“The report of my death has been greatly exaggerated.”
Introduction
1 The integration of the world economy has increased rapidly in recent decades, with world trade growing more than twice as fast as world gdp since 1980. A plausible explanation of this globalization phenomenon that has been set forth is the unilateral trade liberalization and participation in the multilateral trading system undertaken by an increasing number of countries in recent decades. Another one is the decline in trade costs, including transport and communication costs.
2 A decline in trade costs suggests that trade should have expanded geographically. In other words, as trade costs fall, one would expect a larger share of a country’s trade to take place further away from its borders, resulting in an increase in the distance of its trade over time. The declining importance of distance over time associated with declining trade costs has been a widely accepted “stylized” fact, as illustrated by the title of the book “The Death of Distance […]” (Cairncross [1997]). On the other hand, most gravity model estimations seem to contradict the conventional wisdom expressed above. They find that the negative impact of distance on bilateral trade increases over time (e.g. Leamer [1993], Frankel [1997], Smarzynska [2001]). In their review of the literature, Leamer and Levinsohn ([1995], p. 138788) note that “[…] the effect of distance on trade patterns is not diminishing over time. Contrary to popular impression, the world is not getting dramatically smaller”. Disdier and Head [2004] examine 1052 distance effects estimated in 78 papers and find that the negative impact of distance on trade is not shrinking, but increasing slightly over the last century. They conclude that there is a “[…] puzzling persistence of the distance effect on bilateral trade”.
3 Some recent studies have attempted to solve the puzzle using a gravity model in a series of crosssection over 19752000 (Coe et al. [2002]) or in a panel over 19621996 (Brun et al. [2005]). For standard gravity specifications, both studies find rising distance effects. When the model is estimated nonlinearly, Coe et al. [2002] find that the coefficient for the distance variable shows some decline in 19752000. With an augmented transport cost function, Brun et al. [2005] find a decline in the coefficient of distance in a loglinear gravity model, though the decline is largely confined to bilateral trade among rich countries; for developing countries, the coefficient of distance does not decline. Freund and Hummels [2003] find that fdi growth has contributed to increasing proximate trade but has had little impact on the elasticity of trade with respect to distance.
4 Given that the coefficient of distance in a gravity model measures the marginal impact of distance on bilateral trade, an increase in the absolute value of the coefficient of distance over time means that a marginal increase in the distance of trade has become more costly. However, it does not indicate how overall trade costs have changed.
5 The paper proposes an alternative way to solve the puzzle: we show that an increasing distance elasticity of bilateral trade is compatible with falling trade costs and that countries may trade at shorter distances over time even though overall trade costs decrease. The reason is that the decision about what proportion to trade at different distances does not depend on the level of trade costs but on the relative importance of its components [1].
6 The paper contributes to the literature in several ways. First, we opt for a different approach in order to elucidate the puzzle of the increased impact of distance on trade over time and develop a new measure of the distance of trade, dot. The evolution of this measure shows that the importance of distance has increased over time: the distance of trade (dot) appears to have declined between 1962 and 2000 for a majority of countries, with a stronger decline for developing than for oecd countries. To paraphrase Twain, we confirm that “The report on the death of distance has been greatly exaggerated.”
7 Second, the paper provides a simple analytical solution to the puzzle. Trade costs are decomposed into those unrelated to distance –known as dwell costs– and those related to distance. This decomposition is key to show that the dot falls as long as dwell costs fall relative to distance costs, irrespective of the direction of change in total transport costs or in either of its two components. This explanation is supported by econometric analysis, explaining around half of the negative trend.
8 Finally, the paper attempts to explain the entire negative trend of the dot and then proposes some additional determinants by examining the impact of additional determinants, including regional integration, and changes in the geography of growth. The set of variables introduced fully explain the negative trend in the dot for exports and total trade, and 60% of the trend for imports. On a reduced sample, the introduction of the changes in the geography of real exchange rates allows to explain the remaining negative trend for imports.
