The effect of maritime transport costs on the extensive and intensive margins: Evidence from the Europe–Asia trade

This article investigates the determinants of maritime trade. It focuses in particular on the extent to which variations in trade-related costs between Asia and Europe help to explain the surge in Euro–Asian trade in eight of the most emblematic categories of products related to Asian success: textiles, footwear, confection, machinery, electronic products, vehicles, furniture and pharmaceutical products. In marked contrast to other studies that focus only on the determinants of total maritime trade, we decompose trade into two margins: the number of different products exchanged (extensive margin) and the average value of each product (intensive margin). We estimate a trade-augmented gravity model with trade cost factors for specific trade flows and industries and for both margins of trade. Several types of trade costs are considered, namely maritime transport costs, time to export/import, behind-the-border trade costs and distances. The main findings indicate that lower freight costs increase aggregate trade values mainly by increasing the average value of imported varieties, but also by increasing the number of products traded. Our findings suggest that political actions aimed at spurring competition and innovation in the maritime transport industry do have an impact on the volume and composition of international trade.


I n t r o d u c t i o n
This article focuses on clarifying to what extent variations in trade-related costs between Asia and Europe help to explain the surge in Euro-Asian trade in eight of the most emblematic categories of products related to Asian success: textiles (knitted and not knitted), footwear confection, machinery, electronic products, vehicles, furniture and pharmaceutical products. Several categories of trade costs are considered, namely maritime transport costs (MTC), time to export/import, behind-the-border trade costs and distances. In particular, we are interested in the surge of Chinese exports to Europe.
Although the gains from trade are widely accepted, less is known about the magnitude of the penalty faced by countries for which trade is costly. Reducing trade costs has direct and indirect benefits; it promotes trade and also leads to industrial restructuring in the economy, changes in specialisation, factor prices and real income. We focus on international MTCs and on trade facilitation as key aspects of trade costs and analyse how these effects operate and how significant they are.
The relationship between international trade and trade costs has traditionally been estimated using gravity models of trade, which relate bilateral trade flows to the income and population of the trading partners and the geographical distance between them. Recent research has focused on the use of more accurate proxies for transport costs, such as freight rates, infrastructure or customs procedures. In this line, Limao and Venables (2001) analyse the dependency of trade and transport costs on geographical and infrastructure variables and estimate the elasticity of trade with respect to transport costs to take values from 2 to 5, meaning that a reduction of 1 per cent in transport costs increases trade by 2-5 per cent. In addition, Martínez-Zarzoso and Suárez-Burguet (2005) and Martínez-Zarzoso et al (2008) obtained similar results using disaggregated data. Recent studies have found that distance is imperfectly correlated with MTCs (Wilmsmeier and Hoffmann, 2008). Clark (2007) and Martinez-Zarzoso and Nowak-Lehmann (2007) find that distance is a poor proxy for transport costs, but may be a proxy for other types of trade costs, such as familiarity or differences in tastes, and has the advantage of being truly exogenous of the volume of trade in goods. In light of these findings, a number of studies have underlined the importance of obtaining better data on transport costs (Anderson and van Wincoop, 2004).
Yet the evidence suggesting that transport costs are only vaguely related to distance should not be confused with the proven empirical fact that distance is correlated with trade flows. Hilberry and Hummels (2008) note that roughly a quarter of world trade takes place between countries sharing a common border and half of world trade occurs between partners less than 3000 kilometres apart. It is not clear, however, whether the effect of distance on trade volumes can be ascribed either to transport costs or to other trade costs or trade facilitation The effect of maritime transport costs aspects, such as historical ties, cultural proximity, business networks or a combination of and the interrelation between those factors.
The theoretical models used to generate the gravity equation usually assume homogeneous firms within a country and consumer preference for variety. These two assumptions imply that all products are traded to all destinations. However, empirical evidence indicates that only a small number of firms are exporters and export exclusively to a limited number of countries. This empirical fact has led to the development of the so-called new-new trade theories based on firm heterogeneity in productivity and fixed exporting costs (Melitz, 2003). These new theories contemplate the existence of a productivity threshold for each country that firms have to exceed in order to become exporters. As a result, two margins of trade emerge: the number of unique shipments (extensive margin) and the average value of shipments (intensive margin) (Hummels and Klenow, 2005).
