INTRODUCTION

There is a growing interest in the study of the role played by transport costs in models of international trade. Krugman's (1991) seminal work on geography and trade emphasises the crucial importance of the size of trade costs in economic geography models. Deardorff (2001) models local comparative advantage in a partial equilibrium framework and Venables and Limao (1999) show how relative location affects specialisation patterns in a general equilibrium framework.

A better understanding of trade costs might provide insights into a broad range of topics related to international specialisation (Hummels et al, 2001), factor content in trade (Donald and Weinstein, 1996), substitution between trade and foreign investment (Markusen and Venables, 1996) and regional integration. Henderson et al (2001) also emphasise the important role played by transport costs and their influence on trade and income. Kumar and Hoffmann (2002) analyse the mutual relationship between trade and its maritime transport cost and its relevance for globalisation. Hummels (1999) presents evidence of the importance of transport costs in the determination of trade patterns. Rauch (1999) identifies some of the non-conventional cost of trade associated with search for the case of differentiated goods. He claims that these informational costs fall relatively on differentiated goods.

In this paper we aim to shed some light on the investigation of the determinants of transport costs and to study the nature of the relationship between trade and transport costs in the ceramic sector. In most cases, we have no direct way of observing these trade barriers, and therefore we have to rely on indirect measurement and trade modelling in order to assess their relevance. Hummels (1999) has made a significant contribution to the literature in his paper entitled ‘Towards a geography of trade costs’, where clear evidence of the importance of trade costs is shown. His results suggest that, to some extent, import choices are made in order to minimise transportation costs. We apply a variation of Hummel's model to Spanish trade flows using data on Spanish ceramic exports to 76 destinations and their associated overland and maritime transport costs. Our contribution is a unique (though highly specific) data-set containing point of shipment freight rates as opposed to the more common measures, taken from national trade data sources, based on ‘free on board’/‘cost, insurance and freight’ ratios. Since this is the precise type of data collected by Limao and Venables (2001), albeit for 40 foot containers shipped from Baltimore, results presented in the current paper could be useful in validating or refuting earlier findings.

The Section on Measurement of transportation costs presents the different methods used in the recent literature to measure transportation costs. In the section on Estimation of a transport cost function, this is estimated by using data on ceramic tiles. The first section on transport costs and trade presents and estimates a variant of the standard gravity model of trade. The section on concluding comments presents the results of the empirical application.

The sensitivity of trade flows to transportation costs is discussed in Appendix I, taking into account not only distance and infrastructure (Limao and Venables, 2001) but also the existence of back-hauling, special transport conditions (dangerous goods, refrigeration) and number of reloads.

MEASUREMENT OF TRANSPORTATION COSTS

In the recent economic literature, there have been several attempts to measure directly or indirectly transport costs. A number of authors used cif/fob ratios as a proxy for shipping costs (Limao and Venables, 2001; Radelet and Sachs, 1998). Since most importing countries report trade flows inclusive of freight and insurance (cif) and exporting countries report trade flows exclusive of freight and insurance (fob), transport costs can be calculated as the difference of both flows for the same aggregate trade. However, Hummels (2001) showed that importer cif/fob ratios constructed from IMF sources are poor proxies for cross-sectional variation in transport costs and such a variable provides no information about the time series variation. This measure suffers from the fact that it is an aggregate over all products imported. Oguledo and Mcphee (1994) also question the usefulness of cif/fob ratios as a proxy for transportation costs, when averages are used or in the presence of incomplete data.

Hummels (1999),(2001) use data on transport costs from several primary sources including shipping price indices obtained from shipping trade journals, air freight prices gathered from survey data and freight rates (freight expenditures on imports) collected by customs agencies in United States, New Zealand and five Latin American Countries (Mercosur plus Chile). Shipping prices refers to weighted index numbers for ocean shipping with varying coverage of time periods, goods shipped and routes. For example, the time charter index from the Norwegian Shipping News, reports the cost of employing vessels of varying size and speed in many ports world-wide (Appendix II, Hummels, 2001). On the other hand, freight rates refer to freight and insurance charges for each shipment reported by exporter and commodity (Appendix A, Hummels, 1999).