9 The remainder of the paper is organized as follows. Section 2 defines the dot measure and provides information on the average dot over the period 19622000. Section 3 offers evidence on the evolution of the dot in 19622000 for the world, its main regions and representative countries, and for exports, imports and total trade. Section 4 presents theoretical analysis of the evolution of the dot over time associated with changes in the relative components of trade costs, regional integration, the geography of growth and real exchange rates. We also analytically examine the impact on the dot of changes in production, customs and domestic transport costs, as well as in competition and tariffs. An empirical validation of some of the theoretical analysis is provided in Section 5. Section 6 concludes. Appendix 1 describes the data and several variables of interest and Appendix 2 provides information on transport costs.
Definition of the Distance of Trade (DOT)
Measure of the Distance of Trade
10 For each country, region, and for the world, we calculate the dot (and its evolution over time) for exports, imports and total trade. Denote the value of the nonfuel trade flow between countries i and j at time t by Z_{ij}, with Z = M (imports), X (exports) or T (total trade M + X).
11 Denote the share of the trade flows between countries i and j in the total trade of country i at time t by s^{Z}_{ijt}, with: [2]
Denote the distance between countries i and j by d_{ij}. Then, the distance DOT^{Z}_{it} of country i’s trade at time t is:
We compute the distance of exports, imports and total trade for 150 countries over 39 years (19622000) [3] from the comtrade bilateral (nonfuel) trade data and the spherical distance between the main economic cities of any pair of countries. [4] The total number of observations on the dot is 5,777. Data sources and computation are provided in Appendix 1.
Average Distance of Trade
13 The average DOT for 19622000 for various countries and regions is presented in the first two columns of Table 1. [5] What are the main results? First, the is about 50% larger for nonoecd countries (about 6,540 kms) than for oecd countries (about 4,390 kms). [6] Second, within the oecd, the eu15 and Canada have the smallest (about 2,800 kms), followed by the us (6,800 kms), Japan (8,500 kms), Australia (11,850 kms) and New Zealand (12,300 kms).
Average Level and Change in the Distance of Exports DOT^{X}_{i} and Imports DOT^{M}_{i}, 19622000
Average Level and Change in the Distance of Exports DOT^{X}_{i} and Imports DOT^{M}_{i}, 19622000
14 Third, when ranked by continent/region, the is smallest for the eu15 (2,800 kms), larger for the Middle East and North Africa (4,590 kms), over double the eu15 in North America (5,890 kms), followed by SubSaharan Africa (7,790 kms), Asia (8,085 kms), and South America (8,180 kms). Fourth, no country’s is below 5,000 kms in either South America or SubSaharan Africa.
Evolution of the Distance of Trade, 19622000
Trend of the Distance of Trade
15 The evolution of the dot can be examined for individual countries, regions and the world as a whole. We calculate the trend of the dot over time as the estimated value of β in the ols regression (with the White correction for heteroskedasticity):
Average Distance of Trade
Asia
Asia
Latin America and Caribean
Latin America and Caribean
17 and Change, 19622000
usa
usa
Canada
Canada
Evolution of the dot: World and Individual Countries
18 The change ΔDOT^{Z}_{i} is reported in Table 1 (columns 3 and 4) for the World and various regions, countries and trade blocs. We consider the change to be empirically significant if and only if ∣ΔDOT^{Z}_{i}∣ > 5.5%. [7] A country is defined as having a positive (negative) change in its dot if both exports and imports have a significantly positive (negative) change in their dot or if one has a significantly positive (negative) change and the other is not significant. And a country has opposite changes for imports and exports when they are both empirically significant but have opposite signs.
19 According to Table 1, the World presents no empirically significant change in the dot for imports or exports in 19622000, with ΔDOT^{X}_{w} = – 2.5% and ΔDOT^{M}_{w} = 2.9% [8]. We also estimate the trend of the dot for the average country in the world. This is done by running regression (3) on all (individual) country observations in the sample, and is also reported in Table 1 [9]. We find significantly larger (and negative) changes in the dot for the average country (– 12.0% for imports, and – 5.3% for exports) than for the World as a whole. In fact, at the country level, Table 1 (last column) shows a predominance of negative trends in the dot. The difference between the results of the two regressions indicates that countries with negative trends tend to be relatively small in terms of their share in world trade.
20 For the entire sample of 150 countries, we find that i) 77 countries (51.3%) have a significant negative change in the dot; ii) 39 countries (26%) have a significant positive change in the dot; iii) 30 countries (20%) present opposite changes in the dot; and iv) 4 countries (2.7%) have nonsignificant changes. [10] Thus, about twice as many countries show an empirically significant negative change as opposed to a positive change in the dot over time (77 to 39 countries or a ratio of 1.97). [11]
21 The ratio of countries with negative to positive changes in the dot is 1.67 for the oecd and 2.03 for the nonoecd. Thus, the decline in the dot is relatively more frequent for nonoecd countries, and the average annual trend in the dot for imports, exports and total trade is more negative for nonoecd countries than for the oecd (Tables 1 and A.1).
Evolution of dot: Regions and SubRegions
22 Except for the us, with a positive change of 8% for exports and 30% for imports (Fig. 3), and Canada, with opposite changes of – 42% for exports and 36% for imports (Fig. 4), other oecd countries show strong negative trends: the eu15 (– 12% for exports and – 13% for imports), Australia (– 23% and – 20%), Japan (– 17% and – 25%) and New Zealand (– 40% and – 23%).
