 846% greater freight rate than the 5th percentile price; instead, the freight rate is only 9% greater when comparing the 95th and 5th distance percentiles (Hummels et al., 2009, Table 7). 15 1989-2018 US Census imports data and showing that the low elasticity further decreases over time. To this goal, we first estimate the sea-distance elasticity of ad-valorem and unit-cost maritime transport costs τodkt at the HS6 level k: I Modkt ln(τodkt ) = γ0 + γd ln(SeaDistod ) + γ2 ln Yot + γ3 LANGodt + γ4 FTAodt + γ5 ln + θt + νk + ε odkt , Wgtodkt (2) where subscripts o, d, k, t denote country of origin, destination port, product and year, respectively; Y , ( LANG ), ( FTA) are time-varying gravity equation variables from the “Dynamic Gravity Dataset" denoting GDP, common language, and free trade agreement, respectively; (SeaDist) is the CERDI I Modkt country level bilateral sea distance;17 Wgtodkt is the ratio of import charges (CIF) to import weight measured in kilograms; and θt and νk are the year and HS6 product fixed effects, respectively. To explore how it varies over time, we then added the distance-year interaction terms, ln(SeaDistod ) × θt . The Chow-Test confirmed that it is time-variant since the interaction terms significantly improve the model. The full set of estimates is reported in Appendix Table 6. Figure 9 illustrates the results with the year-specific elasticities. The left panel demonstrates that the average decline in the distance elasticity of ad-valorem transportation cost is approximately −0.003 per year or −0.09 over the entire thirty-year interval. The downward trend was even more pronounced for the distance elasticity of the per-unit transportation cost. As demonstrated by the right panel, this elasticity fell, on average, by −0.007 per year or −0.21 over the same three decades. To the best of our knowledge, the decreasing trends in these elasticities and the negative distance elasticity of the transportation costs have not been documented before. This trend is paralleled by the increasing share of low-income and remote countries in global trade. Since low-income and re- mote countries specialize in lighter goods (Lashkaripour, 2020), their growing share may cause the distance elasticity to decline. A more comprehensive explanation of the decreasing trend requires further investigation. To summarize: 17 See https://halshs.archives-ouvertes.fr/halshs-01288748/document. 16 Figure 9: Changing cost/seadistance over time. Source: US Census Bureau, Dynamic Gravity Dataset Gurevich et al. (2018), and authors’ calculations. The left panel, “Ad-valorem," uses the log-ratio of import charges relative to import values (e.g. the CIF-FOB ratio) as a dependent variable. The right panel, “Unit-cost," uses the log-ratio of import charges relative to import weight (in kilograms). Gray bands correspond to ± two (2) standard errors associated with the elasticity estimate. Standard errors are computed via the Delta-method. Stylized Fact 12. Between 1989 and 2019, the sea-distance elasticity of the US transport costs is both low in magnitude and volatile. Within this period, it decreases from 0.2 to 0.1 for the ad-valorem and from about 0.1 to −0.11 for the per-unit transport cost. 3.2 Transportation Costs as the Difference between the c.i.f. and f.o.b. Import Values Several studies have also employed measures of transportation costs constructed using matched partner c.i.f./f.o.b. ratios from IMF and UN data. Hummels and Lugovskyy (2009) found that the ratios from these sources are subject to significant measurement error and are unrelated to actual variation in shipping costs. In a similar vein, Feenstra and Romalis (2014) use the United Nations 17 COMTRADE database to obtain iceberg transport cost by taking the ratio of the importing country’s c.i.f unit values to exporting country’s f.o.b unit values for each 4-digit SITC product for many coun- tries between 1984 and 2008. While this methodology allows for the measurement of transport costs to vary across countries, product, and time, Feenstra and Romalis (2014) reported that there is a large amount of measurement error in the unit values, and there are many instances when the c.i.f unit value is less than the f.o.b. unit value. Stylized Fact 13. Matched partner c.i.f./f.o.b. ratios from IMF and UN data are often inaccurate measures of the ad-valorem freight rates. Recently, the official reporting guidelines for international trade recommend that countries re- port bilateral trade distinguished by five different modes of transport. Thus, the COMTRADE Plus database reports f.o.b. and c.i.f values by mode of transportation. In December 2020, UNCTAD, in collaboration with the World Bank, launched the Global Transport Cost database that has informa- tion on transport costs for 105 importing countries and 200 exporting countries. This dataset has the most extensive coverage of any public transport data so far, but it is available only for 2016. The system that has been put in place will allow for more updates and expansion in the future. 3.3 Direct Measures of Transportation Costs Several data sets contain direct measures of transport costs. We start our discussion of these data with the product-level customs data sets. The United States, New Zealand, and 11 Latin American countries report data on freight expenditures (including insurance) at 10-digit HS product level that are collected using importers’ customs declarations. These data provide a direct measurement of transport costs by mode of transportation together with trade values, quantities, and port of entry (for the US data). Several studies have employed this data to understand the trends in transport costs since the 1950s (Hummels, 2007; Hummels and Schaur, 2010), to model importers’ choice between air and ocean (Hummels and Schaur, 2013), to explore the nature and the determinants of international transport costs and the effect of freight rate on the patterns and quality of trade (Hummels and Skiba, 18 2004; Hummels et al., 2009; Lugovskyy and Skiba, 2015). In addition, the Maritime Transport Costs (MTC) database, prepared by OECD, combines the above-mentioned data on freight expenditures with the COMTRADE dataset and other data sources to provide a comprehensive transport costs database for 43 importing countries from 218 countries at 6-digit HS product level from 1991 to 2007 (Korinek, 2011). Another useful dataset comes from the U.S. Army Corps of Engineers(ACE). It focuses on the U.S. waterborne transport and includes freight expenditures at HS4-product level and U.S. ports of loading and unloading. These data have been used to study the effect of policy on maritime transport costs (Fink et al., 2002) and to estimate port efficiency and its impact on trade flows (Blonigen and Wilson, 2008, 2018; Clark et al., 2004). More recently, researchers can obtain customs data from some countries with freight expendi- tures at the transaction level. For example, Panjiva, a part of S&P Global, provides these data for Chilean and Colombian imports from 2009 onward. These data also contain trade values, quantities, mode of transportation, port of loading and unloading, shipment weight and volume, the identity of the importing firm, information on whether the importing firm or exporting firm is responsible for arranging transportation, and the identity of the carrier or logistic firm arranging transportation. The data on freight expenditures collected from custom declarations or bills of ladings is a direct measurement of transport cost. Therefore, these datasets represent the best sources of data to study the size and nature of maritime transport costs. However, they are available only for a small number of countries. Panjiva also provides the US import bills of lading. These data scope out a large portion of a shipment’s journey; a typical observation includes where a shipment originated, ports of lading and unlading, vessel-specific IMO Number, where the shipment is ultimately sent within the US (shipment destination), and some product characteristics (e.g., shipment HS8 code, TEU volume, and weight). Unfortunately, Panjiva’s US Imports lack direct measures of freight cost, and the value of the shipments are reported for a striking minority of observations. Another useful data supplier is Clarksons Research—the world’s leading shipbroker. Its Shipping 19 Intelligence Network database provides data on vessel-specific fixtures, namely time charter and spot rates. Vessel-specific rates are not particularly representative of a given market segment (i.e., containerships, dry bulk, tankers, auto carriers, etc). For example, Brancaccio et al. (2020) report that Clarksons tend to oversample newer and medium-sized vessels in dry bulk freight markets. That said, Clarksons does provide a representative summary of the freight earnings time series (from an annual to a weekly reporting frequency) for main trade routes and vessel benchmarks: e.g., Feeder Containership 2,7500 TEU gearless 6-12 month Timecharter rates ($/day) or SCFI Shangahi-West Coast America (spot) freight rates ($/FEU). Another readily available data source of transport costs for large cross-sections of countries is the Doing Business (DB) survey, compiled by the World Bank.18 It contains information on international transport for 189 countries since 2006. The section "Trading across Borders" provides data on the number of documents, time, and import and export costs. More specifically, it reports the US dol- lar cost per container to import and export. The data in the DB survey is collected using a World Bank questionnaire completed by trade facilitators and freight-forwarders across countries. For the comparability across countries, the survey presents the respondents with a detailed, stylized import or export case. The cost per container is for dry-cargo, 20-foot, full container load, and the products shipped do not require any special standards. Given the description in the survey, the most likely product categories are textile yarn and fabrics (SITC 65), articles of apparel and clothing accessories (SITC 84), and coffee, tea, cocoa, spices (SITC 07). The respondents are also asked to identify the likely port of loading (Djankov et al., 2010). Therefore, the variation in transport costs across prod- ucts, routes, or firms is rather limited (Blonigen and Wilson, 2018). The advantage of these data is that it allows for a comparison across many countries and some variation over time in the cost of a container and provides a more accurate measure of transport costs than, for example, the previously discussed c.i.f over f.o.b ratio or physical distance. Finally, some studies use more focused data. For example, Simonovska (2015) used the DHL ship- ping rates, Limao and Venables (2001) used shipping rate quotes for a standard 20-container from 18 The survey has been recently discontinued. See WB Statement for more details. 20 Baltimore, Kleinert and Spies (2011) use shipping prices from the UPS. These data are not generally available and focus on the cost of shipping for standardized units of cargo instead of the cost of shipping specific products. 