81273 What Constrains Africa’s Exports? Caroline Freund and Nadia Rocha Africa’s share of global exports has dropped by 50 percent over the last three decades. To stem this decline, aid for trade to the region has increased rapidly in recent years. Assistance can target improvements in three important components of trade facili- tation: transit times, documentation, and ports and customs. Of these, transit delays have the most economically and statistically significant effect on exports. Specifically, a one day reduction in inland travel times leads to a 7 percent increase in exports, after controlling for the standard determinants of trade and potential endogeneity. Put another way, a one day reduction in inland travel times translates into a 2 percentage point decrease in all importing-country tariffs. By contrast, longer delays in the other areas have a far smaller impact on trade. Large transit delays are relatively more harmful because they are associated with high (within-country) variation, making delivery targets difficult to meet. Finally, the results imply that transit times are pri- Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 marily about institutional features—such as border delays, road quality, fleet class and competition and security—and not geography. JEL codes: F13, F14, O55. Sub-Saharan Africa’s share of world merchandise exports fell from 1.6 percent in 1980 to 0.8 percent today.1 Africa’s weak export performance is also appar- ent in an examination of export levels. Controlling for the standard determi- nants of trade, export volumes in Africa are about 16 percent below what is expected. In sum, over the last 30 years Africa has halved its world market Caroline Freund (cfreund@worldbank.org; corresponding author) is the Chief Economist of the Middle East and North Africa Region of The World Bank. Nadia Rocha (nadia.rocha@wto.org) is an Economic Officer in the Economic Research and Statistics Division, of the World Trade Organization. Authors would like to thank Allen Dennis for providing disaggregated data from the Doing Business report and the GPS team at the World Bank for providing detailed GPS data on Sub-Saharan Africa travel distances and times. In addition, authors would like to thank the editor of the journal, three anonymous referees, and seminar participants at the World Bank Trade seminar, the Nottingham Conference on Trade Costs and International Integration in Venice, the Geneva Trade and Development Workshop, and the European Trade Study Group (ETSG) conference for useful comments and suggestions. This article received financial support from the governments of Finland, Norway, Sweden and the United Kingdom through the Multidonor Trust Fund for Trade and Development. The views presented in this article are those of the authors and do not reflect the views of World Bank or the World Trade Organization. 1. In this article, Africa refers to Sub-Saharan Africa. Analysis uses a balanced sample of 167 countries (44 in Africa) reporting trade over the period. THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 3, pp. 361 –386 doi:10.1093/wber/lhr016 Advance Access Publication May 30, 2011 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 361 362 THE WORLD BANK ECONOMIC REVIEW share for exports, and today Africa exports less than expected given its income, population, location and other characteristics. Africa’s relatively poor export performance is worrisome for a number of reasons. First, export growth can substitute for lagging domestic demand. This is especially important in small economies, such as in Africa, where foreign markets are likely to be the main engines for growth (Bhagwati 1996, Krueger 1998). Numerous studies examine the effect of openness to trade on income growth, and in general find positive effects.2 Second, robust export growth yields both more jobs and better jobs. In particular, exporting firms create jobs that pay higher wages and offer better working conditions than otherwise similar import-competing firms.3 Third, strong export growth induces a more efficient production structure. This happens through compositional shifts, as the most productive exporting firms grow most rapidly when exports boom.4 Finally, strong export growth helps prevent against the negative effects of balance of payments crises. This article investigates what constrains Africa’s exports, with a focus on Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 trade facilitation. The results show that long delays in getting export products from the factory gate and onto the ship explain much of Africa’s weak export performance. This is consistent with earlier work that finds that domestic export delays hinder exports significantly, and to a much greater extent than foreign tariffs (Hummels (2001), Djankov, Freund and Pham (2010), and Portugal and Wilson (2009)). This is especially debilitating for Africa’s exports because of extreme and often unexpected delays. This suggests that improving trade facilitation in Africa would significantly boost exports. But there are different ways to accomplish this, as the domestic delay has three distinct components: documentation, transit time, and port handling and customs clearance. In this article it is explored whether these delays are equally burdensome or whether one of these binds relatively more, using detailed data on average trade times from the World Bank’s Doing Business report. This is important from a policy perspective in order to target aid for trade initiatives to their most productive uses. This is especially relevant for Africa—where aid for trade has increased more rapidly than any other region in recent years— making Africa now the second largest recipient (after Asia) of aid for trade (OECD/WTO 2009). Of the various delays, bureaucratic ones are the longest, taking 19 days on average. There is a lot of variation across countries. For example, it takes 36 days to process export documents in countries such as Angola, Zambia and Niger. In contrast, in Swaziland or in the Seychelles, it takes only 7 and 5 days 2. See Winters (2004) for a summary of the literature on trade and growth. 3. Bernard and Jensen (1995) report detailed statistics for the United States. A number of papers followed their approach and find similar results in both developing and developed economies. Schank, Schnabel and Wagner (2007) provide a summary of these papers and offer similar evidence for Germany. Bernard, Jensen, Redding, and Schott (2007) also provide a summary. 