In marked contrast to other studies that focus on the determinants of maritime trade, we use sectoral trade data for eight different selected industries and decompose trade into two margins: the number of varieties exchanged inside each category defined at the Harmonised Standard Classification (HS6) level (extensive margin) and the average value of each variety (intensive margin). This disaggregation shows to what extent trade costs matter in international trade and isolates which trade components are most affected by variations in different types of trade costs.
Our analysis focuses on disaggregated trade between the European Union (EU15) 1 and 15 trading partners representing 2 a total of 225 maritime trade routes over a period of 8 years (1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007), with a special focus on Asia. Freight rates are obtained from the OECD MTCs Database. The database gathers data on unit and ad valorem transport costs for the exports and imports of several sectors between pairs of countries, excluding loading costs. One advantage of this source is that the data are disaggregated at product level (HS2) and precisely define origin-destination and mode of transport for shipments. Therefore, we are able to decompose bilateral trade values into margins and explore how well the variability of each margin is explained by freight rates. In addition, we use a number of trade-and cost-related variables, namely time to export and import and inland transport costs, as proxies for other trade costs related to what the literature has labelled 'trade facilitation'.
By using precise and time-varying transport cost data, we find that transport costs for maritime trade between Asia and Europe have an impact on trade mainly through the intensive margin of trade, at least for the products studied.
The remainder of this article is organised as follows. The next section presents the methodology used to decompose the value of trade into margins and the main hypotheses to be tested. The section after that describes the data. The penultimate section shows the main results. The final section concludes.

D e c o m p o s i n g M a r i t i m e T r a d e : M a i n H y p o t h e s i s
In recent literature the effect of transport costs on trade has been commonly analysed using the gravity model of trade, with the dependent variable being the aggregate/disaggregated value of trade between two countries. Some recent studies for aggregate trade include Limao and Venables (2001), Sánchez et al (2003) and Martinez-Zarzoso and Suarez-Burguet (2005), while those focusing on disaggregated trade include Martínez-Zarzoso et al (2003), Martinez-Zarzoso et al (2005), Martinez-Zarzoso (2009) andMartinez-Zarzoso and. This approach relies on a model that assumes iceberg trade costs 3 and symmetric firms. In this setting, aggregate trade values react to trade costs in exactly the same way as disaggregated trade (firm-level) quantities and consumers buy positive quantities of all varieties.
In this context, we can express the quantity of a variety from origin country i to destination country j (q ij ) as where E j denotes country j's total expenditure on the differentiated product, (p i t ij ) is the price of product i at destination j, p i varies across destinations due to positive iceberg transport costs, t ij . e P j ¼ P i p i t ij À Á 1 -σ is a price index and σ is the elasticity of substitution, which is constant across varieties 4 (constant elasticity of substitution (CES)). 5 As the quantity traded of each variety is in most cases not observable, adding two assumptions, namely, all varieties in the origin are symmetric and the destinations will consume all the varieties in equal quantity, total trade values are obtained as the product of three variables: the quantity per variety traded (q ij ), the price of the variety (p i ) and the number of varieties(n i ). The outcome is In equation (2), the quantity per variety is the only component of T ij that has bilateral variation. Following Hillberry and Hummels (2008), we are able to examine each of the components of total trade values in a more flexible way because our data contain not only quantities, but also prices and the range of products varies depending on the origin and destinations. With this aim, some of the assumptions made above are relaxed. Prices may vary across destinations, if the elasticity of substitution is not constant or if transport costs are not iceberg The effect of maritime transport costs costs (Hummels and Skiba, 2004). Consequently for a given year t, we can assume: At least three reasons have been suggested in the literature to explain why the range of trade products might vary with trade costs (Feenstra and Kee, 2005). First, goods produced in different locations (origin and destination) can be homogeneous. In this case, if production costs at origin and destination are very similar or trade costs are sufficiently large, these goods will not be traded. In addition, the higher the transport costs, the more likely that the products are going to be non-traded goods. Second, if goods are differentiated by country of origin, each country producing a different variety has to incur a fixed cost to sell the product in each destination country. Therefore, not all the varieties will be shipped to each destination and the number of varieties traded will depend negatively on the magnitude of trade costs. Finally, not all varieties are consumer goods. Intermediate inputs that are used in the production of final goods would only be exported to destination j if country j produces the final good. Owing to 'just in time' production processes intermediate goods are more likely to be traded over short distances. With the methodology described below we aim to shed some light on the validity of each of these explanations that justify why not all the varieties are shipped to each destination and why both trade margins depend negatively on the magnitude of trade costs.