3.4 Other Useful Data Sources In studying the determinants of shipping costs to the U.S. and port efficiency, Clark et al. (2004) used quality of port infrastructure and port handling costs from the Global Competitiveness Report. The Global Competitiveness Report, published annually by the World Economic Forum, provides the quality of port infrastructure index for 140 countries between 2007 and 2017. The index mea- sures business executives’ perception of their country’s port facilities in terms of development and accessibility. These data are collected through online surveys or in-person interviews. The Service Trade Restrictiveness Index developed by the OECD measures 44 countries’ restric- tions on service sectors, including maritime transportation, from 2014 to 2020 (the value of the index has little variation over time). This index allows researchers to study the cross-country variation in maritime sector policy restrictions and their impact on transport costs (Bertho et al., 2016). Using data from surveys of logistic professionals, the World Bank provides The Logistic Perfor- mance Index (LPI) that ranks countries on six dimensions of trade. The index includes information on customs performance, infrastructure quality, and timeliness of shipments. The LPI is more use- ful for a cross-country comparison in trade facilitation. It does not allow researchers to identify the international transport aspects of trade separately from the domestic infrastructure. We end this section by providing basic facts about selected datasets in Table 1. 21 Table 1: Databases providing international transport data Database Institution Country Frequency Aggregation Time Measure of Coverage Level Transportation BlueWater BlueWater Global Biweekly Liner 2012- TEU capacity per Reporting Reporting Rotation rotation/carrier/vessel COMTRADE United Nations 200+ Annual HS6 1984- c.i.f and f.o.b values countries Global Transport United Nations, 105 importing, Annual HS6 2016 c.i.f and f.o.b values Costs Dataset World Bank 200 exporting transport cost by mode and countries U.S. Imports of Mer- U.S. Census Bureau U.S. Monthly HS10 1990- freight expenditures chandise Annual by mode, entry port International Trans- ECLAC 11 Latin Ameri- Annual SITC 1999- freight expenditures port Database (BTI) can countries Rev. 3 by mode New Zealand Cus- New Zealand Cus- New Zealand Monthly, HS10 freight expenditures tom Data toms Service Annual U.S. Waterborne U.S. Army Corps of U.S. Annual HS4 2000- freight expenditures by U.S. Commerce Data Enginneers (ACE) 2019 port of (un)loading 22 Panjiva Panjiva Chile Monthly, HS8, 2009- freight expenditures by mode, Colombia Annual Firm level port of (un)loading Panjiva Panjiva U.S. Monthly, HS8 2009- by mode, Annual Firm level port of (un) loading Shipping Intelli- Clarksons Research Global Annual- Aggregate, 1989- Spot/Time gence Network Weekly route, Charter Rates vessel Global Competitive- World Economic Fo- Annual Aggregate 2007- port quality ness Report rum 2017 Doing Business(DB) World Bank 189 countries Annual Aggregate 2006 - US dollar cost per container 2015 Service Trade OECD 44 countries Annual Aggregate 2014- maritime transport policy re- Restrictiveness strictions Index (STRI) Notes: 11 Latin American countries covered by BTI are: Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Paraguay, Peru, Uruguay, Venezuela The Waterborne Commerce Data is available at https://www.iwr.usace.army.mil/About/Technical-Centers/WCSC-Waterborne-Commerce-Statistics-Center-2/WCSC- Waterborne-Commerce/ 3.5 Ad-valorem vs. Specific Transportation Costs As discussed above, the cost of transporting goods can be expressed in an ad-valorem (or iceberg) Freight Charges form as a percentage relative to the value of the transported goods: τ = Value of Transported Goods . Al- ternatively, it can also be expressed in a specific form, as transportation charges per unit or weight: Freight Charges f = Quantity (or Weight) of Transported Goods ; or as a mix of ad-valorem and specific components of trans- portation cost. An empirically-relevant way to distinguish between these forms is by evaluating ∂f p the price elasticity of freight rate, ∂p f , where p is the factory (i.e., exclusive trade costs) price of the good and f is the per-unit freight rate as defined above. The unitary elasticity corresponds to the ad-valorem form, zero elasticity—to the specific form, and the elasticity between zero and one—to having both components present. This section will discuss the literature utilizing both measures and the importance of the functional form of transportation cost for international trade. For many decades, the workhorse empirical model to estimate the effect of trade costs on bilateral trade was the gravity equation. A rich line of literature derives the gravity equation from a variety of theoretical models (e.g., Krugman, 1980; Eaton and Kortum, 2002; Melitz, 2003). These models assume a convenient ad-valorem form of transportation costs, also known as the Samuelson iceberg. Therefore, the ad-valorem formulation became a natural default for transportation costs in the em- pirical literature. In addition, most datasets with the direct measure of transportation charges report the value but not the quantity of the imported goods.19 This further strengthened the reasons to utilize the ad-valorem transportation cost as the per-unit measure could not be calculated without the information on the units. The knowledge about the functional form of transportation costs has been greatly enriched by a seminal work by Hummels and Skiba (2004), who utilized the direct measures of transportation costs using the data from six countries (Argentina, Brazil, Chile, Paraguay, Uruguay, the US) importing from the rest of the world. They showed that the data do not support the widely assumed ad-valorem 19 In the past, several studies have also employed measures of transportation costs constructed using matched partner c.i.f./f.o.b. ratios from the IMF Direction of Trade Statistics (DOTS) and UN COMTRADE data. Hummels and Lugovskyy (2009) found that the ratios from these sources are subject to significant measurement error and are unrelated to actual variation in shipping costs. 23 form of the transportation costs and that the freight rate contains the specific component.20 They also were the first to point out that the assumption of ad-valorem transport costs is not innocuous as it underplays the effects of transportation costs and tariffs on the quality composition of exports and imports. This paper has instigated the importance of the functional form of the transportation cost for multiple aspects of trade. Feenstra and Romalis (2014) used the specific transport costs to estimate traded quality. Due to the presence of the specific component in transport costs, remote economies were predicted and found to specialize in higher quality by Lugovskyy and Skiba (2015) and in lighter varieties by Harrigan (2010) and Lashkaripour (2020).21 Wong (2021) showed that specific transport costs reduce welfare and trade by more than an equal-yield ad-valorem transport costs because they are more price-distorting. We summarize the advantages and disadvantages of the ad-valorem measure as follows. Stylized Fact 14. The advantages of the ad-valorem measure of transport costs are: (i) it is in line with many trade models (ii) it is available for many countries and years; and (iii) it is intuitive and comparable with tariffs. At the same time, the functional form of the transportation cost contains a significant specific (per unit) component. This fact is important for estimating the traded quality and welfare gains from trade. The data on the per-unit measure of transport costs are relatively sparse. Earlier papers on the specific freight rates explored the determinants of the per-weight freight rates. Mainly because in the BTI dataset—the pioneering dataset for measuring specific freight rates—other units were not reported. Lashkaripour (2020) made a point that consumers buy units, not kilos of manufactured goods, and thus the per-unit measure is more relevant from the perspective of trade models. The per-unit freight rates are available only in a handful of datasets: e.g., the US Imports by Census and 20 They estimated the price elasticity of freight rate to be 0.6. The elasticity consistent with the ad-valorem functional form would have to be equal to one. 21 Theoretically, remote in this case refers to the higher transportation costs to the destination markets. Due to the lack of transportation costs for many countries, the measure of remoteness is often based on the average distance or trade- weighted average distance. Lugovskyy and Skiba (2015) constructed the MQC (Multilateral Quality Compensation) index that combines the trade-weighted importer’s preferences for quality and the composition of transportation costs at the exporter-HS6 level and discusses their distribution (see their Fig 1). 24 transaction-level Chilean and Colombian imports by Panjiva or Datamyne. The distinction between the per unit and per kilo freight rates is also important because, for the majority of products, more expensive goods are of greater weight (Lashkaripour, 2020). This may explain why the per-unit transportation costs have some ad-valoremness embedded in them. The binding constraint for a given container is almost always volume, not weight (Holmes and Singer, 2018). Therefore, it is desirable to have the volume per unit to better understand freight rates across goods and varieties of the same good. However, the volume measure data are extremely rare for manufactured goods. The only two papers utilizing the volume per unit are: Holmes (1989), who use Bills of lading filed with the U.S. Department of Customs and Border Protection (CBP); and Ardelean and Lugovskyy (2021), who use Chilean Imports reported by Panjiva. Figure 10: Chile Imports, 2009-2019 (Source: Panjiva) To shed new light on how different measures of freight rate compare to each other, we compare levels and trends for three measures of freight rates for Chilean imports between 2009 and 2019: ad- valorem, specific per weight, and specific per volume. These freight rates are calculated for three types of goods: manufactured goods transported in regular containers, goods requiring low temper- atures transported by refrigerated containers, and dry-bulk commodities transported by dry-bulk vessels. The data are presented in Figure 10, which illustrates several essential facts.22 22 Surprisingly, the reefer per-unit cost is lower than that of the general cargo. One potential explanation can be trade 25 Stylized Fact 15. The ranking of freight rates depends on their measure. In ad-valorem terms, the dry- bulk freight rate is up to 5 times greater than the containerized (non-refrigerated) freight rate. However, the specific (per kilo) dry-bulk freight rate is up to 10 times lower than the containerized one. Stylized Fact 16. The freight rate trends of the containerized and dry-bulk freight rates are not syn- chronized, implying that the cost and demand shocks are either asynchronous or have asymmetric effects on transport costs across commodity types. Stylized Fact 17. The aggregate per-kilo and per-TEU trends for a given commodity type are synchro- nized. The last fact is important from a practical standpoint since per-kilo freight rate measures are available in many more datasets. These similarities in trends indicate that, in the absence of the per-TEU measures, the per-kilo trends are good proxies for the per-TEU trends. Finally, some datasets provide the per-container charges. In some cases, these datasets contain the spot-market, but not the annual-contract charges (e.g., Drewry, as described by Wong, 2020). The US Bills of Lading data provide per-container charges, but only for a limited sample of the shipments.23 These data limitations impede utilizing the per-container freight rates in the context of international trade. 