4. See Bernard, Jensen, Redding, and Schott (2007) for a summary of the literature. Freund and Rocha 363 respectively to produce all necessary export documents. Bureaucratic delays may be especially burdensome if they change often, making them difficult to predict, or if officials use them as means to extract rents. In contrast, documen- tation procedures may be less problematic if they are predictable and can be done in advance, or if there is learning by doing. Customs and ports delays are the second longest, taking on average 9 days. They are less variable than documents. Customs and ports could be especially restrictive if there is a hold-up problem. Once the goods arrive, customs and port authorities could extract high rents by delaying goods. In contrast, if customs and ports are reliable (but slow) or if exporters can pay for faster service they may cause fewer problems. Transit costs are on average the shortest, taking 7 days. But, again, there is a lot of variation. For example, it takes 31 days for the goods to be shipped from the city to the port in Chad and only one day within Gabon. Transit costs may be less burdensome if economic activity has developed endogenously, close to ports and borders when transit costs are large. However, they may be more Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 constraining if there is a lot of uncertainty that cannot be avoided. The main contribution of this article is to understand whether different types of export costs affect trade differently. A modified gravity equation that controls for importer-fixed effects and exporter remoteness is used. An impor- tant concern with this approach is that the volume of trade may directly affect trade costs. The marginal value of investment in trade facilitation is higher when trade volumes are large since cost savings are passed on to a larger quan- tity of goods. In addition, many time-saving techniques, such as computerized container scanning, are only available in high-volume ports. Alternatively, increased trade volumes could increase congestion and lessen the efficiency of trade infrastructure. Thus, while more efficient trade facilitation may stimulate trade, trade is also likely to directly influence trade facilitation. Three distinct strategies are used to deal with the potential effect of export volumes on export times. First, the effect of trade facilitation on trade in new products is examined. These are goods that have not been exported in the past. The intuition is that trade in new products cannot affect the quality of trade facilitation infrastructure or the bureaucracy that is in place for exporting. Second, the effect of requirements in the transit country on exports from land- locked countries is analyzed. This controls for endogeneity because trade facili- tation in transit countries is likely to be exogenous from the perspective of a landlocked country. Finally, it is tested whether lengthy delays have a greater effect on exports of time-sensitive goods. The intuition is that these products make up a small share of total trade so are unlikely to affect trade facilitation. All three different techniques used to analyze the effect of export times of key components on trade values lead to the same conclusion: inland transit delays have a robust negative effect on export values. The estimation results imply that a one day increase of inland transit times reduces export values by about 7 percent. This effect is higher for time-sensitive goods with respect to 364 THE WORLD BANK ECONOMIC REVIEW time-insensitive goods. In contrast, the effects of documents, customs and ports on exports are much smaller. Why would delays in one area affect trade relatively more than in other areas? One potential answer is the variability of each time component. To evaluate this explanation, the effect of within-country export-time uncertainty on export values is examined for each type of delay. While an increase in transit-time uncertainty has a negative and significant effect on trade values, the other time components show no such effect. This suggests that long and unexpected delays in transit make it especially difficult for producers to meet export deadlines. These results have important implications for policy. While reducing bureau- cratic delays and improving ports and customs may have positive effects on trade, the binding constraint in most African countries to expanding exports is inland transit. Improving inland transit is unlikely to be easy or cheap, but it is likely to boost exports and have broad positive economic effects. Beyond these direct implications for policy, the results presented in this Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 article also contribute to the broader debate about the influence of geography versus institutions on income. This literature has focused on the effects of climate versus governance on income, and potential interactions between the two.5 Here, the focus is on a single component of income (exports), and the variables of interest reflect geography and institutions to different extents. The dominance of transit time in hindering exports seems to suggest that geography is the main culprit. To test this, data from a GPS system on geo- graphical distance from the port to the economic center and on the estimated time of travel is gathered and included in the regression equation. The differ- ence between travel time in the GPS data and the Doing Business data is that the former are based solely on travel distance and estimated speed of travel by type of road ( paved or unpaved). These data do not incorporate delays due to the type of vehicles, borders, security, traffic, or other road conditions. The results show that GPS distance negatively affects exports, but GPS travel time does not. Moreover, neither the economic effect nor the statistical significance of the Doing Business inland transit-time variable changes when these vari- ables are included. This suggests that the problem for inland transit lies in the quality and security of the roads, border delays and the efficiency of security checkpoints, the age of the truck fleet and competition in trucking. These are factors which are more closely linked with institutions than geography. The article proceeds as follows. The next section discusses the data. Section II presents the estimation strategy. Section III describes the main results and robustness checks. Section IV examines the effect of uncertainty on exports. Section V determines the importance of purely geographical transit costs. Section VI concludes. 