The methodology we use to decompose the aggregate value of trade into its various components is based on Hummels and Skiba (2004). Unique shipments are indexed by s and the total value of shipments from country i to country j is given by where N ij is the number of unique shipments (extensive margin of trade) and PQ ij is the average value per shipment (the intensive margin). Hence, total trade value is decomposed firstly into extensive and intensive margin As there can be multiple unique shipments within an origin-destination country pair, the number of shipments can be further decomposed into the number of distinct standard international trade classification (SITC) products shipped, N ijk , and the number of average shipments between a country of origin and a destination country, N ij F . N ij F >1 means that we observe more than one unique shipment per commodity travelling from country i to country j.
The average value per shipment can also be further decomposed into average price and average quantity per shipment: By substituting equations (6) and (7) into equation (5) we can decompose total trade between two countries into four different components: Quantities are measured in tons for all commodities. Using a common unit allows us to aggregate different products and compare prices (proxied with import unit values) across all commodities.
We now have two decomposition levels. The first is given by equation (5), which decomposes total trade value into the number of products traded and the average value per product. The second, given by equation (8), further decomposes each of these two components into another two. The extensive margin is decomposed into the number of distinct SITC goods shipped and the number of average shipments between a country of origin and a destination country. The intensive margin is decomposed into average price and average quantity. Taking logs of the firstand second-level decompositions in order to have a linear model, and adding the time dimension for empirical purposes, t, we obtain: Next we analyse how each of the components of equation (10) co-vary with distance and with other trade-related costs (MTCs, time to export/import, cost to export/import).
The estimating equation takes the following form: where γ K and λ t are industry and year fixed effects and α i and β j are importer and exporter fixed effects. ε ijkt is an error term and ln(X ijkt ) is the log of exports of product k from country i to country j in period t or each of its components: the log of the average value per shipment (intensive margin) and the log of the range of shipments (extensive margin), as described in equation (9). GDP it and GDP jt denote the Gross Domestic Product of the importer and exporter country in year t, respectively and GDPh it and GDPh jt denote the respective Gross Domestic Product per capita. D ij is the geographical distance between the trading countries' capitals and TC ijkt denote the freight rates of transporting product k from country i to country j 6 in period t. timem jt and timex jt are, respectively, the time to import from and the time to export to a given destination j. cosm jt and cosx jt are behind-the-border costs 7 to import from and export to a given country j.
As equation 11 is linear in the parameters, the coefficient of total imports will be equal to the sum of the coefficients of the two margins. A further decomposition can be performed, using each of the components in equation (10) as a dependent variable in equation (11). We then test the following alternative specification, which controls for time-sectoral and time-and country-specific effects: where λ it , θ jt are year-country fixed effects, δ kt is time-sectoral fixed effects and ε ijkt is an error term. This specification is introduced in order to control for multilateral resistance effects. Anderson and van Wincoop (2003) describe these effects as the impact of changes in prices caused by variations in trade costs between a given country i and all its trading partners. However, not only the variation of bilateral trade costs matters when determining trade flows between two countries (i and j), but also the variation of these costs in comparison to other existing trade costs linking these two countries to their other trading partners. In order to control for these effects and obtain the direct effect of trade cost reductions, some authors (Feenstra and Kee, 2004;Baldwin and Taglioni, 2006) recommend introducing year-country fixed effects in order to capture the indirect impact of trade cost reductions. In addition, some characteristics related to the sectors we selected may impact bilateral trade flows between countries over time, namely, comparative advantage in a broad sense. Time effects that are sector-specific control for unobserved heterogeneity that is sector specific but varies over time and is common to all countries, for example, technological shocks.