23 Nonoecd countries trade significantly closer to home over time, with a decrease in the dot of 14% for imports and 7.4% for exports. However, there is much variation within that group, with negative changes in the dot in the two largest developing regions, lac (– 23% for exports and – 10% for imports; see Fig. 2) and nonoecd Asia (– 9.8% and – 26%), and positive changes in the smaller regions of SubSaharan Africa (2.9% and 12%) and mena (57.3% and 20.5%). [12]
24 The main results obtained thus far are:
 though there was little change in the dot for the World as a whole in 19622000, the dot fell for the average country;
 the number of countries for which the dot fell in 19622000 is close to double the number of countries for which the dot increased; and
 the dot fell more strongly in nonoecd countries than in the oecd.
Impact of Trade Costs and Other Factors on the Distance of Trade
25 The fact that, despite the decline in transport and communication costs, the dot fell for the average country and fell in many more countries than it rose over time, is puzzling. This section sets out a number of hypotheses about factors that are likely to affect a country’s or region’s dot and its evolution.
Transport Costs
26 The analysis focuses first on transport costs. Divide transport costs (TC) into two components, those unrelated to the distance traveled and which are referred to as “dwell” costs (L), and those related to the distance traveled, i.e., distance costs (DC). Dwell costs include port storage costs, the cost –including time– of loading and unloading ships, the time cost of queuing outside the port waiting to be serviced, and all other port costs. Total transport costs TC equal the sum of these two components, i.e.:
The log derivative of transport costs TC, i.e. the relative variation of , for a trip of given distance m, is
What do the data tell about the evolution of dwell costs relative to distance costs? There is little information on that, though some changes in technology point to a decline in relative dwell costs. For instance, containerization started in 1966 on North Atlantic routes, then spread to North AmericaAsia and EuropeAsia routes by the early 1970s. The share of world tonnage shipped by container increased from 2% to 55% in 19701996 and it increased faster and earlier in the us, from 40% in 1970 to 55% by 1979 (Hummels [1999]). Containerization lowered port labor costs and time in port, and though it probably also lowered distance costs, the cost reduction was most likely larger for the dwell (port) component. Containerization also reduced another component of dwell costs, namely the cost of the inland movement of goods by facilitating their transfer between different shipping modes. In that case, d(logL) [lt] d(logC_{m}) [lt] 0, implying ∂(d logTC)/∂m > 0 (equation (10)) and a reduction in the dot. Further details on the evolution of transport costs are provided in Appendix 2.
Fluctuations in the price of oil would also be expected to affect the dot. Prices jumped at the time of the oil embargo in 1973 and again in 1980, resulting in higher distance costs and an expected fall in the dot. Real oil prices have declined since the early 1980s (until 2000 when our sample period ends), with an expected increase in the dot.
Exchange Rates
28 Another issue is the effect of exchange rate policy on the dot. Many dwell costs are in local currency (e.g., port labor costs) while distance costs are typically quoted in us dollars. Thus, one can rewrite equation (7) to include the exchange rate as:
Note also that trade depends on bilateral real exchange rates and so that its impact on dot will depend on the distance between countries i and j and on the shares traded. To capture that effect, we define an index that captures the geography of the change in the real exchange rate (see p. 000).
Regional Integration
30 Regional integration agreements (rias) or trade blocs are typically formed between neighboring countries. [14] Given that the dot for intrabloc trade is typically smaller than for extrabloc trade and that rias tend to raise intrabloc trade by making it privately more beneficial, rias tend to reduce the dot of its member countries.
31 Since Viner’s [1950] classic work, the static economic effects of rias have been examined in terms of the concepts of trade creation and trade diversion. Whether trade creation or trade diversion dominates also affects the impact of rias on the dot. Trade creation increases trade among members of the ria (without affecting trade with excluded countries) and, given their relative proximity, reduces the dot. The negative effect of a ria on the dot is stronger with trade diversion because it also reduces trade with more distant excluded countries. Thus, for a given increase in trade among member countries, the greater the degree of trade diversion, the larger the reduction in the dot.
The Geography of Economic Growth
32 Another issue that can affect the dot over time is economic growth. Countries belonging to a region that experiences a high rate of economic growth will find it beneficial to trade relatively more with countries of the region. This will tend to lower these countries’ dot. This is the case, for instance, for the East AsiaPacific (eap) region: there is a negative correlation between eap’s growth rate relative to that of the world and its dot. This is confirmed by Frankel and Wei [1996] who, with the help of a gravity model, find that the increase in trade within East Asia “… can be entirely explained by the rapid growth of the countries.”