4 Containerized Maritime Shipping: Freight Rate Determinants Containerized maritime shipping is a multi-billion industry with complex and sophisticated technol- ogy. While an exact functional form of transportation cost is difficult to derive, it includes both fixed and variable cost components. The variable cost depends on the market structure, network organiza- tion, distance and geography, volume of trade and directional imbalances, good- and variety-specific imbalance: e.g., empty reefer boxes are needed for the Chilean exports, and therefore the empty returns to Chile are very cheap. 23 These data have been used by Holmes and Singer (2018) and Terner (2021). The value of the transported goods is also often missing from the data. 26 characteristics, shipping technology, fuel costs, and economies of scale at the vessel level. We discuss these components in Sections 4.1–4.3. The fixed cost component is a function of the vessel prices and port infrastructure. We discuss the port infrastructure in Section 4.4.24 4.1 Market Structure: Network Organization, Industry Concentration, Annual- Contract versus Spot Markets Containership transportation is organized similar to bus services, with the routes labeled as rotations and ports serving as bus stops. Major rotations include multiple ports, with the total duration of a ro- tation often exceeding 60 days. Both ports and ships vary in their capacity and cost of transportation. As discussed above, the average containership has become much larger over the last decades. These larger ships require automatic and modern equipment such as engines and cranes to load and unload them. The developments in shipping technology have created the need for ports to develop their in- frastructure to handle these larger ships, such as special berths, bigger cranes, more storage space for the increased cargo size, and more developed and efficient port logistics. Since the investment in port infrastructure requires a large amount of capital and human resources, the largest container ports handled a disproportionate volume and value of world trade. Notteboom (2018) calculated that in 2015, the 20 largest container ports handled 312.5 million TEU, which is 45% of world trade. The network structure of the containerized shipping was explored in detail by Heiland et al. (2019) and Ganapati et al. (2020). The latter also showed that a handful of ports play a disproportionately significant role:25 Stylized Fact 18. Larger ports serve as hubs accommodating larger ships, making hub-to-hub trans- portation significantly cheaper and hubs essential for maritime trade. Most papers analyzing networks in liner shipping focus on cross-sectional data analysis. The literature is scant when it comes to documenting the network evolution of containerized shipping 24 The discussion on (i) crewing and registration and (ii) trade and transport facilitation is beyond the scope of this review. 25 They also document that, for a given leg, a 1% increase in traffic reduces transportation costs by over 0.05%. 27 over time. Ducruet and Notteboom (2012) tackle the evolution of shipping networks using Lloyd’s Marine Intelligence Unit data from 1996 and 2006. They argue that, over time, the shipping network converges to a structure with densely concentrated and connected neighborhoods of ports. These neighborhoods are distant from one another and are connected via hubs (see Appendix 6.1 for more details). Hoffmann and Hoffmann (2020) compare the global 2020 network structure of ports to that of 2006. One of their main findings is that the network center has shifted from Europe to Asia with the major shifts being associated with Chinese ports (e.g., Nansha, Ningbo, and Shekou). The amount of market power affects the freight rates in two ways. First, the markup over the marginal cost of transportation increases in market power: According to Hummels et al. (2009) it is lower on thick routes with more carriers and greater traffic. Second, larger traffic prompts carriers to use larger vessels, which minimizes the average cost of transportation (Asturias, 2020; Ganapati et al., 2020). Both effects work in the same direction lowering the observed freight rates on thicker routes with larger traffic. Potentially, over-reliance on larger ports may increase the risk of the lack of resilience of global transportation system to various shocks, which should become a topic of future research explorations. Containerized maritime transportation is very concentrated. In 2018, the top 9 carriers (Maersk, MSC, Cosco-OOCL, GMA CGM, Evergreen, Hapag-Lloyd, ONE, and Yang Ming) organized in three alliances controlled 80% of the market (Merk et al., 2018). Furthermore, there are many “thin” routes, i.e., routes served by either a single or very few carriers. Hummels et al. (2009) reported a high degree of the within-route concentration, with the majority of importer-exporter pairs served by three or fewer ships, often owned by a single carrier. Using BlueWater Research data for years 2012- 2021, Ardelean and Lugovskyy (2021), reported that up to 60% of rotations26 connecting to the US were served by 4 carriers or less. Naturally, we can expect relatively high markups charged by the carriers, especially on the thin routes (see Hummels et al., 2009, for a discussion of the variation in route-specific markups). 26 Rotationsare defined as ports on a given schedule, similar to having bus stops on a bus schedule. Rotations often include over two ports in multiple countries. 28 Stylized Fact 19. Containerized shipping is very concentrated. Carriers exercise market power by charging higher markups on thinner routes. From a legal perspective, the industry is unique in that collusive agreements were allowed and enforced by the US government until 1998. Specifically, up until 1998, all carriers within a given conference were expected to post a common freight rate for each route and commodity, with the deviations allowed only in the form of an independent action by a carrier, which had to be publicly announced and offered to all qualified shippers. Any secretive deviation from the common freight rate was subject to a stiff fine by the Federal Maritime Commission (FMC). Naturally, this type of price-fixing agreement, known as conference agreements, severely limited within-route price varia- tion in liner shipping (Clyde and Reitzes, 1995). From the modeling perspective, these agreements justified the assumption of symmetric transportation costs in international trade literature. Indeed, the vast majority of trade models assume transportation costs to be symmetric for a given route. The difference is only in whether they are symmetric in ad-valorem or specific fashion.27 In 1998, the Ocean Shipping Reform Act (OSRA) abolished the price-fixing requirement and dras- tically changed the market structure. Specifically, the vast majority of the US and worldwide markets quickly switched to the confidential annual contracts between carriers and trading firms (FMC, 2001; OECD, 2015). The discussion of the recommended freight rates is still allowed among carriers. How- ever, the recommended freight rates are not part of the contracts, so they cannot be legally enforced. Keeping contracts confidential made the detection of these deviations unlikely, which further weak- ens the possibility of enforcing the recommended rates by other means of repeated interactions.28 Therefore, the freight rate dispersion within a route became more likely. Unfortunately, the direct empirical analysis of these markets is problematic due to the confidentiality of the annual contracts, 27 For example, Krugman (1980); Eaton and Kortum (2002); Melitz (2003); Arkolakis et al. (2012) used ad-valorem transport costs, while Hummels and Skiba (2004); Feenstra and Romalis (2014); Wong (2021) used specific transport costs. Importantly, in all these papers, transport costs are symmetric for a given route. 28 Other non-rate-fixing cooperative agreements between carriers still exist. They include vessel sharing and slot char- tering arrangements and global alliances. These agreements are motivated by improving the quality and efficiency of maritime transport due to economies of scale, but they also raise antitrust issues (UNCTAD, 2018). 29 which blocks researchers’ attempts to get the direct contract data.29 In parallel to the annual contract market, there are spot markets for containerized shipping, which serve the residual demand, including smaller firms without annual contracts. In addition, the spot markets are better suited for empirical analysis as the relevant data are available to researchers (e.g., Wong, 2020). In the spot markets, information about the freight rates is widely available to all partic- ipants. As a result, the within-route freight rates in these markets tend to be unified, in line with how transport costs are viewed in trade models. Even the literature on the market power in shipping, trade imbalances, and network structure primarily tends to assume uniform within-route pricing. It focuses on explaining the variation in freight rates across the routes (e.g., Hummels et al., 2009; Asturias, 2020; Wong, 2020; Ganapati et al., 2020). Shipping practitioners and trading firms are vividly aware of the differences between the two markets and closely monitor them. Indeed, while we do not have the exact data on the shares of the spot and annual-contract markets, some studies indicate that the annual-contract market is a dominant one (FMC, 2001; OECD, 2015). The primary role of the annual-contract market is further supported by the anecdotal evidence from the Journal of Commerce (JOC), a reputable source in the transportation and logistics industry. According to the JOC, the Shanghai Containerized Freight Index spot rate has increased by striking 135% between the 4th quarter of 2019 and 2020. Mean- while, the average corresponding freight rates and earnings of individual carriers increased by much smaller amounts: the ONE’s average Pacific rate increased by 35% (JOC, Jan 29, 2021), while Happag- Lloyd’s average freight rate has increased by 9.5% in the fourth quarter with its annual freight rate increasing by only 4% (JOC, Jan 27, 2021). These facts prompt several observations. First, the annual contract market is much larger than the spot market.30 Second, by design, the annual-contract freight rates are much less volatile than 29 For example, Wong (2020) described her unsuccessful attempt to get the contract data. We are aware of only one public reporting of (aggregated) contract freight rates: UNCTAD (2021a) reported average inter-regional contract freight rates for years 2018-2020, obtained from TIM Consult Market Intelligence. 