5. See, for example, Hall and Jones (1999), Acemoglu, Johnson, and Robinson (2000), and McArthur and Sachs (2001). Freund and Rocha 365 I . D ATA Data on trade times is based on answers to a comprehensive World Bank ques- tionnaire completed by trade facilitators at freight-forwarding companies in 146 countries in 2007 and is collected as part of Doing Business, a World Bank project that investigates the scope and manner of business regulations. Freight-forwarders are the most knowledgeable to provide information on trade costs because most businesses use their services to move their products across borders. They are estimated to handle approximately 85% of global international trade. Their services range from finding the most appropriate route for a shipment, preparing documentation to meet customs and insurance requirements, arranging payments of fees and duties, and advising on legisla- tive changes and political developments that could affect the movement of freight. Overall, 345 trade facilitators participated in the survey, with at least two per country.6 To document the procedures and export times, the survey respondents are asked about a stylized transaction. The exporter is a local business (100% Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 owned by nationals), has 201 employees, and is located in the country’s most populous city. The exporter does not operate within an export-processing zone or an industrial estate with special export privileges. Each year, more than 10% of its sales go to international markets, i.e., management is familiar with all the trading rules and requirements. The purpose of defining the exporter specifically is to avoid special cases. Assumptions are also made on the cargo to make it comparable across countries. The traded product travels in a dry-cargo, 20-foot, full container load. It is not hazardous and does not require refrigeration. The product does not require any special phytosanitary or environmental safety standards other than accepted international shipping standards, in which cases export times are likely to be longer. Finally, every country in the sample exports this product category.7 The questionnaire also asks respondents to identify the likely port of export. For some countries, especially in Africa and the Middle East, this may not be the nearest port. For example, Cotonou, Benin’s main port, is seldom used due to perception of corruption and high terminal handling fees. The novel part of the data used for this investigation is on the distinctions by type of trade time. The data provide detailed information on the different kinds of costs an exporter faces when moving his goods from the principal city to the port of exit. More precisely, the survey asks respondents the average and 6. Follow-up conference calls were conducted with all respondents to confirm the coding of the data. As a further quality check, surveys were completed by port authorities and customs officials in a third of the sample (48 countries). 7. These assumptions yield three categories of goods: textile yarn and fabrics (SITC 65), articles of apparel and clothing accessories (SITC 84), and coffee, tea, cocoa, spices and manufactures thereof (SITC 07). 366 THE WORLD BANK ECONOMIC REVIEW F I G U R E 1. Export Procedures by Category Source: Authors’ calculations on Doing Business data. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 the maximum times in calendar days it takes for completing a series of export procedures. Each procedure can be classified into one of four main categories: documentation, inland transportation, customs, and ports. The first category records the time it takes for an exporter to complete all documentation activities such as securing a letter of credit, assembling and pro- cessing export and international shipping certificates and realizing all pre- shipment inspections and clearance. Inland transportation includes the time it takes for the merchandise to be moved from the principal city to the port of exit, as well as the time spent arranging transport and waiting times for the merchandise’s pick up and loading into a carriage. For landlocked countries, total transport times also include waiting times at the crossing border. The customs category includes the time necessary to realize the technical controls of the merchandise. In addition, for landlocked countries this cat- egory comprises the total time it takes from the submission of request of clearance until the completion of the inspection and clearance procedure in the transit country. Finally, the ports category represents terminal handling times, including storage if a certain storage period is required, the waiting times for loading the containers into the vessel and customs inspection and clearance times. An example illustrates the data. An exporter in Rwanda spends 43 days on average to complete all requirements for shipping its merchandise abroad: 17 days each on delays resulting from documentation and inland transit, while port and custom procedures take respectively 6 and 3 days on average (see Figure 1). Freund and Rocha 367 T A B L E 1 . Times to Export Descriptive Statistics by Geographic Region Region Statistics Documents Customs and ports Inland transit East Asia & Pacific (23) mean 12 8.5 3.9 sd 9.7 4.9 3.1 min 1 2 1 max 39 19 15 Europe & Central Asia (25) mean 13.8 7.6 9.6 sd 9 6.4 14 min 1 2 1 max 32 34 58 Latin America & Caribbean (30) mean 11.2 7.3 3.9 sd 6.7 3.6 3.1 min 4 2 1 max 30 18 18 Middle East & North Africa (12) mean 10.3 6.1 3.6 sd 3.4 2.7 1.9 min 5 3 2 max 18 13 8 OECD (24) mean 5 3.1 2 Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 sd 3.2 1 1 min 2 2 1 max 14 6 4 South Asia (8) mean 16.3 8.6 7.6 sd 11.5 2.6 6.9 min 9 5 1 max 44 12 21 Sub-Saharan Africa (45) mean 18.7 9.4 7.2 sd 9 4.2 7 min 5 2 1 max 41 28 31 Notes: 1. The unit of measure is number of days. 2. Number of countries for each region in parenthesis. Source: Authors’ analysis based on Doing Business data. The summary statistics for each of the components representing the total time8 necessary to fulfil all the requirements for exporting by region and regional arrangement are presented in Table 1. The data show that across regions, docu- mentation procedures times are the longest. Furthermore, while getting a product from the factory to the ship is relatively quick in developed countries, this is not the case for Sub-Saharan Africa, where all time costs categories are on average higher compared to all the other regions. Customs and ports procedures and inland transportation take on average three times more in African countries than in OECD countries. In addition, documentation procedures take four times longer in African countries compared with developed countries. 