33 We also find a negative correlation between a region’s differential growth rate with the world and the trend in that region’s dot. For instance, nafta’s growth rate was lower than the world’s average before 1990 and higher in 19902000, and its dot increased in 19621989 and fell in 19902000. The mercosur region grew slightly faster than the world in 19621979, much slower than the world in 19801989, and faster than the world in 19902000, with the dot trend equal to – .05 in the first period, .20 in the second one, and – .76 in the third one.
Additional Determinants of Changes in dot
34 Other traderelated costs as well as nontrade costs affect the dot and are examined here. The cost to consumers in country j of a product imported from country i, P_{ji}, is:
Anderson and van Wincoop [2004] estimate the costs of shipping a good from a foreign producer to a domestic final user. These costs include transport, borderrelated and distribution costs, or (P_{ji} – C_{i}) in equation (11). The authors argue that these costs amount to 170% of production costs (with (Pji – C_{i})/C_{i} = 1.7) in rich countries and amount to significantly more in developing countries.
Collecting in equation (11) the nondistance costs, NDC, on the one hand, and the distance costs, DC, on the other hand, we have:
What about the effect of the advalorem tariff factor τ_{j}? A given reduction in τ_{j} has a larger proportional effect on DC than on NDC. The effect is equiproportional for DC (equation (14)) but is less than equiproportional for NDC because some of its terms are not affected by a reduction in τ_{j} (equation (13)). This raises the dot. On the other hand, the reduction in τ_{j} raises the degree of international market contestability and leads to a reduction in markups. This has the opposite effect of lowering the dot. Which effect dominates is ambiguous a priori. [15]
Other aspects are also examined in Carrère and Schiff [2003], including counterseason trade, international production fragmentation, and the increasing value of time in trade because of the increasing importance of the ability to respond to fluctuations in demand and supply.
Note that lack of data prevented empirical estimation of the effects presented in section 4.5. on the dot. However, an empirical analysis of the impact of the determinants developed in sections 4.14.4 is now provided.
Estimation of the Determinants of the DOT
36 In this section, we estimate for the full sample the impact on the dot of dwell costs, distance costs, regional integration and economic growth, and for a reduced sample the additional effect of real exchange rates. The mean value of these variables and of the dot, as well as their minimum, maximum and standard deviation, are presented in Table 2. The correlation between the variables is mostly negative and very low, with the largest (in absolute value) equal to – .30. Estimation in this section is carried out using a panel model with country fixed effects (i.e., the “Within” estimator). [16]
Trend
37 The regression of dot on a time trend variable t for 19642000 for the entire sample of 150 countries is shown in Table 3 (first columns of exports, imports and total trade). The estimated annual trend β is – .10% for exports, – .21% for imports and – .14% for total trade, significant at the 5% level for imports and total trade and at the 10% level for exports.
Determinants of the Distance of Trade (Within Estimator)
Determinants of the Distance of Trade (Within Estimator)
Dwell and Distance Costs
38 Detailed data on the evolution of dwell and distance costs are not available for most countries. We use the evolution of the real price of oil as a proxy for the evolution of distance costs, and changes in the countryspecific infrastructure index based on Canning [1996] and Limao and Venables [2001] as a proxy for changes in dwell and domestic transport costs. An increase in the price of oil raises the relative cost of the more distant trade, while an increase in the infrastructure index lowers the relative cost of the more proximate trade, with both expected to lead to a reduction in the dot.
39 The regression of the dot on the real price of oil, the infrastructure index and a time trend is also shown in Table 3 (second columns for exports, imports and total trade). The coefficient for the price of oil is negative, significant at the 5% level for imports and total trade and not significant for exports. The coefficient of infrastructure is also negative, significant at the 5% level for imports and total trade and at the 10% level for exports. Thus, the empirical results support our hypothesis. Note that these variables explain close to 50% of the trend for imports, exports and total trade (the trend coefficients fall by close to 50%). This suggests that dwell costs have fallen relative to distance costs. [17]
Regional Integration
40 We have two dummy variables for Regional Integration Agreements (rias), one for all rias except the eu, and one for the eu. The reason for a separate dummy variable for the eu is that the latter is a much deeper ria and we want to test whether it has had a stronger impact on the dot. The results are provided in Table 3 (third columns of exports, imports and total trade). The ria dummy is negative, significant at the 10% level for exports and total trade and not significant for imports. The eu dummy is negative, significant at the 5% level for imports and total trade and at the 10% level for exports. Note that, as expected, the impact of the eu on the dot is significantly larger than that of the other rias.