Limao and Venables (2001) use, besides cif/fob ratios, shipping company quotes for the cost of transporting a standard container (40 ft) from Baltimore to 64 destinations. The data were provided by a single firm, the packing is loose and the cost does not include insurance. The authors point out that it is not clear how the experience of Baltimore generalises, since the freight rates are affected by specific characteristics (particular routes and opportunities for back-hauling and for exploiting monopoly power). These published data may not be representative of the majority of transactions where the price is confidential. Micco and Pérez (2001) and Sánchez et al (2002) use data from the US Import Waterborne Databank (Department of Transportation). Transport cost is defined there as ‘the aggregate cost of all freight, insurance and other charges (excluding US import duties) incurred in bringing the merchandise from the port of exportation to the first port of entry in the US’.

In the empirical application of this paper we use an additional source: data on transportation costs obtained from interviews held with logistics operators in Spain. A total of 15 operators were interviewed, of which five were overland transport operators and 10 maritime transport operators (Appendix II). The data comprise exports of ceramic tiles from Spain to 76 destination cities, 14 of which are in landlocked countries (Appendix II) (Table 1 shows a summary of the data). Logistics operators were asked for information on quantity (m2) and value (US$) exported, transport costs (20 ft container (TEU), US$ excluding insurance costs) and transport mode (road, ship). An important part of our data corresponds to maritime transport since only in 16 cases out of 76 was the mode road, naturally European countries. When the transport mode was ship, the port of origin was Valencia in most cases, given the proximity of this port to Castellón, the main location of ceramic factories in Spain. In 12 cases the origin was Barcelona, in three cases Bilbao and in two cases Algeciras.

Table 1 Descriptive statistics of the data on transport costs, trade and distance
Full size table

Ceramic tile manufacturing is one of Spain's most dynamic industries. The second-largest ceramic tile producing country, Spain maintains a world-wide market share of more than 14%. More than 80% of the country's manufacturers are clustered together in the province of Castellón. Those companies make up an industrial conglomeration that benefits both from secondary services and a competitive ancillary industry that helps ensure continuing innovation and development.

Table 2 shows a list of countries receiving more than 1% of total exports and their corresponding freight rates. In all, 10 countries are European Union members, but more distant destinations are also found in the list such us Singapore, Australia and Colombia. We observe that in 1999, destinations with low transport cost rates (below average=$1785 per TEU), as USA, France, UK, Portugal or Germany, enjoy large shares with the exception of Poland which is considered a landlocked country (being landlocked normally adds extra costs). Nevertheless, the data also show destinations with very low transport cost rates, as Singapore, Greece and Italy who enjoy medium shares (around 1–3% of total exports).

Table 2 Main markets and transport costs in 1999
Full size table

Figure 1 shows the dispersion of freight rates versus trade shares for all 76 countries in the sample. Contrary to Hummels (1999), our data do not clearly show that transport costs play an important role in allocating trade over partner countries. In our data set we have two cases where freight rates are extraordinarily high: Peru and the Côte d’Ivoire (the cost of transporting a 20 foot container from Valencia to Callao and to Abidjan is US$ 6203 and US$ 5299, respectively). We will estimate the empirical model without these countries to avoid problems associated with distorted measurement of transport costs.

Figure 1
figure 1

Transport costs versus share in total exportsNote: SHARE=Share in total exports is in percentage, TC=Transport costs is in US$ per TEUSource: Own elaboration with data obtained from interviews held with logistics operators