8. The time delays reported in the survey are probably at the lower end of the time it takes to move the average product from factory to ship. This is because the products are chosen so that they do not require cooling or any technical inspections based on use of hazardous materials. 368 THE WORLD BANK ECONOMIC REVIEW The rest of the data are from standard sources. The trade data are both from the UN Comtrade database and the IMF Direction of Trade database. GDP and Population are from the World Bank’s World Development Indicators. Gravity variables such as country-pair distances, language and colony are taken from the Mayer and Zignago dataset. Country’s Capital abundance information is avail- able for 2005 and comes from GTAP 7 database. Simple average tariffs at 6 digit level are taken from the TRAINS dataset. Aid for trade data is collected for the 2006 and comes from the OECD/DAC Creditor Reporter System on disburse- ments. Table 2 presents correlations between the variables used in the analysis. II. ME T H O D O LO GY In order to examine the effect of trade cost an augmented gravity equation is estimated as a first step: LnExportsij ¼ b1 Ln GDPi þ b2 Ln Popi þ b3 LnDistij þ b4 Ln Remotei ð1Þ Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 þ b5 landlocki þ Xij þ mj þ 1ij ; where the i and j subscripts correspond to the exporter and the importer, respectively.9 The dependent variable is the log of bilateral exports from country i to country j. The standard determinants of trade are: mj, importer- fixed effects, which control all importers-specific characteristics; GDPi and POPi are respectively the Gross Domestic Product and the total population of the exporting country; Distij is the distance between country i and country j. Xij is a vector of dummy variables associated with the exporter and the impor- ter such as sharing the same official language or border or past colony/coloni- zer relationship. Landlocki is a dummy variable equal to one if the exporter country is landlocked and zero otherwise. Remotei is a measure for the exporter’s remoteness and is calculated following Head (2003), Remotei ¼ P GDP 1 .10 To examine whether African trade is different, this j =Distij j equation is estimated on all available data and a dummy that is one if the exporter is an African country is included in the regression. A negative coeffi- cient would imply that Africa undertrades relative to other countries. The main purpose of this article, however, is to investigate how three diverse trade costs—completing documentation, inland transit delays, and customs and ports times—affect Africa’s trade volumes. Longer time delays act as a tax on 9. The times representing terminal port handling and customs and technical control were aggregated due to their very high correlation (See table 2). 10. It is important to control for remoteness in the regressions for two reasons. First, there is evidence that a country’s trade with any given partner is dependent on its average remoteness to the rest of the world (Anderson and Van Wincoop (2003)). Furthermore, remoteness is correlated with factory-to-port time delays hence not including it into the regression would produce biased estimates of the impact of trade facilitation on export volumes. T A B L E 2 . Correlation of Explanatory Variables Travel Total Inland Aid GPS Dist. time Uncert Uncert. export Inland Docs Customs Ports transp. for city to city to Uncert custom & Inland GDP POP time Docs Customs Ports transp. (TC) (TC) (TC) (TC) Remote Trade port port Docs Ports Transit GDP 1 POP 0.73 1 Total Export time 0.04 0.25 1 Documents 0.01 0.19 0.84 1 Customs 0.22 0.22 0.31 0.10 1 Ports 0.28 0.17 0.35 0.13 0.39 1 Inland transport 2 0.11 0.12 0.72 0.35 0.07 2 0.02 1 Docs (TC) 0.03 0.14 0.47 0.66 0.20 0.26 2 0.06 1 Customs (TC) 0.16 0.08 0.26 0.14 0.76 0.36 2 0.01 0.35 1 Ports (TC) 0.22 0.05 0.33 0.16 0.28 0.89 2 0.01 0.27 0.36 1 Inland transp. 0.07 0.17 0.36 0.18 0.18 0.24 0.34 0.26 0.15 0.14 1 (TC) Remote 2 0.15 2 0.08 2 0.24 2 0.11 2 0.23 2 0.12 2 0.23 2 0.10 2 0.10 2 0.13 0.02 1 GPS dist. city 0.03 0.21 0.65 0.48 0.09 2 0.07 0.71 2 0.03 0.00 0.03 2 0.06 2 0.38 1 to port GPS travel time 0.00 0.19 0.68 0.47 0.06 2 0.07 0.80 2 0.06 2 0.04 2 0.02 2 0.02 2 0.34 0.98 1 city to port Aid for Trade 0.37 0.55 0.16 0.07 0.26 0.00 0.17 2 0.16 0.09 2 0.11 0.10 2 0.11 0.27 0.23 1 Uncert. Docs 0.01 2 0.18 2 0.19 0.01 2 0.23 2 0.17 2 0.25 0.18 2 0.27 2 0.11 2 0.03 0.69 2 0.29 2 0.24 2 0.12 1 Uncert. Custom 0.45 0.14 2 0.01 0.14 2 0.53 2 0.13 2 0.03 2 0.11 2 0.55 0.26 2 0.10 0.27 0.16 0.17 2 0.06 0.32 1 & Ports Uncert. Inland 0.05 2 0.22 0.57 0.58 2 0.19 0.10 0.26 0.14 2 0.27 0.13 2 0.27 0.09 0.44 0.44 2 0.13 0.09 0.24 1 Transit Notes: TC stands for transit country. Freund and Rocha Source: Authors’ analysis based on data sources discussed in the text. 369 Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 370 THE WORLD BANK ECONOMIC REVIEW exports, especially on high-value goods, since they are effectively depreciating during the delay. In addition, the exporter must spend capital on the exporting process and storage/transport of the goods during the delay. For this analysis, equation (1) is estimated including the Doing Business (DB) variables: LnExportsij ¼ b1 Inland transit i þ b2 Customs & Portsi þ b3 Documentsi þ b4 Ln GDPi þ b5 Ln Popi þ b6 LnDistij ð2Þ þ b7 Ln Remotei þ b8 landlocki þ Xij þ mj þ 1ij : The variables of interest are the export times for transit, customs and ports, and documents. The coefficient on each represents the effect in percent of trade of a one day increase in that component. The variables are analyzed in levels, so that the coefficients are comparable—the effect of a one day change. However, for robustness, the regression with the three variables in logs is also estimated. The previous specification could be subject to omitted-variables bias given that the error term might include the effect of country-specific policy