41 Moreover, the trend variable no longer has any explanatory power for the dot of exports and total trade, and the trend coefficient for the dot of imports has been reduced by 60% in absolute value (– .08% versus – .21%) and its significance has been reduced from 5% to 10%. Thus, our explanatory variables fully explain the trend in the dot for exports and total trade and explain around 60% of the trend in the dot for imports.
42 Carrère and Schiff [2003] examined regional integration and the dot in more detail and found that all have a negative impact on the dot of their member countries (Detailed estimation results on the individual impact of eight rias are provided). Nevertheless, as shown in table A.2, the ratio of positive to negative trends of the dot was found to be close to twice as large for ria than for nonria countries. One hypothesis is that there are some positive externalities associated with increasing trade with neighboring countries, such as increased security and other political and institutional benefits (Schiff and Winters [2003]), and that such externalities provide an incentive for countries with the more positive dot trends to use regional agreements to capture them.
Geography of Economic Growth
43 We test the impact of uneven economic growth on the dot by constructing an index that indicates for each country whether high growth occurred mainly in proximate or in distant countries. For each country i, the index of relative growth REG_{it} is:
One would expect that if the absolute change in gdp is larger in more distant countries and reg increases, the dot for exports and total trade also increases. The results are shown in Table 3 (last column of exports, imports and total trade). As expected, the coefficient of reg is positive, significant at the 10% level for exports and total trade, and not significant for imports. The elasticity is larger for exports than for total trade because the elasticity for imports is small and not significant. The latter is to be expected since the measure reg relates directly to the demand for a country’s exports. The high growth in distant countries need not raise a country’s imports. We obtain similar but statistically less significant results with the current rather than the lagged reg variable.
Real Exchange Rate
45 We examine here the impact of the real exchange rate (rer) on the dot, in addition to the variables examined above. The results are not directly comparable with those above because the sample used here is less than two thirds of the size of the full sample (2,713 observations rather than 4,119).
46 Data on exchange rates are from the ifs data base (imf) and the nominal exchange rate is defined as the number of units of domestic currency per us dollar. An increase in the rer means a depreciation of the domestic currency and a decrease in nondistance costs (many of which are in domestic currency) relative to distance costs. This lowers the relative cost of trade at closer distances and lowers the dot. This is shown in Table 4 (columns 2 for exports, imports and total trade). The coefficient of the rer variable is negative, significant at the 5% level for exports and total trade and either significant at the 10% level or not significant for imports.
Impact of Real Exchange Rate on the Distance of Trade (Within Estimator)
Impact of Real Exchange Rate on the Distance of Trade (Within Estimator)
Geography of Real Exchange Rates
47 Trade also depends on bilateral real exchange rates RER_{ij} between countries i and j. If a country j devalues its currency relative to that of home country i, the appreciation of RER_{ij} is likely to result in a decrease (increase) in exports from i to j (imports from j to i). The impact on DOT_{i} will depend on the distance between countries i and j and on the shares traded. To capture that effect, we define an index that captures the change in the RER_{ij} of country i relative to all its trading partners j, weighted by the bilateral distance d_{ij} and trade share s_{ij}:
Conclusion
49 It has been widely argued that the importance of distance has declined with the reduction in transport and communication costs and the integration of the global economy. On the other hand, gravity models find an increasingly important impact of distance on trade. This paper examines this puzzle and makes several contributions. First, it develops a new measure of the distance of trade (dot) and presents findings on its evolution for individual countries, regions and for the world. We find that the dot falls over time for the average country in the world, and that the number of countries with declining dot is close to double those with increasing dot. In other words, distance has become increasingly important over time for a majority of countries.
50 Second, the paper examines analytically a number of hypotheses in order to explain the evolution of the dot. One of the conclusions is that the negative evolution of the dot is compatible with falling trade costs as this evolution is unrelated to that of the overall level of trade costs but depends on the relative evolution of its components. Specifically, the dot falls over time as long as dwell costs fall relative to distance costs, irrespective of the direction of change in total transport costs or in either of its two components.
51 Third, the paper shows that reductions in production, domestic transport and customs costs, and increases in competition and the real exchange rate, result in a decline in the dot and that the impact of a change in advalorem tariffs is ambiguous.