D a t a D e s c r i p t i o n a n d V a r i a b l e s

Maritime transport costs
The main data source for MTCs is the OECD MTC database. The MTC covers annual transport statistics (ad valorem transport costs, unit transport costs in dollars, total transport costs in dollars) for a vast number of trade routes according to the type of good (2-digit HS) and the type of vessel (container ship, tanker, dry or dirty bulk vessel) used to ship the goods. It is widely accepted that trade openness has increased over the last three decades. The cornerstone of this statement is the well-documented fall of tariff barriers (Hummels, 2001). Is the same trend noticeable for transport costs? Figure 1 from Korinek (2009) shows the evolution of international MTCs since 1980. The general trend appears to be a slight decrease in the ad valorem equivalent of international MTCs over the period 1980-2005. Some differences appear between developing and developed countries. Ad valorem transport costs are higher for developing countries in every single year with the only exception of 2002. Transport costs decreased rapidly in the 1980s for developed countries before a period of stabilisation in the 1990s, recording a slight increase during the last decade. With regard to developing countries, ad valorem freight rates remained around 9 per cent until 1995, fell steadily between 1995 and 2001 and increased sharply after 2001. Figures 2 and 3 detail the evolution of ad valorem transport cost between the EU15 and China and between the EU15 and    the United States in the textile sector and the footwear industry (62, 63, 64) and in the machinery, electrical equipment and vehicles sectors (84, 85, 87), respectively. Large differences can be noticed at first glance between the two partners of the EU15 and between import and export ad valorem transport costs. Ad valorem transport costs are higher for Chinese imports of textiles in comparison with the costs faced by EU15 exporters (9.3 per cent and 2.05 per cent, respectively in 2007). In the case of the EU15-USA trade in textiles, the difference between export and import ad valorem transport costs is less significant (4.08 per cent and 2.6 per cent, respectively in 2007). The same can be said of sectors 84, 85 and 87. Ad valorem transport costs are remarkably higher for Chinese exports to the EU15 than for EU15 exports to China (6.9 per cent and 0.7 per cent, respectively in 2007). Otherwise, ad valorem transport costs for these sectors and for trade between EU15 and the USA are quite similar (1.3 per cent and 1.8 per cent, respectively in 2007). Marked differences between the costs to import from and export to China reveal perhaps the large trade imbalance that exists between Europe and China (Behrens and Picard, 2011). Evidence showing decreases in transport costs calculated at the equivalent ad valorem tariff are significant for vehicles (87) in the direction China-EU15 (from 17.15 per cent in 1999 to 8.97 per cent in 2007, which implies a decrease of 52 per cent). We have carried out the same exercise for the other EU15 partners in our sample. In general, developing Asian countries show the same trends as China with regard to the evolution of their ad valorem MTCs with the EU15 (with a decrease of 40 per cent in freight rates for vehicles). A sharp decrease in ad valorem transport costs in vehicles, which also occurred in the case of developed Asian countries (32 per cent), is the most noticeable trend.

Gravity variables
Trade data were obtained from Eurostat. We use a detailed Eurostat database, which covers both extra-and intra-EU trade. The products are classified according to the HS codes at the HS6-digit level. Products within eight broad categories (at 2-digit level) of manufactured products are taken into consideration (categories 30, 62, 63, 64, 84, 85, 87 and 90 as described in Table A1 in the Appendix). The extensive and intensive margins of trade, as well as the average prices of products traded between the EU15 and 15 partners have been calculated over the period 1999-2007 using export values and export quantities. We count the number of products (6-digits HS) exported within each 2-digit HS category from each exporter to each importer on a yearly basis. On average, our sample contains 77 varieties of goods exchanged within each category.