30 If the annual-contract ONE has not changed, the difference between the 135% and 35% implies that only 25.9% of the freight rates were determined on the spot market (35/135*100%). For the Happag-Lloyd’s, the share of the spot market is also likely to be low. Still, we do not have an apple-to-apple comparison since a 9.5% increase is an average increase across all of its routes, whereas the spot rates’ dynamics vary across regions. 30 the spot market rates—the fact which has not been explored in the trade literature. In practice, importers closely watch annual contract rates as they determine their trade volumes for the en- tire year.31 Third, if annual-contract markets dominate containerized shipping, the within-route price dispersion should be detectable in the transaction-level datasets, conditional on them reporting freight rates. Two papers have recently conducted this type of analysis: Holmes and Singer (2018) and Ardelean and Lugovskyy (2021). Stylized Fact 20. There are two types of containerized shipping markets: long-term contract market and spot market. The rates may differ significantly between the two. Both papers document pronounced freight rate dispersion and link it to the firm and shipment sizes: larger firms and larger shipments face lower freight rates. The papers use different datasets and provide different interpretations of these findings. Holmes and Singer (2018) use Bills of lading filed with the U.S. Department of Customs and Border Protection (CBP). They attribute the lower freight rates faced by larger trading firms to the container indivisibility and the ability of larger firms to pack shipments fuller than smaller companies.32 Ardelean and Lugovskyy (2021) use Chilean Imports reported by Panjiva. They explicitly model price-discriminating carriers with trading firms getting quotes from carriers, where each quote is subject to a fixed cost of processing. Larger firms have greater bargaining power in this setting as they have greater incentives to obtain more quotes. This advantage disappears if the number of carriers is limited so that all firms, large and small, obtain the same number of quotes. Empirically, they find that the 90th percentile importing firms pay up to 23% lower freight rates than the 10th percentile firms. In line with their theoretical predictions, they also find that price discrimination disappears in the thin markets—i.e., in markets with three carriers or fewer. These results indicate that reducing market power distortion in the upstream industry (containerized ship- ping) may amplify the informational friction distortion in the downstream industry (freight rate 31 See https://www.freightwaves.com/news/importers-face-massive-hike-in-ocean-contract-costs 32 The main advantage of their dataset is in observing container types and sizes and specifying whether a container is filled to capacity. The disadvantage is in not observing important variety-specific characteristics, such as price and weight per unit. 31 asymmetry across importers). This type of price discrimination also results in the asymmetric ac- cess to imports between larger and smaller importers, increasing the observed firm heterogeneity.33 Therefore, this result may have important implications for the welfare calculations and policy analy- sis (e.g., Han, 2021). Stylized Fact 21. Ceteris paribus, larger firms face lower freight rates than smaller firms. 4.2 Product Type and Variety Characteristics Different products demand different transportation costs as they vary in volume, units of measure- ment, dimensions, weight, time sensitivity, and import demand elasticity. Thus, for accurate mea- surement of transport costs, it is necessary to control for the product type. In the literature on the freight rate determinants, it is typically done with the fixed effects at the HS6, HS8, or HS10 product level. Even within a product, there is a significant variation in variety-specific characteristics, such as price, weight per unit, time sensitivity, and dimensions. Schott (2004) has shown that across the 1994 U. S. manufacturing imports, the within-product unit value dispersion was on average 24 times. Lashkaripour (2020) found that, for 95% of goods, price is positively correlated with the weight per unit. Not surprisingly, more expensive varieties of the same good are subject to higher transporta- tion costs. Indeed, most studies find the price elasticity of the freight rate to be between 0.4 and 0.6 (Hummels and Skiba, 2004; Hummels et al., 2009; Ardelean and Lugovskyy, 2021). Hummels et al. (2009) found that after controlling for the product type, variety price explains by far the most variation in freight rates. Ardelean and Lugovskyy (2021) demonstrated that omitting product and variety characteristics in freight rate regressions results in the pronounced omitted variables bias of other freight rate determinants. 33 In many models, that would increase the shape parameter θ in the Frechet productivity distribution of firms. 32 Stylized Fact 22. Product-level controls (at least at the HS6 level) and the within-product variety characteristics (e.g., price, weight per unit) are essential for the unbiased estimation of maritime freight rates. 4.3 The Effect of Distance on Freight Rates As discussed in Section 3.1, distance is often used as a proxy for transport costs, where the dis- tance effect is typically estimated as a single parameter for all goods and destinations elasticity of freight rates. However, there are at least two strong reasons why the effect of distance is likely to be heterogeneous. First, geographic distance between countries affects only a portion of the total transportation costs. This fact alone suggests an essential source of heterogeneity of the distance ef- fects on the total cost of transporting goods because the size of the distance-related portion depends on other factors such as port efficiency (Humphreys et al., 2019; Blonigen and Wilson, 2008) or fuel price (Brancaccio et al., 2021; Behar and Venables, 2011). Precisely, the total cost of delivering goods consists of voyage costs incurred en route (such as the cost of fuel) and terminal costs incurred at the point of departure or destination (such as the cost of unloading). Halaszovich and Kinra (2020) present indirect evidence of such heterogeneity by showing that the effect of geographic distance on trade is affected by the domestic port and road infrastructure. The imports and exports of countries with better port infrastructure are less affected by the geographic distance to trading partners. The road infrastructure weakens the estimated effect of distance on imports but strengthens the negative effect of distance on exports. Taken together, the results in Table 9 of their paper illustrate that the effects of distance on trade vary across countries. Second, the heterogeneity of the distance effects can also be caused by technology. For example, Behar and Venables (2011) find that effect of fuel costs, which are related to the distance, on the ship- ping costs is smaller for containerized cargo than that one for bulk cargo. This fact is consistent with the idea that a larger part of the cost of shipping containers is not related to the distance. Another reason for heterogeneity is the varying degree of market power (see our discussion in section 4.1). 33 Thus, the effect of distance on the freight rate will be mixed with the effect of switching to a different route, which potentially has a different degree of market power. Ignoring the resulting heterogeneity in distance effects is materially important if the research question goes beyond the average effect of distance. The average effect is sufficient to capture the negative impact of distance on trade. However, focusing on the average effect of distance is of less interest if we are interested in the effects of distance on particular industries and countries with specific and observed characteristics. The rest of the section focuses on determining the extent of the distance heterogeneity across countries and products. 4.3.1 Heterogeneous Distance Elasticity of Transport Costs: US Imports Using the US Census imports data, we explore whether the distance elasticity of transportation costs varies across export origins and products. For the country-specific heterogeneity, we leverage the fact that the US is geographically large, with ports allocated to different Custom Districts.34 We first calculate the sea distances between various countries of origin and each of the US Custom Dis- tricts.35 Next, we use these distances when employing the following model featuring the most direct approach to the estimation of the heterogeneity in distance elasticity: ICodkt I Modkt ln = β 0 + β 1 ln(seadistanceodkt ) + β 2 ln(shipsizeodkt ) + β 3 ln I Modkt Weightodkt + γ0 1{θo } + γ1 1{θo } ln(seadistanceodkt ) + τt + ηd + ε odkt (3) ICodkt I Modkt ln = β 0 + β 1 ln(seadistanceodkt ) + β 2 ln(shipsizeodkt ) + β 3 ln I Modkt Weightodkt + γ0 1{νk } + γ1 1{νk } ln(seadistanceodkt ) + τt + ηd + θo + ε odkt , (4) 34 Therefore, the same country-exporter to the US has sufficient variation in sea distances to the US in large part due to the difference in sea distances to the coasts. 35 These calculations are described in detail in Terner (2021). These distance measurements reflect the average traveled distance of containerships from country o to Custom District d across 2015-2017. 34 where o, d, k, t are subscripts for the country of origin, destination (US) Customs District, HS6 prod- IC uct, and year, respectively; IM is transportation cost expressed in ad-valorem terms, seadistance is IM the sea distance between o and d, shipsize , Weight is the value-to-weight ration of the shipment. The terms 1{θo } in equation (3) and 1{νk } in equation (4) capture the full set of origin and HS6 product fixed effects respectively. The interaction of these effects with the log(seadistance) provide direct es- timates of origin country and product specific differences in the distance elasticities. We present the summary of the two dimensions of distance elasticity heterogeneity in Table 2: country in column 1 based on (3) and HS6 product in column 2 based on (4). The results of estimation with origin country heterogeneity suggest an average distance elasticity of 0.116 with a standard deviation of 0.07. Our HS6-based model produces a lower average distance elasticity of 0.07 with a standard deviation of 0.02. Figure 11 plots both distributions. To summarize, Stylized Fact 23. The distance elasticity of ad-valorem transport cost faced by US importers is hetero- geneous across countries of origin and across HS6 products. Figure 11: Table 2 results in chart form. In the appendix, we also provide robustness checks of heterogeneity in distance effects on the OECD ad-valorem and unit cost maritime trade costs for two-digit HS2 level products (by mode of 35 Country Model HS6 Model (1) (2) (Intercept) −5.904∗∗∗ −5.968∗∗∗ (0.304) (0.284) log(seadistance) 0.116∗∗∗ 0.069∗∗∗ (0.017) (0.010) log(shipsize) 0.007 0.001 (0.010) (0.010) log(value-to-weight) −0.463∗∗∗ −0.459∗∗∗ (0.002) (0.003) AIC 816522.218 804847.066 BIC 816835.895 805840.379 Log Likelihood −408231.109 −402328.533 Num. obs. 256, 766 256, 766 Num. groups: country 66 sd: country (Intercept) 0.63058 sd: country×log(seadistance) 0.07421 sd: Residual 1.18527 1.14571 Num. groups: HS6 3973 sd: HS6 (Intercept) 0.32145 sd: HS6×log(seadistance) 0.01625 Included Fixed Effects: Year, District Year, District,Country Distance Elasticity CV: 0.466 0.079 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1, sd - standard deviation Table 2: Heterogeneity of Distance Elasticity (based on US imports) maritime transport) for 223 importers and 23 partners annually for 2006 and 2007. 