52 Fourth, the paper provides an empirical analysis of the dot for exports, imports and total trade. Explanatory variables (and their sign) include a trend variable (negative), regional integration (negative), a variable reg measuring geographically uneven growth (positive), an infrastructure index whose value increases as dwell and domestic transport costs fall (negative) and the price of oil as a proxy for distance costs (negative). Thus, all variables have the expected sign, most are significant, and their inclusion fully explains the negative trend in the dot for exports and for total trade and explains 60% of the trend for imports. In a smaller sample, we add the real exchange rate (negative impact on the dot), and a geographic variable relrer measuring changes in bilateral real exchange rates (positive for exports and negative for imports). These variables also have the expected sign, are significant, and explain the remaining negative trend for imports.
Data, Sample and Computations
53 This study is based on nonfuel trade data from 1962 to 2000 of 150 countries [19] from the comtrade (un). The list of the available countries is in Table A.1. These countries account for more than 90% of world trade. The distance of trade (dot) is computed for each country and year using these trade data and the spherical distance between the main economic cities of any pair of countries. The source for the location of capitals is the cia World Factbook. The calculations of the spherical distances are our own.
54 To overcome missing data problems, when a country’s import data are not available, mirror estimates (export data reported by the partner countries) are used (and similarly for missing export data). This approach has the advantage of covering almost all the missing data. [20] Once the dot per country and year is computed using the database with mirror estimates, we have 5,777 observations (98.5% of the potential number of data points), [21] rather than 4,641 for imports and 4,670 for exports in the data base without mirror estimates. Information on the number of data per country and year are available from the authors.
55 The infrastructure index includes the density of roads, paved roads, railways, and telephone lines for each country and year (see Limao and Venables [2001]; Brun et al. [2005]) [22]. This index (in annual percentage change) captures both the impact of the evolution of domestic transport costs and of the evolution of dwell costs. The correlation between this infrastructure index and a port efficiency index, for a sample of 44 developing and oecd countries for which data on port efficiency in 1998 are available, is 0.70. [23] Similarly, the correlation between this infrastructure index and a custom clearance index is – 0.59. [24]
Trend (in percentage) in the Distance^{a)} of:
Trend (in percentage) in the Distance^{a)} of:
Countries per Category (number in RIA/number of countries/) = (96/150)
Countries per Category (number in RIA/number of countries/) = (96/150)
Transport Costs
56 Information on general or liner cargo does not distinguish between dwell and distance costs, though that for charter shipping does. Hummels [1999] argues that for charter shipping bulk commodities (on a worldwide basis) as well as for general or liner cargo (for ships loading and unloading in Germany and the Netherlands), including containerized vessels, the cost per value shipped has risen since 1952. However, Lundgren [1996] concludes that the constant dollar price of shipping bulk commodities fell substantially between 1950 and 1993, though not the advalorem barrier of shipping bulk commodities (Hummels [1999]). Since the figures for charter shipping do not include port costs, the increase in charter shipping distance costs should have a negative impact on the dot. On the other hand, the evidence on US air cargo rates indicates very large distance cost reductions between 1955 and 1977, which may explain the increase in the us dot over time.
57 The bulkiness (and/or weight) of many tradable products has fallen over time, resulting in a fall in C_{m} and an increase in the dot for any given mode of transport. [25] With the fall in air transport costs as well as in many products’ bulkiness, there has also been a gradual shift from ocean to air transport over time, with a further increase in the dot. In a model of choice between ocean and air transport, Carrère and Schiff [2003] show that a fall in bulkiness leads to a rise in air relative to ocean travel.
58 International trade between neighboring countries is typically made over land. Glaeser and Kohlhase [2003] find that that US overland transport costs have declined, with the cost of moving a ton a mile by rail falling by 2.5% a year since 1890, and trucking costs falling by 2% a year since the Motor Carrier Act of 1980. They attribute this to improved transport technologies and to the fact that the value of goods lies increasingly in quality rather than quantity. They also find a positive relation between products’ value per ton and the distance hauled. Indirect evidence also suggests that overland transport costs in the US declined relative to ocean transport costs (Hummels [1999]). The fall in us overland shipping costs provides an incentive to increase overland trade, resulting in an increase in the dot over land but in a reduction in the overall dot (due to the increased share of overland trade).
Us Trade by Transport Mode
Us Trade by Transport Mode
Notes