variables Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 that affect both trade flows and the time-to-export variables. A potential vari- able of concern is aid for trade. Aid affects trade directly through productive capacity assistance or technical assistance for trade policy and also through trade-related infrastructure assistance. To control for this, a lagged variable capturing the total amount of aid-for-trade assistance received by the exporting countries is included. There is a potential reverse causality problem in the regressions because time-to-export variables are likely to be correlated with country exports. An improvement of infrastructure and administrative time costs has positive effects on exports. However, countries that export more may have higher returns to enhance local trade facilitation and invest more in time efficient means. In addition, some types of export processing might only be available in high volume ports. Hummels and Skiba (2004), for example, provide evidence that trade volumes affect the adoption time of containerized shipping, which greatly reduces shipping costs. Finally, it might be the case that in countries with higher volumes of trade, export procedures will be affected by congestion effects. In this case, the presence of reverse causality will lead to an underesti- mation of the coefficient on time costs. To control for the possibility that more trade leads to improved trade facili- tation, the effects on exports of new products are investigated.11 The intuition is that exports of new products cannot have had an impact on the historical develop- ment of infrastructure or the type of bureaucratic procedures in place. In addition, because they are a very small share of total trade, they are unlikely to be associ- ated with congestion effects. The approach of Djankov, Freund, and Pham (2010) is also followed. First, trade times of transit countries are used as instruments for 11. New products are defined as those products that were not exported in the years 2002-2004 and that entered into the export market in the time interval 2005-2007. Freund and Rocha 371 trade costs in landlocked countries and second, it is examined whether trade times affect time-sensitive goods relatively more. II I. R ES ULT S As a first step, Africa’s exports are examined in comparison with the rest of the world. The augmented gravity equation (1) is estimated on all countries with available data, and a dummy for sub-Saharan Africa is introduced. Recall, con- trols for income, population, distance, common border, language, colonial heritage, landlocked, remoteness and importer-fixed effects are included in this regression. Results are reported in Table 3. The negative and significant value of the Africa dummy (column 1) implies that countries in the region export about 16 percent (e20.18 2 1 ¼ 2 0.16) less than expected.12 Column (2) includes the level of the total export time and the coefficient on the Africa dummy falls from Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 2 0.18 to 2 0.11 and becomes significant at the 10 percent level only. This implies that trade costs are a significant part of the explanation for Africa’s weak performance. In the next two columns, the effect of trade costs in all countries is compared with the effect of trade costs in Africa. Column (3) reports the basic regression with export times for all countries and column (4) reports the same regression only for sub-Saharan Africa. The time to export is a more important determinant of African trade than overall trade. Finally, the last two columns report the results with the time variable in logs and the nature of the results remains—time is a much bigger deterrent to African trade than general trade. While it would be interesting to perform this type of analysis for each region, this article focuses on Africa because export growth has been relatively weak and trade costs are especially important. In addition, it is likely that the 12. Rodrik (1997) and Coe and Hoffmaister (1999) find that Africa’s low trade can be explained with enough control variables. This is not inconsistent with our results, as the Africa dummy loses significance when trade costs are included. In addition, if only income, distance, and partner effects are included, the coefficient on the Africa dummy more than doubles in magnitude, highlighting that Africa’s weak trade performance is in part explained by the large number of landlocked countries and the general remoteness of the region. Still, there are some important distinctions between the methodology used in this article methodology and theirs. Both Rodrik and Coe and Hoffmaister examine exports and imports together, and many African countries run persistent trade deficits. Rodrik uses exports plus imports relative to GDP (openness) as the dependent variable, which makes the regression less theoretically based and not a test of export performance. In addition, he controls also for Latin American, East Asian, and OECD countries with dummies—so his measure of Africa’s performance is as compared with the Middle East, Eastern Europe and Central and South Asia, other traditionally weak trade performers. Coe and Hoffmaister examine only trade flows between South and North countries. Still, they find very similar results in their basic specification: Africa’s trade is significantly below what is expected in a regression with income, population, and distance. They show that if you control for openness to trade, type of exports, language and a number of fixed effects it may go away. However, even these variables do not change their other main result, that Africa’s trade performance has significantly weakened over recent decades. 372 T A B L E 3 . Is Africa Different? All countries All countries All countries Sub-Saharan Africa All countries Sub-Saharan Africa Dependent Variable: (levels) (levels) (levels) (levels) (logs) (logs) ln (Aggregate exports) (1) (2) (3) (4) (5) (6) GDP 1.248*** 1.136*** 1.149*** 1.080*** 1.104*** 1.060*** [0.014] [0.019] [0.017] [0.053] [0.020] [0.053] Population 2 0.098*** 0.013 0.001 0.004 0.042* 0.047 [0.017] [0.021] [0.020] [0.050] [0.022] [0.051] THE WORLD BANK ECONOMIC REVIEW Distance 2 1.474*** 2 1.474*** 2 1.474*** 2 1.186*** 2 1.474*** 2 1.170*** [0.027] [0.027] [0.027] [0.127] [0.027] [0.128] Sub-Saharan Africa dummy 2 0.180*** 2 0.114* [0.069] [0.069] Total time to export 2 0.024*** 2 0.024*** 2 0.056*** 2 0.626*** 2 2.005*** [0.002] [0.002] [0.006] [0.055] [0.203] Observations 16,085 16,085 16,085 3,494 16,085 3,494 R-squared 0.675 0.678 0.678 0.525 0.678 0.524 Notes: 1. Robust standard errors in brackets. ***p , 0.01, **p , 0.05, *p , 0.1. 