4.4 Port Infrastructure Ports infrastructure is vital to the integration of countries in the global economy via international trade. Consequently, there is an ongoing discussion of port reform and challenges associated with it (Haralambides, 2017). This discussion requires careful estimation of port efficiency, its sources, and its effect on transportation costs. Blonigen and Wilson (2008) classify the literature on evaluating port efficiency into three major groups: (i) Production function approaches, (ii) Surveys of port effi- ciencies, and (iii) Estimating port efficiencies from international trade data. In what follows, we will 36 describe these literatures, compare some measures of port efficiency, and discuss the links between port efficiencies and transportation costs. 4.4.1 Production Function Measures and Cost Analysis This operations research study approach is based on the production function estimation and involves several linear-programming applications. It focuses on (i) identifying an efficiency frontier based on observed inputs and outputs; and then (ii) calculating the distance between a productive-unit (read: port) efficiency and the efficiency frontier. A popular linear-programming application includes the Data Envelopment Analysis (DEA), a non-parametric and deterministic approach sensitive to vari- able selection and measurement errors. The competing production function estimation approach is represented by the Stochastic Frontier Analysis (SFA)—a parametric and stochastic approach that requires careful decisions over the functional form of the production function. Blonigen and Wilson (2008) provide an informative literature review of this approach. The main advantage of estimating the port efficiency as the production function is in pinpointing the marginal capacity gained from investing in any particular part of the port infrastructure. This analysis then allows identifying the bottlenecks for any given port and performing an informed cost-benefit anal- ysis when evaluating various investments in ports. However, this type of analysis tends to omit the demand side and thus by itself is not providing a direct link with the transportation costs associated with utilizing the ports. Recent work by Herrera Dappe and Suárez-Alemán (2016) linked this approach to transporta- tion cost.36 This paper first estimated individual ports efficiency using the DEA methodology for each year37 and then merged the average DEA-methodology annual port efficiencies with the OECD Maritime Transport Cost dataset and UN Commtrade US exports from 2000-2007. Next, building on 36 This paper leverages the data on nearly fifty (50) ports across East Asia, the Middle East, Southeast Asia, and South- ern/Eastern Africa. Herrera Dappe et al. (2017) further focused on countries and ports around the Indian Ocean. 37 It also decomposed productivity changes into technological progress, pure efficiency, and scale efficiency. This de- composition is performed by the Malmquist Total Factor Productivity (TFP) Index a la Caves et al. (1982). Other papers, such as Suárez-Alemán et al. (2016), also use the Malmquist TFP Index to track and decompose sources of productivity gains over longer port-efficiency times series. 37 Fink et al. (2002), they regressed (log) transport price p on the (log) marginal transport cost mc and (log) markup µ with mc being a function of the average country-specific port efficiency and µ being determined by the product fixed effects and liner connectivity index:38 p jkt = mc( j, k, t) + µ( j, k, t) where subscripts j, k, and t index country-importer, HS2 product, and year, respectively. Their estimation results indicate that a 0.1 increase in port efficiency reduces the (unit) maritime transport cost of exporting to the US by 2.3 percent. Put differently, moving from the 25th to the 75th percentile reduces transport costs by 3.5%. With similar data sources and empirical strategies, Herrera Dappe et al. (2017) extended this analysis by focusing on the impacts of port efficiency on bilateral trade flows with the US: The same 25th to the 75th percentile efficiency increases the value of exports to the US by roughly 0.56 percent. Furthermore, Herrera Dappe et al. (2021) find that increasing port efficiency (based on Stochas- tic Frontier Analysis) from the 25th-75th percentile lowers transport costs by approximately 4 per- cent. Their point estimate of the unit transportation cost elasticity with respect to efficiency, namely −0.23, is consistent with the Data Envelopment Analysis efficiency measures of Indian/Western Pa- cific Ocean ports which are between −0.22 and −0.24 (Herrera Dappe and Suárez-Alemán, 2016; Herrera Dappe et al., 2017). 4.4.2 Surveys of Port Efficiencies and Linking Them to Transportation Costs Clark et al. (2004) used the US Import Waterborne Databank alongside the World Economic Forum’s 1999 Global Competitiveness Report (WEF GCR hereafter) to show that moving from the 25th to the 75th percentile of port efficiency reduces transport costs by approximately 12 percent. Their main variables were: (i) liner transportation charges per weight at the HS6 product level and (ii) Country- 38 Marginal cost is also a function of (i) containerization, (ii) transportation distance multiplied by fuel costs, (iii) economies of scale, (iv) trade imbalances, (v) value per weight, (vi) national income (GDP), (vii) average country-specific port efficiency, and(viii-x) country, commodity, and time fixed effects. 38 level GCR port efficiency scores, indexed from one to seven (7 is the highest score).39 To illustrate how their analysis may be applied to predict trade flows, the authors developed a country-specific maritime cost index (driven to a large degree by WEF GCR port efficiency) and used it in the standard bilateral gravity model of trade. They concluded that increasing maritime transport costs from the 25th to the 75th percentile would decrease bilateral trade by 22 percent. 4.4.3 Transport Costs as Measures of Port Efficiency Blonigen and Wilson (2008) expanded upon Clark et al. (2004) by estimating port-specific efficiency rates. Whereas Clark et al. (2004) utilizes country level measures of port efficiency, which are based on survey results that are unsuitable for longitudinal analysis, Blonigen and Wilson (2008) used time- varying port fixed effects to capture an individual port’s contribution to maritime shipping costs. Using US import data sourced from the National Data Center of the Army Corps of Engineers, they produced estimates for the top fifty (50) American ports and top one hundred (100) foreign ports (by import value from the US perspective). Foreign port efficiency measurements are next aggre- gated into a country-level estimate and then are used in a standard gravity trade model. Without controlling for country-level fixed effects, the elasticity of bilateral trade with respect to port effi- ciency is 1.27. Since port efficiency estimates are time-varying, country fixed effects can be included. Thus, controlling for unobserved country-specific confounds reduces the bilateral trade port effi- ciency elasticity to 0.32. Based on these, moving a country from the 25th to 75th percentile of port efficiency increases bilateral trade by a more modest 5 percent. Feenstra and Ma (2014) used this approach to measure trade facilitation. They used port infras- tructure to analyze the extensive margin—or product variety—on both import and export trade. A 10 percent increase in port efficiency increase export variety by 1.5-3.4 percent but only increases the intensive margin by 0.2-1 percent; this result is broadly consistent with a literature, best exemplified by Bergin and Lin (2008), which documents how and why intensive and extensive margins respond 39 The scores included a country’s overall level/quality of infrastructure, cargo handling restrictions, mandatory port services, and severity of the organized crime. These four factors can explain more than 77 percent of WEF GCR variation. 39 differently to the same reduction in trade costs. Blonigen and Wilson (2018) utilize a longer time window (e.g., 1991-2009) over US import data to re-estimate their port efficiency measures. We compare their top hundred (100) non-US port mea- sures40 against the survey-based WEF GCR port infrastructure indicators. The results are presented in Figure 12 and they show strong correlation of ρ = 0.65 (with p.value < 10−5 ) our comparison of their Blonigen and Wilson (2008)’s country-level efficiency estimates against the WEF GCR indicators results in a twice weaker correlation of ρ = 0.33 (with p.value < 0.1). Figure 12: Country level port efficiency comparison Notes: x-axis values correspond to Blonigen and Wilson (2018) country level port efficiency estimates; the original fixed effects are transformed such that the x-axis measures the import charge (cost) improvement relative to Rotterdam (e.g. it costs roughly 40% more to send goods through Filipino ports relative to Rotterdam). y-axis values correspond to GCR quality of port infrastructure. 5 Concluding Remarks We conclude by discussing the gaps in the literature on the policy questions that need further re- search. One key question is: what is the optimal degree of regulation of liner shipping? Uninter- rupted, reliable shipping services are essential for world trade, especially in today’s highly integrated 40 We aggregated them to the national level by taking a weighted average using US import share as weights. 