[*]
Université de Lausanne, Département d’Économétrie et d’Économie politique (deep), Hautes Études Commerciales (hec), Email: Celine. Carrere@ unil. ch

[**]
Development Research Group, World Bank, 1818 H St., NW, Washington, DC 20433, Email: mschiff@ worldbank. org; and cerdi, Université d’Auvergne.

[1]
The idea of the asymmetric evolution of different trade cost components is mentioned in Brun et al. ([2005], appendix A.4) but the gravity model does not allow a test of this hypothesis.

[2]
“Trade flows” always refer to “nonfuel trade flows”.

[3]
In reality, we cannot state s_{ijt} ≠ 0 for all the countries i whatever j and t. Hence, the number of trading partners changes from country to country and over time. However, the measure of Distance of Trade (DOT_{ijt}) is comparable across countries and time as it does not depend on the number of trading partners. Actually, we are interested in the share of trade of each country i at various distance, whatever the number of partners concerned for a given distance.

[4]
In an analysis at a highly disaggregated level, Berthelon and Freund [2004] find no impact of compositional changes in trade on the distance elasticity of trade over time. This increases our confidence in the adequacy of using aggregate data.

[5]
Carrère and Schiff ([2003], Table A.1) provide details about each country’s sample with and without mirror data.

[6]
The oecd is defined here as the oecd in 2000, with 23 member countries (and 22 observations because Belgium and Luxembourg are considered as one country in the comtrade database).

[7]
The cutoff value of 5.5% is arbitrary. Qualitative results remain unchanged with cutoff values of 10% or even 15%.