Full size image

ESTIMATION OF A TRANSPORT COST FUNCTION

In the recent literature there have been several attempts to investigate not only the trade-transport costs relationship, but also the determinants of international transport costs. Limao and Venables (2001) and Radelet and Sachs (1998) undertake regressions to explain transport costs. The explanatory variables used in their analysis are basically related to distance and connectivity, such as if countries are landlocked, or if trading partners are neighbours, and to country characteristics such as GDP per capita. Micco and Pérez (2001) and Sánchez et al (2002) analyse the impact of port reform on transport costs, and study possible determinants of port reforms in Latin America. Hummels (1999) studies the determinants of transport costs by dividing the trade costs implied by trade flows into three different components: explicit measured costs given by tariffs and freight rates; costs associated with common proxy variables such as distance, sharing a language, sharing a border or being an island, and implied but unmeasured trade costs given by geographical position, cultural ties or political stability. His estimation results indicate that explicit measured costs are the most important component. He offers three alternative explanations for the costs associated with language, distance and adjacency: trade barriers, preferences and production composition. The author develops a potential method for separating the three alternatives paying attention to functional form and level of aggregation.

When adjacency and distance effects are interpreted as direct trade barriers, the estimated coefficients are not significant. The language effects are significant for almost the same set of goods and are of similar size as when interpreted as preference parameters. The distance coefficient might also be identifying an endogenous production response for goods characterised by high transport costs. Production of specific varieties is placed close to locations where these varieties are strongly preferred.

Since the results present a mixed picture and the fit of the equations is poor, the author concludes that the regression results do not allow disentangling among the various interpretations.

Fink et al (2000) investigate how liberalisation in trade and transport services leads to further reductions in transport costs, which in turn lead to a further promotion of trade in goods. Kumar and Hoffmann (2002) consider the mutual relationship between trade volumes, transport costs, and the quality of transport services. They find that the market for maritime transport services is growing and observe increased concentration in the maritime industry and, at the same time, more competition. Although transport unit costs decline, the incidence of the maritime transport costs in the final value of the good increases since many components are purchased internationally. Some preliminary research for Intra-Latin American trade suggests that higher quality of service implies higher transport costs, yet also promotes trade. Economies of scale from high trade volumes have a strong negative (ie, decreasing) impact on transport costs. The authors state that the strong relationship between trade and transport costs detected by Limao and Venables (2001) does not only reflect the elasticity of trade towards transport costs, but might be also reflecting the economies of scale through which higher volumes lead to lower costs of transport. For the case of Intra-Latin American trade, Hoffmann (2002) analyses the impact of a number of factors on transport costs. The results suggest that the unit value of traded goods and economies of scale appear to have a stronger impact on transport costs than distance.

We are aware that recent changes in restrictive business practices in liner shipping have been also affecting the evolution of transport costs. This is discussed later in the text.

We estimate a linear equation where transportation costs are specified as a function of distance, unit values, infrastructure and various dummies. Initially, we also considered a variable capturing economies of scale, that is the level of trade that goes through a particular route. This variable could be calculated in terms of volume or in terms of value. We have data on total volume exported from country i to country j but not for trade that goes through a particular route. The direct incorporation of the available data in the estimation presents a problem of endogeneity. We estimate the transport cost function using instrumental variables estimation (importer GDP and importer GDP per head were used as instruments) and only dropping the infrastructure variable the coefficient was significant at 13% level and the magnitude was (−0.11).

An obvious determinant of transport costs, as confirmed by the literature, is the distance between trading partners. Traditionally, distance has been used in gravity equations as a proxy for transport costs since a higher distance implies a longer journey and an increase in the associated costs. Figure 2 shows the relationship between transport cost and distance found in our data.

Figure 2
figure 2

Transport costs versus distanceNote: TC denotes Transport costs in US$ per TEU and DIST denotes distance in kilometresSource: Own elaboration with data obtained from interviews held with logistics operators

Full size image

Recent studies (Hummels, 1999; Kumar and Hoffmann, 2002) show that freight rates also depend on the unit value of the goods being shipped. Although the reasons are not obvious, insurance costs, modal switching (transfers between modes) or discrimination by shipping cartels may bear some responsibility for this. Since our shipment units are homogeneous, the mode is known and insurance costs are omitted from the freight rates, we attempt to find out effectively what the most plausible reason is to determine why there is a relationship between freight rates and unit values. Infrastructures in the exporting country and in the transit countries have also proved to be important determinants of transport costs (Limao and Venables, 2001). The infrastructure measures are related to the quality of communications and transport infrastructures that countries possess.