40 global economy. Shipping services became the first-order part of the overall infrastructure in our economy, comparable to other essential services such as electricity, international transportation, and communication. Providing ocean shipping services proves to be challenging because the demand is much more volatile than the ability of carriers to adjust capacity on the supply side. Therefore, on the one hand, carriers require a sufficient profit margin to absorb negative shocks while providing sufficient capacity during significant demand spikes. On the other hand, research documents signif- icant market power by carriers which they utilize for both across- and within-route discrimination across trading firms. Designing optimal policy becomes even more complicated in the international setting. Oceanic transportation is directly related to the environmental costs of international trade. Envi- ronmental regulation has had a negligible effect on the carriers’ costs and freight rates in the past. Therefore, the literature has mainly focused on estimating the extent of carbon emission by oceanic shipping, the related welfare analysis, and the comparison of emissions across transportation modes (Cristea et al., 2013; Shapiro, 2016). Any significant additional environmental regulations, such as, for example, operational efficiency regulations by IMO, will affect freight rates, fleet size, modes of transportation, technological choices, and eventually patterns of trade (UNCTAD, 2021a,b). New research can leverage new detailed datasets to understand better how changes in maritime freight rates will affect the technological and geographic (routes) choices of carriers and trading firms on the intensive and extensive margins, including the choice of the mode of transportation. References Adina Ardelean and Volodymyr Lugovskyy. It pays to be big: Price discrimination in maritime shipping. Mimeo, Indiana University, Bloomington, 2021. Costas Arkolakis, Arnaud Costinot, and Andres Rodriguez-Clare. New Trade Models, Same Old Gains? Amer- ican Economic Review, 102(1):94–130, February 2012. Jose Asturias. Endogenous transportation costs. European Economic Review, 123:1033–66, 2020. Alberto Behar and Anthony J. Venables. Transport Costs and International Trade. 2011. Paul R Bergin and Ching-Yi Lin. Exchange rate regimes and the extensive margin of trade. Technical report, National Bureau of Economic Research, 2008. 41 Dariusz Bernacki. Assessing the link between vessel size and maritime supply chain sustainable perfor- mance. Energies, 14(11), 2021. ISSN 1996-1073. doi: 10.3390/en14112979. URL https://www.mdpi. com/1996-1073/14/11/2979. Daniel M. Bernhofen, Zouheir El-Sahli, and Richard Kneller. Estimating the effects of the container revolution on world trade. Journal of International Economics, 98:36–50, 2016. Fabien Bertho, Ingo Borchert, and Aaditya Mattoo. The trade reducing effects of restrictions on liner shipping. Journal of Comparative Economics, 44(2):231–242, 2016. Bruce Blonigen and Wesley Wilson, editors. Handbook of International Trade and Transportation. Edward Elgar Publishing, 2018. Bruce A. Blonigen and Wesley W. Wilson. Port Efficiency and Trade Flows. Review of International Economics, 16(1):21–36, February 2008. Giulia Brancaccio, Myrto Kalouptsidi, and Theodore Papageorgiou. Geography, Search Frictions and Endoge- nous Trade Costs. Econometrica, 88(2):457–691, 2020. Giulia Brancaccio, Myrto Kalouptsidi, and Theodore Papageorgiou. The Impact of Oil Prices on World Trade. Boston College Working Papers in Economics 1030, Boston College Department of Economics, April 2021. URL https://ideas.repec.org/p/boc/bocoec/1030.html. Benjamin Bridgman. Why containerization didn’t reduce ocean shipping costs, at first. 2021. Douglas W Caves, Laurits R Christensen, and W Erwin Diewert. Multilateral comparisons of output, input, and productivity using superlative index numbers. The economic journal, 92(365):73–86, 1982. Thomas Chaney. The Gravity Equation in International Trade: An Explanation. NBER Working Papers 19285, National Bureau of Economic Research, Inc, August 2013. Ximena Clark, David Dollar, and Alejandro Micco. Port efficiency, maritime transport costs, and bilateral trade. Journal of Development Economics, 75(2):417–450, December 2004. Paul S. Clyde and James D. Reitzes. The Effectiveness of Collusion Under Antitrust Immunity. The Case of Liner Shipping Conferences. Bureau of economics staff report, Federal Trade Commission, December 1995. Anca Cristea, David Hummels, Laura Puzzello, and Misak Avetisyan. Trade and the greenhouse gas emissions from international freight transport. Journal of Environmental Economics and Management, 65 (1):153–173, 2013. doi: 10.1016/j.jeem.2012.06.00. URL https://ideas.repec.org/a/eee/jeeman/ v65y2013i1p153-173.html. Kevin Cullinane and Mahim Khanna. Economies of scale in large container ships. Journal of transport economics and policy, pages 185–207, 1999. E. Delhaye, T Breemersch, and K. Vanherle. Compass. the competitiveness of european short-sea freight ship- ping compared with road and rail transport. Final report; european commission dg environment: Brussels, belgium, 2010. Anne-Célia Disdier and Keith Head. The Puzzling Persistence of the Distance Effect on Bilateral Trade. The Review of Economics and Statistics, 90(1):37–48, February 2008. URL https://ideas.repec.org/a/tpr/ restat/v90y2008i1p37-48.html. 42 Simeon Djankov, Caroline Freund, and Cong S Pham. Trading on time. The Review of Economics and Statistics, 92(1):166–173, 2010. César Ducruet and Theo Notteboom. The worldwide maritime network of container shipping: spatial struc- ture and regional dynamics. Global networks, 12(3):395–423, 2012. Jonathan Eaton and Samuel Kortum. Technology, Geography, and Trade. Econometrica, 70(5):1741–1779, September 2002. Pablo Fajgelbaum and Stephen J. Redding. Trade, Structural Transformation and Development: Evidence from Argentina 1869-1914. Mimeo, Princeton University, 2021. Lixian Fan, Xinlu Li, Sijie Zhang, and Zimeng Zhang. Vessel size, investments and trade. In Handbook of International Trade and Transportation, pages 518–541. Edward Elgar Publishing Limited, 2018. Robert C Feenstra and Hong Ma. Trade facilitation and the extensive margin of exports. The Japanese Economic Review, 65(2):158–177, 2014. Robert C. Feenstra and John Romalis. International Prices and Endogenous Quality. The Quarterly Journal of Economics, 129(4):477–527, 2014. Carsten Fink, Aaditya Mattoo, and Ileana Cristina Neagu. Trade in international maritime services: How much does policy matter? The World Bank Economic Review, 16(1):81–108, 2002. FMC. The Impact of the Ocean Shipping Reform Act of 1998. Tech. report, Federal Maritime Commission, 2001. Sharat Ganapati, Woan Foong Wong, and Oren Ziv. Entrepôt: Hubs, Scale, and Trade Costs. Technical report, University of Oregon, 2020. Tamara Gurevich, Peter Herman, Nabil Abbyad, Meryem Demirkaya, Austin Drenski, Jeffrey Horowitz, and Grace Kenneally. The dynamic gravity dataset: Technical documentation, 2018. Tilo F. Halaszovich and Aseem Kinra. The impact of distance, national transportation systems and logistics performance on fdi and international trade patterns: Results from asian global value chains. Transport Policy, 98:35–47, 2020. ISSN 0967-070X. doi: https://doi.org/10.1016/j.tranpol.2018.09.003. URL https: //www.sciencedirect.com/science/article/pii/S0967070X17303670. Zheng Han. Price Discrimination in Transport Industry and the Gains from Trade. Technical report, SSRN: https://ssrn.com/abstract=3862530, 2021. Hercules Haralambides. Globalization, public sector reform, and the role of ports in international supply chains. Maritime Economics & Logistics, 19(1):1–51, March 2017. C Knick Harley. Ocean freight rates and productivity, 1740-1913: the primacy of mechanical invention reaf- firmed. Journal of Economic History, pages 851–876, 1988. James Harrigan. Airplanes and comparative advantage. Journal of International Economics, 82(2):181–194, November 2010. Keith Head and Thierry Mayer. Gravity equations: Workhorse, toolkit, and cookbook. In Handbook of interna- tional economics, volume 4, pages 131–195. Elsevier, 2014. 43 Inga Heiland, Andreas Moxnes, Karen-Helene Ulltveit-Moe, and Yuan Zi. Trade From Space: Shipping Net- works and The Global Implications of Local Shocks. CEPR Discussion Papers 14193, C.E.P.R. Discussion Papers, December 2019. Matías Herrera Dappe and Ancor Suárez-Alemán. Competitiveness of South Asia’s container ports: A comprehen- sive assessment of performance, drivers, and costs. World Bank Publications, 2016. Matias Herrera Dappe, Charl Jooste, and Ancor Suárez-Alemán. How does port efficiency affect maritime transport costs and trade? evidence from indian and western pacific ocean countries. World Bank Policy Research Working Paper, (8204), 2017. Matias Herrera Dappe, Tomás Serebrisky, and Ancor Suárez-Alemán. On the Historical Relationship between Port (In)Efficiency and Transport Costs in the Developing World. Interamerican development bank, 2021. Jan Hoffmann and Julian Hoffmann. Ports in the global liner shipping network: Understanding their position, connectivity, and changes over time. Tech. report, 2020. Thomas J. Holmes. The Effects of Third-Degree Price Discrimination in Oligopoly. The American Economic Review, 79(1):244–250, March 1989. Thomas J. Holmes and Ethan Singer. Indivisibilities in Distribution. NBER Working Papers 24525, National Bureau of Economic Research, Inc, April 2018. David Hummels. Transportation costs and international trade in the second era of globalization. Journal of Economic Perspectives, 21(3):131–154, September 2007. doi: 10.1257/jep.21.3.131. David Hummels and Volodymyr Lugovskyy. International pricing in a generalized model of ideal variety. Journal of Money, Credit and Banking, 41(s1):3–33, 02 2009. David Hummels, Volodymyr Lugovskyy, and Alexandre Skiba. The trade reducing effects of market power in international shipping. Journal of Development Economics, 89(1):84–97, May 2009. David L. Hummels and Georg Schaur. Hedging price volatility using fast transport. Journal of International Economics, 82(1):15–25, September 2010. David L. Hummels and Georg Schaur. Time as a Trade Barrier. American Economic Review, 103(7):2935–59, December 2013. David L. Hummels and Alexandre Skiba. Shipping the good apples out? an empirical confirmation of the alchian-allen conjecture. Journal of Political Economy, 112(6):1384–1402, 12 2004. Martin Humphreys, Aiga Stokenberga, Matías Herrera Dappe, and Olivier Hartmann. Port Development and Competition in East and Southern Africa: Prospects and Challenges. World Bank Publications, 2019. David S. Jacks and Martin Stuermer. Dry bulk shipping and the evolution of maritime transport. (NBER WP 28627), March 2021. Myrto Kalouptsidi. Time to build and fluctuations in bulk shipping. American Economic Review, 104(2):564–608, February 2014. doi: 10.1257/aer.104.2.564. URL https://www.aeaweb.org/articles?id=10.1257/ aer.104.2.564. 