[8]
Why is there a difference in the change in the dot for exports and imports for the World as a whole? The first reason is that some countries are missing in our sample because of definitional changes during the period (e.g. the 15 exussr countries). Second, the difference between cif and fob values in the weights of DOT^{M}_{w} (distance weighted by the cif value of imports) and DOT^{X}_{w} (distance weighted by the fob value of exports) combined with higher cif/fob ratios at greater distances results in DOT^{M}_{w} > DOT^{X}_{w} and thus in the likelihood that ∆DOT^{M}_{w} ≠ ∆DOT^{X}_{w}.

[9]
The results reported in Table 1 are obtained with ols. They are not qualitatively different when we use a “within” estimator by introducing country fixed effects in equation (3).

[10]
The full list of countries in each category is provided in Table A.2.

[11]
The number of countries with negative changes remains much larger than that with positive changes when we consider a cutoff point of 10% rather than 5.5% (70 to 41 or a ratio of 1.72). The results are similar for total trade (as compared to those for imports and exports). With a 5.5% cutoff point, we find that 80 (43) countries have a significant negative (positive) change in the dot, with a ratio of 1.86.

[12]
The trend in the dot in trade blocs is provided in Table A.1 and that across subperiods is examined in detail in Carrère and Schiff [2003].

[13]
Note that C_{m} need not be constant.

[14]
On the actual 208 Preferential Trade Agreements in force in 2004 (i.e. notified to the gatt/wto), 160 (77%) are implemented between countries of a same region. Source: World Trade Organization secretariat and Author’s calculation.

[15]
The us applies its advalorem tariff on the fob value of the product. The fob value does not include transport costs D_{ij} (or dwell costs L_{j} in j). In that case, equation (14) becomes DC = D_{ij}, and a reduction in the us advalorem tariff factor τ_{us} lowers NDC but not DC. This reduces the dot for us imports. On the other hand, us tariffs have been low for a while, with a decline from 3.8% to 1.8% in 19892001 (World Bank [2003]), so that this effect on the us dot is likely to have been small.

[16]
We tested for unit roots for panel data. The test (Im, Pesaran and Shin [2003]) significantly rejects the null hypothesis of unit roots.

[17]
Brun et al. [2005] estimate a gravity model for bilateral imports which includes the price of oil and an infrastructure index. They do not examine the interaction effects between these variables and the distance elasticity.

[18]
In other words, the implicit assumption is that the marginal propensity to import is constant and the same for all countries. Note that all countries in the sample are included, whether country i trades with them or not.

[19]
Actually the sample covers more than 150 countries as data concerning BelgiumLuxembourg and sacu (Southern African Customs Union) is not presented for each individual country.

[20]
Mirror statistics also have some shortcomings, especially for trade between developing countries where they do not always match the original data.

[21]
The potential number of data points is 5850 (= 150*39).

[22]
Each country’s infrastructure is measured by an index constructed from four variables from the Canning [1996] dataset: km of road, km of paved road, km of rail (each per sq. km of country area), and telephone main lines per person. We took the mean over the four variables (each being normalized to have a mean equal to one, See Limao and Venables [2001], Appendix 1). As the final year of the Canning [1996] dataset is 1995, we used the predicted value of the infrastructure index for 1996 to 2000 according to a quadratic trend estimated by country on the 19621995 available data.

[23]
The port efficiency index goes from 1 (inefficient port) to 7 (most efficient port) and is based on surveys of representative firms in each country. Source: The Global Competitiveness Report, various years [19962000]; also available in Appendix B in Clark, Dollar and Micco [2002] for 1998.

[24]
The customs clearance index corresponds to the time (median number of days) needed to clear customs, based on surveys performed (by the World Bank) with respect to importers in each country. Source: Appendix B in Clark, Dollar and Micco [2002] for 1998.

[25]
The share of light manufactures in the exports of developing countries to developed ones increased over time, from 5% in 1955 to 58% in 1992 (Hillman [2003]), reducing the average bulkiness of trade.

[26]
Note that as far as the choice between shipping modes is concerned, the evolution of total transport costs matters rather than that of dwell versus distance costs.