The costs of the journey between countries are influenced by other geographic characteristics such as adjacency, being an islandFootnote 1 Footnote 2 or being landlocked. Countries sharing a common border usually have better communication network connections and more possibilities for back-hauling, since they trade more extensively, allowing the fixed costs to be shared over two trips and reducing total costs. Some cultural similarities, such as a common language, could also be considered as determinants of transport costs, assuming that this will facilitate trade transactions. Furthermore, being landlocked normally adds extra costs since commodities transported by ship have to switch transport mode. We added several dummies according to the mode of transport and the exiting port. The basic specification is given by:

where TC j denotes transport costs, D j denotes distance, Xuv j is the export unit value, Inf j denotes infrastructure of country j, Inftra j is infrastructure of the transit countries to reach country j and j denotes the destination country. Mode is a dummy which takes the value one when goods are transported by road and zero when they are transported by ship. Barce takes the value one when exports are from Barcelona and zero otherwise. Similar dummies are introduced for Bilbao (Bilb), and Algeciras (Alg). The estimated coefficients of the dummies will be proxies for the efficiency of these ports in comparison with the port of Valencia (which is the default).

All the variables except dummies are in natural logs. u j denotes the error term that is assumed to be independently normally distributed. The variables Inf j and Inftra j are constructed as an indexFootnote 3 (taking information on roads, paved roads, railroads and telephones) differentiating between importer and transit countries’ infrastructure as explanatory variables of transport costs. Our index is comparable to that of Limao and Venables (2001). The difference is that a rise in our index indicates better infrastructures whereas a rise in Limao and Venables’ index indicates poorer infrastructures.

Estimation results are shown in Table 3. We tried several specifications, by testing for the significance of the explanatory variables. The first two models show results which exclude the infrastructure variables. A number of conclusions were reached. First, the distance coefficient has the expected positive sign showing that a 1% increase in distance increases transport costs in approximately 0.25% (Model 2), similar in magnitude to those found in other studies for different commodities. Hummels (1999) finds commodity specific distance coefficients clustered in the 0.2–0.3 range and Kumar and Hoffmann (2002) found a distance elasticity of 0.24 for the case of Intra-Latin American trade. Secondly, the unit value coefficient is not significant when added in Model 1. The reason may be that this variable interacts with the mode dummy. This will be investigated below.

Table 3 Determinants of transport costs (variables in natural logs)
Full size table

Thirdly, the estimated coefficient of the dummy ‘mode’ has a positive sign showing that transporting the product by road increases transport costs. Finally, the significant and positive sign of the coefficient of the port of export dummies indicates that transport costs rise when the commodity is not loaded in the port of Valencia. This result cannot be a consequence of the proximity from the ceramic factories to this port since this factor is already taken into account by the distance variable. The port of Valencia has recently developed a logistics management centre which considerably improves the efficiency of its services. This new development may be an explanation of our result.

Infrastructure variables are added in Models 3 and 4. The importer infrastructure variable shows a statistically significant coefficient with the expected negative sign. An improvement of 1% in the infrastructure of the destination country lowers transport costs by 0.14%. We can interpret the importer infrastructure variable as a proxy for port infrastructure when the mode is ship. The estimated coefficient indicates that poor partner infrastructure notably increases transport costs. As Limao and Venables (2001) also find, the inclusion of infrastructure measures improves the fit of the regression since the adjusted R2 changes from 0.20 to 0.31 (Models 2, 3), corroborating the importance of infrastructure in determining transport costs. However, we find that the transit infrastructure variable is not significant at conventional levels and is positive signed. This variable is highly correlated with the mode dummy variable (r=0.83) and we expect a positive sign in the estimated coefficient because overland transport is more expensive (‘Mode’ takes the value 1 for overland transport and zero otherwise). Since multicollinearity poses problems to identify the separate effects of the two variables, only the mode variable is added in Models 5 and 6 in Table 3. Moreover, the non-significance of the transit infrastructure variable might also be due to the small number of observations. Limao and Venables (2001) find that this variable was significant and with the correct sign. However, they find that splitting the distance variable into the overland and sea components made the coefficient for transit infrastructure smaller and insignificant. The reason for this is the variable's high positive correlation with land distance. Once again, multicollinearity poses problems to identify the separate effects of the two variables. The authors confirm the importance of the transit variable testing for the joint significance of this variable with either own infrastructure or land distance.