44 Joern Kleinert and Julia Spies. Endogenous Transport Costs in International Trade. IAW Discussion Papers 74, Institut für Angewandte Wirtschaftsforschung (IAW), July 2011. URL https://ideas.repec.org/p/ iaw/iawdip/74.html. Jane Korinek. Clarifying trade cost in maritime transport, 2011. OECD. Paul R. Krugman. Scale economies, product differentiation, and the pattern of trade. American Economic Review, 70(5):950–59, December 1980. Ahmad Lashkaripour. Weight-based quality specialization. Journal of International Economics, 127, 2020. Marc Levinson. The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger. Princeton University Press, 2008. Nuno Limao and Anthony J. Venables. Infrastructure, Geographical Disadvantage, Transport Costs, and Trade. The World Bank Economic Review, 15(3):451–479, October 2001. Volodymyr Lugovskyy and Alexandre Skiba. How geography affects quality. Journal of Development Economics, 115(C):156–180, 2015. Volodymyr Lugovskyy and David Terner. Shipping towards greener horizons: How market structure affects firm-level response to environmental regulation. Indiana University, Mimeo, 2021. Inmaculada Martínez-Zarzoso and Felicitas D. Nowak-Lehmann. Is distance a good proxy for transport costs? the case of competing transport modes. The Journal of International Trade & Economic Development, 16(3): 411–434, 2007. doi: 10.1080/09638190701527186. Marc J. Melitz. The impact of trade on intra-industry reallocations and aggregate industry productivity. Econo- metrica, 71(6):1695–1725, November 2003. Olaf Merk, Lucie Kirstein, and Filip Salamitov. The Impact of Alliances in Container Shipping. Technical report, OECD, International Transport Forum, 2018. Saif I Shah Mohammed and Jeffrey G Williamson. Freight rates and productivity gains in british tramp ship- ping 1869–1950. Explorations in Economic History, 41(2):172–203, 2004. Theo Notteboom. Trade and transport modes: the search for global connectivity through transport networks. In Handbook of International Trade and Transportation, pages 541–566. Edward Elgar Publishing Limited, 2018. Theo Notteboom and Jean-Paul Rodrigue. Re-assessing port-hinterland relationships in the context of global com- modity chains. Routledge, 2007. Theo E. Notteboom and Bert Vernimmen. The effect of high fuel costs on liner service configuration in con- tainer shipping. Journal of Transport Geography, 17(5):325–337, 2009. OECD. Competition Issues in Liner Shipping. Note for item iv of the 59th meeting of the working party no. 2, OECD Secretariat, 2015. Luigi Pascali. The wind of change: Maritime technology, trade, and economic development. American Economic Review, 107(9):2821–54, 2017. 45 Stephen Redding and Anthony J. Venables. Economic geography and international inequality. Journal of Inter- national Economics, 62(1):53–82, 2004. David H. Romer and Jeffrey A. Frankel. Does trade cause growth? American Economic Review, 89(3):379–399, June 1999. Peter K. Schott. Across-product versus within-product specialization in international trade. The Quarterly Journal of Economics, 119(2):646–677, May 2004. Joseph S Shapiro. Trade costs, co 2, and the environment. American Economic Journal: Economic Policy, 8(4): 220–54, 2016. Ina Simonovska. Income differences and prices of tradables: Insights from an online retailer. The Review of Economic Studies, 82(4):1612–1656, 2015. Martin Stopford. Maritime Economics, volume 3rd edn. London:Routledge, 2009. Ancor Suárez-Alemán, Javier Morales Sarriera, Tomás Serebrisky, and Lourdes Trujillo. When it comes to container port efficiency, are all developing regions equal? Transportation Research Part A: Policy and Practice, 86:56–77, 2016. Stig Tenold. Norwegian Shipping in the 20th Century: Norway’s Successful Navigation of the World’s Most Global Industry. Springer Nature, 2019. David Terner. Off Course? Knot anymore. Cost Efficiencies and the Panama Canal Expansion. Mimeo, 2021. Jan Tinbergen. Shaping the World Economy. The Twentieth Century Fund, New York, 1962. UNCTAD. Review of Maritime Transport. Tech. report, United Nations Conference on Trade and Develop- ment, 2015. UNCTAD. Review of Maritime Transport. Tech. report, United Nations Conference on Trade and Develop- ment, 2017. UNCTAD. Review of Maritime Transport. Tech. report, United Nations Conference on Trade and Develop- ment, 2018. UNCTAD. Review of Maritime Transport. Tech. report, United Nations Conference on Trade and Develop- ment, 2021a. UNCTAD. UNCTAD Assessment of the Impact of the IMO Short-Term GHG Reduction Measure on States. Tech. report, United Nations Conference on Trade and Development, 2021b. Woan Foong Wong. The Round Trip Effect: Endogenous Transport Costs and International Trade. Technical report, University of Oregon, 2020. Woan Foong Wong. The Round Trip Effect: Endogenous Transport Costs and International Trade. American Economic Journal: Applied Economics, 2021. 46 Figure 13: Panjiva Trends Figure 14: US use/share of world containership capacity Table 3: Dry Bulk Average Spot Rate ISO Code Economy Name TC rate DWT TC rate/DWT Count AE United Arab Emirates 13,623.29 59,745.92 0.24 73 AR Argentina 14,335.54 58,280.62 0.27 166 AG Antigua and Barbuda 13,000.00 61,150.00 0.21 1 AU Australia 11,140.62 70,412.25 0.19 16 BE Belgium 15,085.06 68,998.51 0.23 77 BJ Benin 16,250.00 42,805.50 0.39 2 BD Bangladesh 13,142.21 56,433.39 0.23 77 BG Bulgaria 15,050.00 45,462.50 0.37 2 BH Bahrain 11,980.77 46,041.54 0.27 13 BO Bolivia, Plurinational State of 14,500.00 74,117.00 0.20 1 47 Table 3: Dry Bulk Average Spot Rate ISO Code Economy Name TC rate DWT TC rate/DWT Count BR Brazil 15,107.65 70,584.86 0.23 605 BN Brunei Darussalam 10,500.00 52,223.00 0.20 1 CA Canada 13,144.44 59,302.11 0.23 9 CL Chile 14,250.00 60,656.00 0.24 2 CN China 11,733.57 85,474.21 0.15 2,495 CI Côte d’Ivoire 21,750.00 59,589.50 0.36 2 CM Cameroon 19,375.00 50,629.50 0.41 4 CD Democratic Republic of the Congo 15,750.00 53,413.00 0.29 1 CG Republic of Congo 15,000.00 56,925.00 0.26 1 CO Colombia 16,094.29 62,087.26 0.29 35 CV Cabo Verde 11,583.33 48,251.00 0.24 3 CR Costa Rica 31,000.00 61,384.00 0.51 1 CU Cuba 19,333.33 31,686.67 0.63 3 DE Germany 15,684.12 77,911.42 0.21 149 DK Denmark 11,882.56 51,600.42 0.26 43 DO Dominican Republic 14,750.00 42,179.00 0.36 4 DZ Algeria 14,667.50 56,995.81 0.27 21 EC Ecuador 14,915.69 78,756.74 0.19 838 EG Egypt, Arab Rep. 17,569.03 73,046.89 0.25 88 ES Spain 15,806.75 74,979.96 0.23 126 FI Finland 14,100.00 37,985.00 0.37 1 FR France 14,880.95 68,593.94 0.24 85 GA Gabon 14,020.00 59,442.40 0.24 5 GB United Kingdom 15,205.85 77,411.61 0.21 124 GE Georgia 13,416.67 34,405.33 0.39 3 GH Ghana 18,285.71 58,057.57 0.32 7 GI Gibraltar 15,983.66 88,155.99 0.19 231 GN Guinea 10,501.32 69,028.05 0.16 19 GR Greece 14,387.50 51,635.89 0.31 28 GT Guatemala 16,500.00 38,246.50 0.43 2 HK Hong Kong, SAR China 10,770.85 76,276.70 0.15 293 HR Croatia 14,350.00 81,895.00 0.18 1 HT Haiti 14,000.00 56,982.00 0.28 2 ID Indonesia 11,270.00 64,128.69 0.18 270 IN India 13,881.78 70,522.49 0.20 782 IE Ireland 15,898.53 80,095.97 0.20 34 IR Iran, Islamic Republic of 16,150.00 56,655.00 0.28 2 IQ Iraq 17,958.33 60,391.17 0.30 6 IS Iceland 13,250.00 75,200.00 0.18 1 IL Israel 17,500.00 89,092.67 0.28 3 IT Italy 16,444.26 84,897.43 0.21 122 JM Jamaica 13,183.33 79,954.83 0.20 12 JO Jordan 16,325.00 62,847.25 0.26 4 JP Japan 12,267.62 80,874.57 0.16 421 KE Kenya 12,150.00 60,729.80 0.20 5 KH Cambodia 12,750.00 55,564.10 0.23 10 KR Korea, Rep. 11,234.89 84,478.56 0.14 427 KW Kuwait 13,930.77 55,481.38 0.25 13 LY Libya 45,000.00 174,316.00 0.26 1 LK Sri Lanka 12,237.93 74,459.17 0.17 29 48 Table 3: Dry Bulk Average Spot Rate ISO Code Economy Name TC rate DWT TC rate/DWT Count LT Lithuania 11,125.00 63,082.75 0.16 4 LV Latvia 12,500.00 37,100.00 0.34 1 MA Morocco 15,648.33 69,660.23 0.24 75 MX Mexico 16,057.14 43,079.14 0.38 7 ML Mali 12,750.00 56,452.00 0.23 1 MT Malta 16,875.00 79,612.00 0.21 2 MM Myanmar 9,950.00 56,951.57 0.18 7 MZ Mozambique 11,725.00 57,209.22 0.22 18 MR Mauritania 15,500.00 48,619.20 0.33 5 MU Mauritius 11,500.00 81,889.00 0.14 1 MY Malaysia 12,612.71 71,957.76 0.18 118 NA Namibia 12,000.00 57,945.00 0.21 1 NC New Caledonia 10,125.00 69,970.38 0.15 8 NG Nigeria 17,035.71 53,050.14 0.36 7 NL Netherlands 15,619.89 82,984.95 0.20 220 NO Norway 9,375.00 41,154.25 0.24 4 OM Oman 18,500.00 71,810.83 0.26 6 PK Pakistan 14,894.30 60,602.02 0.25 57 PE Peru 14,500.00 59,675.50 0.25 2 PH Philippines 12,489.61 69,267.46 0.18 308 PL Poland 18,052.56 74,680.23 0.25 39 PR Puerto Rico 13,875.00 45,995.50 0.32 2 PT Portugal 14,805.00 69,885.77 0.27 30 QA Qatar 10,975.00 48,714.83 0.23 6 RO Romania 12,931.25 48,255.50 0.28 16 RU Russian Federation 12,135.29 59,408.41 0.24 17 SA Saudi Arabia 16,093.75 52,346.83 0.35 24 SD Sudan 18,227.50 53,586.10 0.32 10 SN Senegal 17,068.75 43,760.12 0.41 8 SG Singapore 12,591.70 71,002.03 0.19 594 SL Sierra Leone 12,750.00 57,200.00 0.23 2 SV El Salvador 13,437.50 58,492.50 0.25 4 SR Suriname 12,500.00 32,616.00 0.38 2 SK Slovak Republic 11,708.33 81,401.75 0.15 12 SI Slovenia 31,000.00 90,266.00 0.35 2 SE Sweden 21,333.33 66,789.83 0.32 6 TG Togo 18,875.00 50,870.00 0.38 4 TH Thailand 10,268.98 56,779.10 0.19 108 TT Trinidad and Tobago 14,500.00 81,462.00 0.18 1 TN Tunisia 12,500.00 48,379.50 0.27 2 TR Turkey 15,022.26 57,638.42 0.27 266 TW Taiwan, China 11,091.10 81,341.17 0.14 292 UA Ukraine 33,062.00 112,093.56 0.28 25 US United States 16,065.54 63,978.53 0.27 436 VE Venezuela, RB of 15,250.00 53,368.60 0.29 5 VN Vietnam 11,816.67 66,780.01 0.19 204 YE Yemen, Rep. 13,250.00 56,220.33 0.23 3 ZA South Africa 12,783.28 69,796.60 0.19 290 49 6 Appendix 6.1 Additional Details and Terminology on the Networks in Shipping Ducruet and Notteboom (2012) explicitly model the evolution of containerized shipping over time and at a global scale. They use Lloyd’s Marine Intelligence Unit data to analyze containership fleet service allocation in 1996 and 2006. The authors construct containership networks via one of two approaches. The first approach is called a graph of direct linkages (GDL hereafter) whereby ports A and B are connected if there exists a direct connection between A and B (the orientation of the direction does not matter). The second approach is called a graph of connections (GAL hereafter) whereby ports A and C are connected if there exists any path that connects A and C; for instance, a port call sequence of A → B → C would translate to a GDL of port pairs ( A, B) and ( B, C ) whereas the same sequence would to a lead GAL graph with connections ( A, B), ( B, C ), and ( A, C ). GDL and GAL present competing views of the (i) fundamental character of containerized networks; and (ii) its evolution over time. GDL networks can be thought of as polarized/scale-free networks since the Power- law coefficient |γ| > 1; a small number of ports have a large number of connections (e.g. hubs or entrepôts) whereas the lion’s share of ports have few connections. GAL networks are better thought of as small-world networks: most ports are not necessarily (directly) connected but a port’s neighbors are likely to be neighbors (e.g. if port A has neighbors B and C, then B and C are likely to be neighbors of each other). In short, the GDL networks can be thought to represent a world in which hub ports dominate secondary ports, whereas GAL networks rationalize densely connected shipping regions. Table 4: Topological properties of the global maritime network Source: Table 2, (Ducruet and Notteboom, 2012, p.23) GDL GAL Percent Change Index Measure 1996 2006 1996 2006 GDL GAL No. vertices 910 1205 910 1205 32.4 32.4 No. edges 5,666 9,829 28,510 51,057 73.5 79.1 Max. degree 165 226 437 610 37.0 39.6 Avg. degree 12.787 17.027 64.178 87.521 33.2 