Income and population data are taken from the World Development Indicators Database 2008 and distances from capital cities are taken from Centre d' Etudes Prospectives et d' Informations Internationales (CEPII) (www.cepii.fr/ anglaisgraph/bdd/distances.htm). Trade facilitation variables, namely, time needed to export/import and inland transport costs paid to export/import come from the World Bank Doing Business Data set.
A description of the main variables, sources and units in which the variables are measured is presented in Table 1 and summary statistics of the main variables are presented in Table 2.

M T C s a n d t h e T w o M a r g i n s o f T r a d e
The gravity model of trade presented above is estimated for bilateral trade and also for both trade margins for exports and imports of EU15 to 15  (11). We decompose our results according the position of the EU15 as an exporter (Table 3) or importer (Table 4).  Notes: Importer, year and sectoral fixed effects control for unobserved sources of variability linked to countries, sectors and time variant characteristics. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics robust to heteroskedasticity and autocorrelation are in brackets. l denotes natural logarithms. The variables time to import (timem) and cost to import (cosm) are only available after 2003. As a result, the estimation is restricted to the period 2004-2007.
Equation (11) is estimated using a least squares dummy variable estimator that introduces different sets of dummy variables to control for unobservable heterogeneity as described in the section 'Decomposing maritime trade: Main hypothesis'. The first column of Table 3 shows the results when the dependent variable is sectoral trade value, while Columns 2 and 3 display the dependent variables, namely, the extensive and the intensive margins, respectively. Finally, the last two columns show the results for a further decomposition of the intensive margin into average quantity (Column 4) and average price (Column 5). The same structure is used for Tables 4-11.
The estimates shown in Table 3 concerning the target variables (freight costs, time to import and inland transport costs to import) show a significant and negative impact on EU15 exports, with distance also displaying a negative and significant coefficient with an elasticity that is higher than unity. These results strongly support the finding obtained in several studies (Wilmsmeier and Martínez-Zarzoso, 2010) that distance has an impact on trade after controlling for transport costs using more direct proxies. Hence, distance may reveal other characteristics of bilateral relations between countries that influence trade, such as trust or information. With regard to the two margins of trade, the decomposition of the influence of the trade cost variables on each margin of trade shows that while distance effects work Notes: Exporter, year and sectoral fixed effects control for unobserved source of variability linked to countries, sector and year characteristics. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l is for natural logarithms. Data availability of the variables time to export (timex) and cost to export (cosx) restrict the period to 2004-2007. exclusively through the intensive margin (Columns 2 and 3, Table 3), ad valorem freight rates, time to import and inland transport costs have an impact on both margins of trade, indicating that they affect the fixed and variable costs of exporting. GDP, as a proxy for the size of the economy of the partners of the EU15, has a positive impact on trade and its margins. When the EU15 is the exporter, GDP and GDP per capita display the expected positive sign. However, this is not the case for EU15 imports (Table 4). GDP per capita is negative, indicating perhaps that the type of products imported are labour intensive. This could be owing to the composition of our sample being largely dominated by Asian countries because of data availability. Exports to the EU15 from these countries are Notes: Importer and year and sector and year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per capita, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007. Notes: Exporter and year and sector and year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per head, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007.
dominated by low-value added products mainly produced by countries with lower levels of GDP per capita. The estimates concerning trade cost variables for EU15 imports (Table 4) are remarkably different to those obtained for EU15 exports (Table 3). The main differences concern distance, which shows a positive and significant coefficient in Table 4, indicating that the EU imports more from more distant destinations. This positive distance effect cannot be due to factors that are time invariant, as they are controlled for by adding country dummies. However, this could be showing the increasingly important role played by China as one of the main EUtrading partners. Important differences are also found for ad valorem freight rates, which record a higher impact on EU imports (almost double) than on EU exports. Indeed a 10 per cent decrease in ad valorem transport costs is associated to an increase in EU imports of 6.3 per cent (3.3 per cent for exports) and the effect works only through the intensive margin. The variable time needed to import is now not statistically significant for any of the dependent variables (different trade margins) and inland transport costs display very high elasticity with respect to EU imports, which is almost 10 times the elasticity found for EU exports. It is worth noting that EU imports in the sectors considered are dominated by countries in Asia in which reductions in internal transport costs could considerably impact their exports to the EU and other destinations.