Model 5 adds several dummy variables as a proxy for certain geographical and cultural characteristics, namely, being neighbours, being an island, being landlocked or sharing a language. Dummy variable coefficients are not significant for the adjacency, language, island and landlocked dummies. It may be that these dummies enter directly in the trade equation. A possible interpretation is that they do not represent direct trade costs. Alternatively, they may indicate cultural similarities, preferences or production effects. We also considered the possible influence of other variables such as port efficiency, as do Micco and Pérez (2001) and Sánchez et al (2002). Although the estimated coefficient have the expected sign (elasticity=−0.22) it is not significant. The lack of data for many destinations might be one reason, since our sample was reduced to 44 observations.

Finally, since not only the levels of freight rates might be affected by the mode of transport, but also the distance and unit value elasticities,Footnote 4 we introduce interaction variables (mode*distance and mode*unit value) in Model 6. Only the (mode*unit value) variable shows a significant coefficient which indicates that the unit value is only a relevant variable when tiles are transported by road. The greater the unit value ratio, the lower the overland transport cost is. Possibly road is cheaper for valuable ceramics since breakages might be more likely when loading and unloading at ports. However, the (mode*distance) coefficient is not significant. The coefficient of the adjacency dummy becomes significant when the interaction variable mode*unit value is added, indicating that sharing a border decreases transport costs by 30.23% [exp(−0.36)−1]. The landlocked dummy coefficient shows an unexpected negative sign (significant only at 10% level), whereas the language and island dummy coefficients show the expected signs but are again insignificant.

TRANSPORT COSTS AND TRADE

We now look at the relationship between trade and transport costs. In order to assess the relative importance of transport costs in trade, we need an appropriate theoretical framework. We base our application on Hummels’ (1999) model, derived from the commonly accepted Dixit-Stiglitz (1977) model of imperfect competition.

Let us assume that consumer preferences are represented by a utility function with constant elasticity of substitution over varieties within a sector,

where j denotes the importer and σ denotes the elasticity of substitution between varieties and C j denotes consumption of variety j.

Suppose that each firm produces a different variety of the good and has monopoly power over this variety (the markup of price over marginal cost equals ((σ − 1/σ)). Each country produces a number of varieties which is determined by the labour force, L i , the size of the fixed costs, a, and the elasticity of substitution,

Assuming iceberg transport costs,Footnote 5 t ij , if p i is the exporter's price exclusive of trade barriers and differences in prices across destination markets depend entirely on t ij , the price faced by the importer is p=p i *t ij .

Consumers in country j import a quantity of each variety produced in the exporter country, i, given by

where Y j is the income in the destination market and is a price index over all varieties demanded by the importer j. Supposing that demands are symmetric for all varieties from country i, the volume of bilateral trade is given by

According to this expression, an increase in transport costs leads to substitution of more expensive varieties for less expensive ones. Since there is only one sector, production cannot adjust across sectors. Moreover, production cannot adjust across varieties given that consumers' utility is the same for different varieties.

Taking natural logarithms of equation 5 and adding a number of dummies, we derive the equations to be estimated. We estimate an import demand model for ceramic exports specified as

where ln denotes natural logarithms, Y j is the income in the destination market, FTC j is the forecasted transport costs obtained from Model 6 in Table 3, Ldl is a dummy for landlocked countries, Lang is a dummy for countries sharing the same language, Isl takes the value 1 when countries are an island and Adj takes the value 1 when countries share the same border.

The model is estimated for a sample of 76 countries with 1999 data. We performed OLS estimation on the double log specification as given by equations 6 and 7. We used the forecasted transport costs as an explanatory variable of ceramics exports (equation 6) since the result from the Hausman (1978) test indicates that transport costs must be considered as endogenous in the import demand equation (IV estimation).Footnote 6 Then, we compare the results with those obtained when using distance as a proxy for transport costs (equation 7).