36.4 Network size Total length (000s km) 5,159 10,813 71,835 130,927 109.6 82.3 Traffic density (TEU/km) 331 407 125 183 23.0 46.4 Max. edge length (km) 10,012 10,018 10,018 10,018 0.1 0.0 Avg. edge length (km) 1,008 1,227 2,900 2,997 21.7 3.3 Diameter 9 8 4 5 -11.1 25.0 Cycles Cyclomatic number 4,757 8,625 27,601 49,853 81.3 80.6 Lattice degree α 0.005 0.012 0.033 0.069 140.0 109.1 Complexity β 6.226 8.156 31.329 42.370 31.0 35.2 Connectivity γ 0.014 0.014 0.069 0.070 0.0 1.4 Scale-free Power-law coefficient -1.351 -1.293 -0.624 -0.647 -4.3 3.7 Avg. clustering coefficient (local) 0.540 0.545 0.744 0.734 0.9 -1.3 Small-world Transitivity (global) 0.266 0.266 0.404 0.435 0.0 7.7 Efficiency Avg. path length 3.253 3.189 2.230 2.219 -2.0 -0.5 In terms of changing “hub structure," Ducruet and Notteboom (2012) presents mixed evidence; the GDL 50 power law coefficient decreases from 1996 to 2006, which signifies that hubs play a less significant role in con- tainerized networks. Network Definitions/Measures • Vertex: Sometimes called a node, in context a vertex is a port. • Edge: Connections between vertices (ports), in context an edge is a scheduled delivery between two ports. • Degree: The number of edges (services) incident to a vertex (port). • Length: Sum of all edge (service) distance. • Diameter: Shortest distance (measured by the number of edges) between the two most distant vertices (ports). Put differently, once the shortest path is computed between all vertices in a graph, the diameter is the longest shortest path. • Cyclomatic number: A measure of graph complexity, M = E − N + 2 P, where E is the number of graph edges, N is the number of vertices, and P is the number of connected components. • α-index: The ratio of the number of circuits (e.g. r = E − N + p) in a network to the maximum possible in that network; values closer to unity (zero) indicate highly connected (disconnected) network. • β−index: the ratio of edges to vertices ( β = E/ N ) • γ−index: ratio of number of edges to the theoretical maximum possible number of edges. • Power-law coefficient: Scale coefficient γ that governs the degree distribution of the network; the proba- bility that a network has vertices of degree k is: ∞ −1 pk = Ck−γ where C = ∑ j−γ j =1 As |γ| → ∞, the more likely it is that single vertices (port) are hubs with large number of connections. • Avg. clustering coefficient : The expected ratio of vertex number of triangles (a sub-graph with 3 vertices and 3 edges) to the number of triples (the number of sub-graphs with 2 edges and 3 vertices); larger values signify densely connected local networks. • Transitivity: The ratio of 3 times the number of triangles in the network to the number of connected triple nodes in the network. • Avg. path length: The average shortest distance (measured by the number of edges) across all vertices. 6.2 Heterogeneous distance impacts on transport costs: OECD Data In this section, we confirm that country and product-based distance effect exists globally. To this goal, we depart from US imports and move towards exploring distance heterogeneity in international perspective. We assemble a new dataset using: • CERDI country level bilateral sea distance.41 • OECD ad-valorem and unit cost maritime trade costs for two digit HS2 level products (by mode of maritime transport) for 223 importers and 23 partners annually for 2006 and 2007.42 Theses data are restricted to HS2 goods transported by containership. 41 See https://halshs.archives-ouvertes.fr/halshs-01288748/document. 42 See https://stats.oecd.org/viewhtml.aspx?datasetcode=MTC&lang=en. 51 • UNCTAD liner shipping bilateral connectivity index (LSBCI).43 Our estimating equations mimic equations (3) and (4), but include the liner bilateral connectivity index to capture the network effects on transport cost: ln(tcodkt ) = β 0 + β 1 ln(seadistod ) + β 2 ln( LSBCIodt ) + γ0 1{θo } + γ1 1{θo } ln(seadistod ) + τt + ηd + νk + ε odkt (5) ln(tcodkt ) = β 0 + β 1 ln(seadistod ) + β 2 ln( LSBCIodt ) + γ0 1{νk } + γ1 1{νk } ln(seadistodkt ) + τt + ηd + θo + ε odkt (6) Results from equations (5) and (6) are reported in Table 5. Unit cost ($ per TEU) distance elasticities are a touch higher than their ad-valorem counterparts: a 10 percent increase in sea distance raises unit transport cost by 3.3 to 3.5% on average with a standard deviation between 0.3-0.4%. The ad-valorem distance elasticities are 0.083 and 0.164 respectively. Cost Type Unit Cost Ad-Valorem Slope Heterogeneity (Country) (HS2) (Country) (HS2) (Intercept) −4.063 ∗∗∗ −4.688∗∗∗ −4.190 ∗∗∗ −4.231∗∗∗ (0.286) (0.209) (0.321) (0.191) log(seadistance) 0.329 ∗∗∗ 0.345∗∗∗ 0.083 ∗∗∗ 0.164∗∗∗ (0.012) (0.012) (0.023) (0.012) log(LSBCI) −0.410∗∗∗ −0.397∗∗∗ −0.423∗∗∗ −0.463∗∗∗ (0.043) (0.045) (0.045) (0.041) AIC 104196.394 104384.098 96575.261 96908.891 BIC 105178.520 105814.211 97557.327 98338.918 Log Likelihood −51984.197 −52026.049 −48173.630 −48288.445 Num. obs. 40, 747 40, 747 40, 726 40, 726 Fixed Effects o , d, k , t o , d, k , t o , d, k , t o , d, k , t Var: i (Intercept) 0.066 3.061 Var: i.1 log(seadistance) 0.001 0.036 Var: Residual 0.736 0.736 0.601 0.613 Num. groups: Commodity 88 88 Var: Commodity (Intercept) 0.359 0.315 Var: Commodity.1 log(seadistance) 0.002 0.002 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1 Table 5: Country level maritime transport cost models (based on all imports) 6.3 Additional Tables 43 See https://unctadstat.unctad.org/wds/TableViewer/tableView.aspx?ReportId=96618. 52 Table 6: US Maritime Trade Cost Models (based on equation 2 with and without distance-time interaction terms ) Model No Interaction Interaction Dependent Variable ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln(seadistod ) 0.151∗∗∗ −0.006∗∗∗ 0.157∗∗∗ −0.002 (0.002) (0.000) (0.009) (0.001) ln( I Modkt /Wgtodkt ) −0.448∗∗∗ 1.036∗∗∗ −0.449∗∗∗ 1.036∗∗∗ (0.001) (0.000) (0.001) (0.000) ln(Yot ) −0.006∗∗∗ 0.006∗∗∗ −0.005∗∗∗ 0.006∗∗∗ (0.000) (0.000) (0.000) (0.000) LANGodt 0.036∗∗∗ −0.001∗∗∗ 0.038∗∗∗ −0.001∗∗∗ (0.002) (0.000) (0.002) (0.000) FTAodt −0.119∗∗∗ 0.003∗∗∗ −0.128∗∗∗ 0.004∗∗∗ (0.003) (0.000) (0.003) (0.000) 1990 0.003 −0.002∗∗∗ −0.121 −0.021 (0.006) (0.001) (0.119) (0.015) 1991 −0.017∗∗∗ −0.002∗∗ −0.330∗∗∗ −0.009 (0.006) (0.001) (0.118) (0.015) 1992 −0.048∗∗∗ 0.000 −0.461∗∗∗ 0.019 (0.006) (0.001) (0.118) (0.015) 1993 −0.046∗∗∗ −0.001 −0.437∗∗∗ 0.027∗ (0.006) (0.001) (0.116) (0.015) 1994 −0.035∗∗∗ −0.000 −0.283∗∗ 0.044∗∗∗ (0.006) (0.001) (0.115) (0.015) 1995 −0.041∗∗∗ −0.000 0.009 0.007 (0.006) (0.001) (0.114) (0.015) 1996 −0.088∗∗∗ 0.004∗∗∗ −0.092 0.034∗∗ (0.006) (0.001) (0.113) (0.015) 53 Table 6: US Maritime Trade Cost Models (based on equation 2 with and without distance-time interaction terms ) Model No Interaction Interaction Dependent Variable ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) 1997 −0.138∗∗∗ 0.009∗∗∗ 0.087 0.020 (0.006) (0.001) (0.112) (0.014) 1998 −0.161∗∗∗ 0.010∗∗∗ −0.136 0.067∗∗∗ (0.006) (0.001) (0.111) (0.014) 1999 −0.146∗∗∗ 0.006∗∗∗ −0.809∗∗∗ 0.126∗∗∗ (0.006) (0.001) (0.111) (0.014) 2000 −0.093∗∗∗ 0.002∗∗∗ −0.712∗∗∗ 0.110∗∗∗ (0.006) (0.001) (0.109) (0.014) 2001 −0.157∗∗∗ 0.004∗∗∗ −0.577∗∗∗ 0.104∗∗∗ (0.006) (0.001) (0.109) (0.014) 2002 −0.165∗∗∗ 0.007∗∗∗ −0.290∗∗∗ 0.081∗∗∗ (0.006) (0.001) (0.108) (0.014) 2003 −0.113∗∗∗ 0.002∗∗∗ −0.418∗∗∗ 0.074∗∗∗ (0.006) (0.001) (0.108) (0.014) 2004 −0.094∗∗∗ −0.001 −0.587∗∗∗ 0.084∗∗∗ (0.006) (0.001) (0.108) (0.014) 2005 −0.073∗∗∗ −0.003∗∗∗ −0.354∗∗∗ 0.071∗∗∗ (0.006) (0.001) (0.107) (0.014) 2006 −0.108∗∗∗ −0.001∗ 0.174 0.033∗∗ (0.006) (0.001) (0.107) (0.014) 2007 −0.139∗∗∗ 0.000 0.261∗∗ 0.042∗∗∗ (0.006) (0.001) (0.108) (0.014) 2008 −0.158∗∗∗ −0.000 0.261∗∗ 0.033∗∗ (0.006) (0.001) (0.108) (0.014) 54 Table 6: US Maritime Trade Cost Models (based on equation 2 with and without distance-time interaction terms ) Model No Interaction Interaction Dependent Variable ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) 2009 −0.246∗∗∗ 0.007∗∗∗ 0.424∗∗∗ 0.008 (0.006) (0.001) (0.109) (0.014) 2010 −0.241∗∗∗ 0.005∗∗∗ 0.231∗∗ 0.017 (0.006) (0.001) (0.109) (0.014) 2011 −0.251∗∗∗ 0.005∗∗∗ 0.349∗∗∗ 0.023 (0.006) (0.001) (0.108) (0.014) 2012 −0.258∗∗∗ 0.005∗∗∗ 0.253∗∗ 0.010 (0.006) (0.001) (0.108) (0.014) 2013 −0.264∗∗∗ 0.006∗∗∗ 0.278∗∗∗ 0.017 (0.006) (0.001) (0.108) (0.014) 2014 −0.302∗∗∗ 0.007∗∗∗ 0.254∗∗ 0.023 (0.006) (0.001) (0.108) (0.014) 2015 −0.283∗∗∗ 0.005∗∗∗ −0.038 0.044∗∗∗ (0.006) (0.001) (0.108) (0.014) 2016 −0.318∗∗∗ 0.009∗∗∗ 0.110 0.026∗ (0.006) (0.001) (0.107) (0.014) 2017 −0.340∗∗∗ 0.009∗∗∗ −0.339∗∗∗ 0.026∗ (0.006) (0.001) (0.107) (0.014) 2018 −0.347∗∗∗ 0.009∗∗∗ 0.004 0.013 (0.006) (0.001) (0.107) (0.014) ln(seadistod )×1990 0.014 0.002 (0.013) (0.002) ln(seadistod )×1991 0.035∗∗∗ 0.001 (0.013) (0.002) 55 Table 6: US Maritime Trade Cost Models (based on equation 2 with and without distance-time interaction terms ) Model No Interaction Interaction Dependent Variable ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln(seadistod )×1992 0.045∗∗∗ −0.002 (0.013) (0.002) ln(seadistod )×1993 0.043∗∗∗ −0.003∗ (0.013) (0.002) ln(seadistod )×1994 0.027∗∗ −0.005∗∗∗ (0.013) (0.002) ln(seadistod )×1995 −0.005 −0.001 (0.013) (0.002) ln(seadistod )×1996 0.000 −0.003∗∗ (0.012) (0.002) ln(seadistod )×1997 −0.025∗∗ −0.001 (0.012) (0.002) ln(seadistod )×1998 −0.003 −0.006∗∗∗ (0.012) (0.002) ln(seadistod )×1999 0.073∗∗∗ −0.013∗∗∗ (0.012) (0.002) ln(seadistod )×2000 0.068∗∗∗ −0.012∗∗∗ (0.012) (0.002) ln(seadistod )×2001 0.046∗∗∗ −0.011∗∗∗ (0.012) (0.002) ln(seadistod )×2002 0.014 −0.008∗∗∗ (0.012) (0.002) ln(seadistod )×2003 0.033∗∗∗ −0.008∗∗∗ (0.012) (0.002) 56 Table 6: US Maritime Trade Cost Models (based on equation 2 with and without distance-time interaction terms ) Model No Interaction Interaction Dependent Variable ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln(seadistod )×2004 0.054∗∗∗ −0.009∗∗∗ (0.012) (0.002) ln(seadistod )×2005 0.031∗∗∗ −0.008∗∗∗ (0.012) (0.002) ln(seadistod )×2006 −0.031∗∗∗ −0.004∗∗ (0.012) (0.002) ln(seadistod )×2007 −0.044∗∗∗ −0.005∗∗∗ (0.012) (0.002) ln(seadistod )×2008 −0.046∗∗∗ −0.004∗∗ (0.012) (0.002) ln(seadistod )×2009 −0.073∗∗∗ −0.000 (0.012) (0.002) ln(seadistod )×2010 −0.052∗∗∗ −0.001 (0.012) (0.002) ln(seadistod )×2011 −0.066∗∗∗ −0.002 (0.012) (0.002) ln(seadistod )×2012 −0.056∗∗∗ −0.001 (0.012) (0.002) ln(seadistod )×2013 −0.059∗∗∗ −0.001 (0.012) (0.002) ln(seadistod )×2014 −0.061∗∗∗ −0.002 (0.012) (0.002) ln(seadistod )×2015 −0.027∗∗ −0.004∗∗∗ 57 Table 6: US Maritime Trade Cost Models (based on equation 2 with and without distance-time interaction terms ) Model No Interaction Interaction Dependent Variable ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) ln( ICodkt / I Modkt ) ln( ICodkt /Wgtodkt ) (0.012) (0.002) ln(seadistod )×2016 −0.047∗∗∗ −0.002 (0.012) (0.002) ln(seadistod )×2017 0.000 −0.002 (0.012) (0.002) ln(seadistod )×2018 −0.038∗∗∗ −0.000 (0.012) (0.002) Num. obs. 2, 338, 744 2, 338, 406 2, 338, 744 2, 338, 406 R2 0.333 0.992 0.334 0.992 R2 (w/o FE) 0.165 0.982 0.165 0.982 Adj. R2 0.331 0.992 0.332 0.992 Adj. R2 (w/o FE) 0.162 0.982 0.162 0.982 FE: HS6 6, 885 6, 885 6, 885 6, 885 ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.1 58