Next, we focus exclusively on the effect of reductions in freight rates on EU trade. Tables 5 and 6 present the results for the specification given by equation (12). It has the advantage of extending the analysis to more years and of isolating the impact of MTCs more effectively after controlling for unobservable sources of variability through a set of time and country and sector and time fixed effects. As before, the dependent variable in the first column is the total imported or exported value from a given country. In the rest of the columns, each of the components of equation (10) is used as a dependent variable. The coefficients have the expected signs in most specifications and ad valorem transport costs display a negative coefficient for all components and for EU15 exports (Table 5) and imports (Table 6).
As previously, ad valorem transport costs have a greater effect on the intensive margin of trade (Column 3 -Tables 5 and 6) than on the extensive margin (Column 2 -Tables 5 and 6) for all sampled products. Approximately 83 per cent of the impact of ad valorem transport costs on trade works through the intensive margin (that is, 0.222/(0.222+0.045)) in the case of EU15 exports and about 99 per cent (that is, 1.166/(1.166+0.004)) in the case of EU15 imports. In comparison to the results shown in Tables 3 and 4, we can conclude that the estimated elasticities are robust to changes in model specification and that controlling for unobserved heterogeneity that is country-and-time specific does not modify the main results. Indeed the elasticity of EU imports with respect to ad valorem transport costs is slightly higher than before, indicating that a 10 per cent reduction in transport costs will increase imports by approximately 12 per cent, a more than proportional increase. Our results for EU exports are strikingly similar to those obtained for intra-Latin American trade by Wilmsmeier and Martínez-Zarzoso (2010), who also obtained an elasticity of (−0.5) for maritime transports costs with respect to the extensive margin and an elasticity of (−0.19) with respect to the intensive margin.
When decomposing the effect of the intensive margin into the impact on the average quantity of each shipment and their average price, the foremost is a decrease in the average price for EU15 imports (78 per cent, that is, 0.838/(0.838 +0.327)) and also for EU15 exports (62 per cent, that is, 0.567/(0.567+0.345)). Notes: Importer and year and sector and year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per head, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007. Notes: Exporter and year and sector and year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per head, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007.
The impact of transport costs on the average quantity shipped is negative and significant (−0.327) only for EU15 imports (Table 6), but positive and significant for EU15 exports (Table 5), indicating that an increase in transport costs leads to an increase in the average quantity of goods shipped, accompanied by a decrease in prices. Summarising, our results indicate that reductions in freight rates increase not only the average quantities of EU exports (intensive margin), but also the number of products exported, whereas for EU imports only the average quantities imported are affected by reductions in freight rates. One possible explanation for this is that the products imported by the EU from the 15 trading partners included in our data set are less differentiated in comparison to the Notes: Importer and year and sector and year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per head, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007. Notes: Exporter and year and sector and year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per head, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007.
products exported from the EU, which would mean the number of varieties that can be produced is limited. We also investigate whether our results are common for all the sectors under study. In order to do so, we have grouped the sectors into four categories of products. The first group includes the textile and footwear sectors (sectors 62, 63, 64), the second includes machinery, construction, vehicles and electronics (sectors 84, 85, 87) and the two last groups correspond to the two remaining sectors, namely, 30 (pharmaceutical products) and 94 (furniture). 8 The corresponding estimation results for the first group are shown in Tables 7 and 8. The estimated coefficients for transport costs are not statistically significant and have the expected signs for EU15 imports. These results are not surprising, as ad Notes: Importer and year and sector and-year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per head, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007. Notes: Exporter and year and sector and year fixed effects are added to control for unobserved sources of variability linked to multilateral resistance effects and sector year characteristics. Hence, country characteristics that changed yearly (in our case GDP, GDP per head, distance, time to export/import, cost to export/import) are omitted. *, **, *** indicate significance at 10 per cent, 5 per cent and 1 per cent, respectively. t-statistics are in brackets. l denotes natural logarithms. The period is from 1999 to 2007.