Table 4 shows our results. Model 1 presents the OLS results for the baseline case, which excludes the forecasted transport costs and dummies. The standard regressors are income and distance variables. The coefficient on income is positive, as expected, and the income elasticity is small (0.35). The coefficient on distance is negative signed but it is not significant. In Model 2 the distance variable is replaced by forecasted transport costs (obtained from estimation of Model 6 in Table 3, excluding the non-significant dummies: language, landlocked and island). We observe the expected sign on the coefficient which is significant at a level of 5%. Its magnitude indicates that a rise of 1% in transport costs leads to a decrease in exports of 1.04%. In Model 3 we add the list of dummies that might influence exports. All of them present the expected signs but only the language variable is statistically significant and thus improves the fit of our equation. The remaining variables have the same sign and similar magnitude as before.

Table 4 Determinants of ceramic exports
Full size table

In Model 4 forecasted transport costs is again replaced by distance to evaluate the validity of the distance variable as a proxy for transport costs when other dummies are added. We can see how the distance coefficient is not significant, as it shows the correct sign but a small magnitude (−0.28) when compared to the coefficient of forecasted transport costs (−1.26). The fit of the equation is also better when transport costs are used.

Finally, in order to compare our results with those obtained by Limao and Venables (2001), Model 5 estimates the import demand model with distance and infrastructure variables given by equation 7. Using estimates from Model 5 we will be able to link trade volumes to transport costs by computing parameter τ, the elasticity of trade volumes with respect to transport costs. We use the coefficients of significant variables included in both the transport cost and the import demand equations. We focus on distance and importer infrastructure. Table 5 presents the parameter estimates for these variables and the ratio of the trade equation to the freight elasticities indicates the elasticity of trade with respect to transport costs.

Table 5 Estimates of import elasticity with respect to transport costs
Full size table

The import elasticities with respect to transport costs implied by the point estimates are −2.4 on the basis of distance and −1.65, for the importer infrastructure. We estimated τ in the same way as Limao and Venables (2001) and we are getting broadly similar results. They show implied elasticities of −2.95 for distance and −2.34 for own infrastructure. Our calculations for the point estimates are only slightly lower than those they find (doubling transport cost leads to a fall in import value between five and six times), particularly for the elasticity based on the infrastructure measure. According to our findings doubling transport costs leads to a reduction in import value of between three and five times.

CONCLUDING COMMENTS

The objective of this paper was to investigate the determinants of transport costs and the role they play in deterring international trade. We estimated a transport costs equation using data on transportation costs for the ceramics sector obtained from interviews held with Spanish logistics operators. We also studied the relationship between transport costs and trade and we estimated an import demand model for ceramic products. Moreover, we discuss on the sensitivity of trade flows and transportation costs to the existence of back-hauling, special conditions for transportation and number of reloads.

Our results from the first estimation show that higher distance and poor partner infrastructure leads to a notable increase in transport costs. Inclusion of infrastructure measures improves the fit of the regression, corroborating the importance of infrastructure in determining transport costs. Transporting the product by road increases transport costs and transport cost rises when the commodity is not loaded in the port of Valencia. The port of Valencia has recently developed a Logistic Management Center which considerably improves the efficiency of its services and helps to explain our result.

Our results from the second estimation show that importer income, as expected, has a positive influence on bilateral trade flows. Income elasticity is very low but positive as predicted by the theory. Higher transport costs significantly deter trade, and distance does not appear to be a good proxy for transport costs in the ceramics sector.

Future estimations for sectors and products with different price-weight ratios will be of interest in order to improve the knowledge of the effects of trans-portation costs on trade flows under diverse conditions of international transport.

The study of modal transport (overland versus maritime) and its differential characteristics are of relevant interest for maritime economists and should be taken into account in economic policy-making. The proven impact of infrastructure on transport costs and trade points towards the importance of investing in new port infrastructures as a way of fostering trade and income in developing countries.