valorem transport costs have barely decreased for textile goods over the period under study (see Figure 2 above). This may also be because of other events, such as the end of the Multifibre agreement in 2005, which influenced exports of textiles to the EU15. It is worth noting that the effect of this agreement, which is sector specific, was controlled for in our results shown in Tables 5 and 6 through the addition of sector-specific time dummies .  Tables 9 and 10 show the results for the second group of products: machinery, construction, vehicles and electronics. The estimated coefficients for transport costs are only significant and have the expected signs for EU15 imports (Table 10), whereas no impact is shown on the value of EU exports. Indeed, a sharp decrease in ad valorem transport costs is observed for vehicles in the direction Asia-Europe.
Finally, so as to investigate to what extent our results are driven by EU trade with Asia, Tables 11 and 12 show the results obtained from estimating equation (12) only for trade flows between EU and Asian countries. The results confirm that the decrease in ad valorem transport costs has a significant and positive effect on trade between the EU15 and the Asian countries in our sample. Once again the effects through the intensive margin and the average price dominate. In particular, for Asian exports to the EU, the intensive margin of trade is much more sensitive to variations in transport costs than for Asian imports from the EU, indicating that Asia will benefit more than the EU from reductions in transport costs.
We have applied several strategies in order to check the robustness of our results. In order to control for the possible endogeneity of the trade cost variables, we have used lagged values and tried alternative sets of fixed effects. In all these cases, we find almost no variations in the results in comparison to those presented in this article. 9

C o n c l u s i o n s
This article focuses on the analysis of the relationship between European maritime trade and trade costs. According to new theories of international trade with imperfect competition and heterogeneous firms, lower trade costs increase bilateral trade through an increase in both margins of trade. We use highly disaggregated trade data to decompose trade into its extensive and intensive margins and to estimate the effects of different sources of trade costs, namely, distance, time needed to trade, inland transport costs and MTCs on each margin.
The decomposition of the influence of the trade cost variables on each margin of trade shows that while distance effects work mainly through the intensive margin for EU exports, changes in ad valorem freight rates, time to import and inland transport costs have an impact on both margins of trade, indicating that they affect both the fixed cost and the variable cost of exporting. In particular, inland transport costs record very high elasticity with respect to EU imports, which is almost 10 times the elasticity found for EU exports. It is worth noting that EU imports in the sectors considered are dominated by countries in Asia where reductions in internal transport costs could considerably impact their exports to the EU and other destinations. This indicates the importance of investing in trade facilitation initiatives in developing countries.
A decrease in freight rates has a substantial and positive impact on trade, particularly on the intensive margin of trade and partly through a decrease in the average price of traded goods and an increase in the average quantity traded. This result indicates that Europe exports more of the same goods at a more competitive price for consumers. To a lesser extent, decreases in MTCs also increase trade in new varieties of goods, in particular for EU exports to Asia. This finding helps to understand how the dynamics of transport costs impact trade. These results deviate significantly from the results obtained when transport costs are approximated using the geographical distance between countries.
Our findings suggest that political actions aimed at spurring competition in the maritime transport industry and supporting innovations in the shipping industry do have an impact on the volume and the composition of international trade. In particular, increasing ship size and limiting the consumption of fuels by ships could reduce freight rates and improve the competitiveness of Asian firms in the EU market and stimulate the creation of new products. These results call for further research on the effects of transport market structures on trade patterns and transport costs.
A c k n o w l e d g e m e n t s The authors would like to thank an anonymous referee and the editor for their very helpful comments and suggestions and gratefully acknowledge the financial support received from the research project (ECO2010-15863) granted by the Spanish Ministry of Science and Innovation.

N o t e s
1 We only considered EU15 because of data availability concerning transportation costs. 2 EU15 stands for: Austria, Belgium, Denmark, France, Finland, Germany, Greece, Ireland, Italy, Luxemburg, the Netherlands, Portugal, Spain, Sweden and United Kingdom. The 15 partners selected are the following: China, Hong-Kong, India, Indonesia, Japan, Malaysia, Philippines, The effect of maritime transport costs