2. Other control variables: partner FE, common language, common border, colony, remoteness, landlocked. Source: Authors’ analysis based on data sources discussed in the text. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 Freund and Rocha 373 relative importance of the type of trade costs varies by region. This would make a single analysis of the three types of costs in five regions quite complex. To focus on the three types of costs in the African sample, the augmented gravity equation from expression (2) is estimated.13 The linear regression results for a sample of 45 Sub-Saharan Africa countries are reported in Table 4. In order to control for the presence of zero trade flows, a nonlinear estimation with censored data is also performed (see column (2) of Table 4).14 The first two columns show the results from estimation on all trade. In both cases all three variables are significant and their coefficients are similar, though it is somewhat higher for inland transit.15 However, this column does not deal with the problem of endogeneity of the right hand side variables. In column (3), results for trade in new goods only are reported. The time variables are less likely to be endogenous to trade in new goods, since this trade was not around in the past when procedures and infrastructure for trade were devel- oped. The results are somewhat different. While the coefficient on inland transit is little changed from column (1), the other coefficients fall consider- Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 ably, suggesting that the previous column was also picking up the effect of trade on documentation procedures and customs and ports. In particular the results imply that a one day increase in transit time leads to a nearly 7 percent decline in exports. In the next five columns, robustness tests are reported. Columns (4)-(6) report the results of each variable independently and total time. This helps to deal with potential multicollinearity between the variables and also informs us whether each variable is significantly different from total time in its effect on exports. Only inland transit has an independent effect on exports. Moreover the total effect of inland transit, equivalent to 0.066 (0.049 þ 0.017), is nearly four times as large as the effect of the other components of time. This outcome holds after the inclusion of foreign import tariffs in the regression (see column (7)).16 Including foreign tariffs also allows interpreting a day in terms of tariffs. A one day delay is roughly equivalent to a 2 percent point reduction in all importer-country tariffs. Importer tariffs are on average 10 percent and on average transit time is 7 days. This implies that cutting the transit time in half would expand trade by 30 percent; while all importers cutting tariffs in half 13. Equation (2) is estimated excluding trade in oils and minerals from the dependent variable to control for the surge in commodities exports that took place in some African countries during the 2007. 14. A Tobit regression with left-censoring at zero is run. 15. Results from OLS and Tobit estimations are qualitatively similar. However, in the second case the magnitude of the coefficients more than doubles. This might be explained by the fact that almost 30% of the observations are left-censored. 16. To control for the fact that some African countries benefit from preferential tariffs a country-pair specific dichotomous variable that takes the value of 1 when the partner grants lower preferences to the reporter through the generalized system of preferences (GSP) program and zero otherwise is also introduced. This variable is never significant, but is included in the regressions as a control (coefficients not reported). 374 T A B L E 4 . The Effect of Export Time Components on Aggregate Exports (OLS Regression) New New New New New New OLS Tobit Products Products Products Products Products Products Dependent variable: (levels) (levels) (levels) (levels) (levels) (levels) (levels) (logs) ln (Aggregate exports) (1) (2) (3) (4) (5) (6) (7) (8) Inland transit time 2 0.073*** 2 0.198*** 2 0.066*** 2 0.049*** 2 0.070*** 2 0.417*** [0.012] [0.025] [0.015] [0.017] [0.024] [0.134] Customs and ports time 2 0.052*** 2 0.129*** 2 0.018** 0.022* [0.007] [0.014] [0.007] [0.013] Documents time 2 0.045*** 2 0.128*** 2 0.016 0.012 [0.015] [0.027] [0.014] [0.016] GDP 1.070*** 1.979*** 0.982*** 1.019*** 0.992*** 0.984*** 1.030*** 0.975*** [0.054] [0.109] [0.063] [0.060] [0.062] [0.061] [0.085] [0.061] THE WORLD BANK ECONOMIC REVIEW Population 2 0.017 0.050 2 0.323*** 2 0.351*** 2 0.328*** 2 0.324*** 2 0.414*** 2 0.281*** [0.055] [0.117] [0.066] [0.065] [0.066] [0.065] [0.090] [0.067] Distance 2 1.210*** 2 3.565*** 2 0.899*** 2 0.889*** 2 0.868*** 2 0.900*** 2 1.029*** 2 0.868*** [0.128] [0.287] [0.161] [0.161] [0.160] [0.161] [0.196] [0.160] Aid for Trade 0.039 0.368*** 0.079* 0.088** 0.074* 0.080* 0.038 0.073* [0.037] [0.074] [0.042] [0.042] [0.041] [0.041] [0.056] [0.041] Total export time 2 0.040*** 2 0.029*** 2 0.017*** 2 0.016* 2 0.502** [0.010] [0.006] [0.006] [0.008] [0.220] Tariffs (simple average) 2 0.048** [0.023] Observations 3,494 5,737 2,054 2,054 2,054 2,054 1,142 2,054 R-squared 0.525 0.425 0.423 0.422 0.425 0.433 0.424 Notes: 1. Robust standard errors in brackets. ***p , 0.01, **p , 0.05, *p , 0.1. 2. Other control variables: partner FE, common language, common border, colony, remoteness, landlocked and GSP (whenever import tariffs are included in the regressions). Source: Authors’ analysis based on data sources discussed in the text. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 Freund and Rocha 375 would expand trade by 25 percent.17 Finally in column (8) results using logs of the time variables are reported. Again, it can be seen that only transit time is independently significant (results for other variables are not reported). The second strategy to deal with the potential endogeneity of the export time variables is to use a sample of landlocked countries and use the variables for the transit country(ies) as the instrument. This follows from Djankov, Freund and Pham (2010).18 Results are reported in Table 5. The first column reports OLS regression results for this sample. With the exception of docu- ments, the coefficients are much larger for this sample than for the full sample (column (1) of Table 4). One explanation is that the endogeneity problem is greater here. For example, when landlocked country trade is small, customs and ports authorities (which must be located in neighboring countries) give them the lowest priority. To control for endogeneity, each time variable is instrumented with the corresponding variable faced by exporters in the transit country. The F-statistics of the first stage regressions (see Table 1A of the Appendix) indicate that none of the instruments is a weak instrument.19 Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 Second stage regression results, with and without foreign tariffs, are reported in levels in columns (2) and (3) and in logs in columns (4) and (5). Transit trade from a given country tends to be a small share of total coastal country trade, on average 27 percent. Still, this can be a problem if large land- locked exporters have more influence on times getting through transit countries. As robustness check, the estimated GPS distance in km from the border between the landlocked and the transit country and the port of the latter20 is also used as an alternative instrument for inland transit times. Results in levels and logs are reported in columns (6) and (7). For landlocked countries, the instrumental variables (IV) results show that only inland transit has a robust negative and significant effect on trade.21 Moreover, the magnitude from column (2) is similar to the result using all 17. From column 7, the aggregate coefficient on transit is 2 0.086 ( 2 0.070 þ 2 0.016) and the coefficient on tariffs is 2 0.048. Thus, the effect of 3.5 days is 3.5* 2 0.086 ¼ 2 0.30, while the effect of a 5 percentage point cut in tariffs is 5*.048 ¼ 2 0.24. 18. There are two differences with this study. First, Djankov, Freund, and Pham use a difference gravity equation on similar exporters. This strategy is not necessary for this investigation since in the regressions only sub-Saharan Africa countries are considered. Second, while they use the actual times in the transit country as instrument, here the time for the transit county’s trade in the transit country is used. 19. Since this regression is perfectly identified, it is not possible to test whether the excluded instruments are not correlated with the error terms. In table A2 of the Appendix the regressions of table 4 are replicated but this time including separate instruments for customs and ports. In this case the Sargan overidentification test supports the validity of the instruments. 20. This variable is computed taking into account geography and type of road. 21. This result does not only reflect the fact these countries are more isolated. Even though delays in inland transport are higher with respect to coastal countries (15 days versus 4 days on average), delays in documentation (24 days on average) and customs and ports (9 days on average) procedures are even higher for exporters in landlocked countries. 376 T A B L E 5 . The Effect of Export Time Components on Aggregate Exports, Landlocked Sample Regression OLS IV IV IV IV IV IV Dependent Variable: (levels) (levels)a (levels)a (logs)a (logs)a (levels)b (logs)b ln (Aggregate exports) (1) (2) (3) (4) (5) (6) (7) Inland transit time 2 0.125*** 2 0.097*** 2 0.082*** 2 1.788*** 2 1.453** 2 0.126*** 2 2.212*** [0.015] [0.020] [0.031] [0.501] [0.726] [0.033] [0.530] Customs and ports 2 0.252*** 0.113 0.066 0.027 2 0.169 0.208 1.193 [0.053] [0.190] [0.228] [1.706] [2.259] [0.204] [3.287] Documents time 2 0.047*** 0.012 0.019 2 0.412 2 0.083 0.053 0.818 [0.012] [0.047] [0.051] [1.054] [1.281] [0.052] [2.087] GDP 0.389** 0.120 0.700* 0.042 0.670** 0.355 0.433* THE WORLD BANK ECONOMIC REVIEW [0.178] [0.257] [0.370] [0.218] [0.329] [0.311] [0.254] POP 0.555** 2 0.322 2 0.493 2 0.051 2 0.245 2 0.836 2 0.512 [0.249] [0.519] [0.654] [0.424] [0.612] [0.610] [0.800] Distance 2 1.102*** 2 0.940*** 2 1.386*** 2 0.995*** 2 1.388*** 2 1.441*** 2 1.466*** [0.293] [0.284] [0.353] [0.288] [0.347] [0.361] [0.357] Tariffs (simple av.) 2 0.055* 2 0.054* 2 0.070** 2 0.070** [0.030] [0.029] [0.029] [0.030] Observations 991 991 479 991 479 479 479 R-squared 0.569 0.542 0.544 0.563 0.560 0.524 0.541 Notes: 1. Robust standard errors in brackets. ***p , 0.01, **p , 0.05, *p , 0.1. 2. Other control variables: partner FE, common language, common border, colony, remoteness, landlocked aid for trade and GSP (whenever import tariffs are included in the regressions). a. Instruments: documents, customs and ports and inland transit times in transit countries. b. Instruments: documents and customs and ports times and GPS distance from border between the landlocked and the transit country and the port of the latter. Source: Authors’ analysis based on data sources discussed in the text. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 Freund and Rocha 377 T A B L E 6 . The Effects of Export Time Components on Time Sensitive Products (OLS regression) Countries exporting at Countries exporting least one product 70% of the products Dependent Variable (levels) (logs) (levels) (logs) ln (Aggregate Exports by industry) (1) (2) (3) (4) Inland transit time*Time sensitivity 2 0.023* 2 0.173* 2 0.038** 2 0.226** [0.013] [0.090] [0.016] [0.101] Customs and ports time*Time sensitivity 2 0.002 0.003 2 0.002 0.039 [0.022] [0.191] [0.024] [0.208] Docs time*Time sensitivity 0.014 0.247 0.015 0.219 [0.009] [0.187] [0.009] [0.196] K abundance*Canned product 0.513** 0.544** 0.689*** 0.715*** [0.222] [0.222] [0.233] [0.231] Observations 626 626 519 519 R-squared 0.519 0.518 0.542 0.541 Notes: Robust standard errors in brackets ***p , 0.01, **p , 0.05, *p , 0.1. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 Source: Authors’ analysis based on data sources discussed in the text. countries and new trade (column (3), Table 4): a one day reduction in transit delays leads to about 9 percent more exports.22 Finally, the effects of documentation, inland transit and customs and ports times on the exports of time-sensitive products are analyzed. All time delays should have a greater effect on exports of time-sensitive goods. To examine the extent to which they are hampered, the methodology in Djankov, Freund and Pham (2010) is followed and an estimation of a difference-in-difference gravity equation is performed using trade data of agricultural ( processed and unpro- cessed) products for which time matters the most and the least. This method- ology reduces the endogeneity problem coming from reverse causality because it controls for country- and industry-fixed effects. In addition, the products considered account for only a small fraction of trade in agricultural goods on average (less than 10 percent) so it is unlikely that they have a large impact on establishing trade facilitation processes. The definition of time-sensitive agricultural products is based on the infor- mation of their storage life (Gast 1991), which includes a range of products going from a minimum storage life of 2 weeks or less, such as apricots, beans, currants, and mushrooms to 4 weeks or longer, for example apples, cranberries and potatoes and canned products. Goods with a very long storage life such as dry fruits with a maximum storage life of between 6 months and one year and canned products with a storage life ranging from 1 to 5 years, depending on 22. When the log of total time is included as the only trade cost variable in the regression, the estimated coefficient is about 2 1, the same as the results for developing countries in Djankov, Freund and Pham (2010), although they use a slightly different approach. 378 THE WORLD BANK ECONOMIC REVIEW the good’s acidity are also included. To measure time sensitivity, the inverse of the median storage life of each product is used. To study the joint effect of industry time-sensitivity and country-time delays on exports the following difference-in-difference gravity regression is estimated LnExportsik ¼ ai þ ak þ b1 ðTime Sensitk Þ Â ðInland transit timei Þ þ b2 ðTime Sensitk Þ Â ðCustoms & Ports timei Þ ð3Þ þ b3 ðTime Sensitk Þ Â ðDocs timei Þ þ b4 ðK abundancei Þ Â ðcanned productk Þ þ 1ik , where ai and ak represent country- and industry-fixed effects. The coefficients b1, b2 and b3 capture the joint effect of time-sensitive products and time delays in inland transit, customs and ports, and documentation on export values. The term (Kabundancei) Â (canned productk) is introduced to control for the fact that more capital abundant countries are more likely to have the necessary resources and technologies to process fresh food into canned products. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 With this specification it is tested whether exports of time-sensitive goods are more responsive to time delays in each of the key components of time to export than exports of time-insensitive products. For example, if b1 is negative and significant, it means that a longer transit delay reduces time-sensitive exports more than time-insensitive exports. The key advantage of this approach is that it controls for industry- and country-fixed effects, so significant results cannot be a result of more trade being associated with more efficient trade facilitation, and shorter delays. The simple correlation between the share of time-sensitive goods23 in trade and transit times is low (0.02), so it also cannot be due to countries with greater exports of time-sensitive goods having lower times. Results for time-sensitive agricultural products controlling for countries’ capital abundance are presented in Table 6. Results are reported for countries exporting at least one product and countries exporting at least 70 percent of the products, and also with the variables in logs and levels. In all cases, the coefficient on the interaction term of inland transit times with time sensitivity is negative and significant, and highly significant when intensive exporters are considered. This implies that an increase in inland transit times reduces exports of time-sensitive goods relatively more than time-insensitive goods. In contrast, interactions with documents and customs and ports times are never significant. Transit delays affect the composition of trade, preventing countries from exporting time-sensitive agricultural goods. Time-sensitive goods also tend to have higher value, implying that some of the effects of inland transit delays on aggregate exports results from countries with poor trade-facilitation programs concentrating on low-value, time-insensitive goods. 23. To calculate the share, products with a storage life of 2 weeks or less are classified as time-sensitive. Freund and Rocha 379 In sum, three different ways to examine the effects of various trade delays on trade flows are used, each of which should reduce the endogeneity problem inherent in the analysis. All three point to the same conclusion: delays during inland transit affect trade flows to a much greater extent than delays because of documentation or at the port. These results imply that reducing time spent on inland transit will significantly stimulate trade in Africa. I V. W H Y D O E S I N L A N D T R A N S I T M A T T E R M O R E ? All else equal, a one day delay should affect exports the same way no matter when it occurs. However, one reason it may not is if there is more uncertainty associated with high delays in some procedures than in others. Uncertainty will reduce exports because it makes delivery deadlines harder to meet. In this section it is investigated whether greater uncertainty related to inland transport times makes costs related with documents, customs and ports become a sec- ondary priority relative to travel costs for existing exporters. The effects of time uncertainty in each component of export times are esti- Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 mated for a sub-sample of 22 Sub-Saharan countries for which there is infor- mation on the maximum and the average number of days it takes for an exporter to complete each of the exporting procedures:24 LnExportsij ¼ b1 Inland transit time uncerti þ b2 Customs & Ports time uncerti þ b3 Docs time uncerti þ b4 Ln GDPi þ b5 Ln Popi ð4Þ þ b6 LnDistij þ b7 Ln Remotei þ b8 Landlocki þ Xij þ mj þ 1ij Time uncertainty is defined as the difference between the maximum time and the average time it takes to conclude each of the different phases representing the total time to export. Unfortunately, since estimates are only from two or three freight forwarders in each country, it is not possible to use more sophisticated measures of uncertainty like the standard deviation of times in the country. Results from Table 7 show a negative and significant impact of inland transit time uncertainty on trade values, with a one day increase in this variable leading to a reduction of exports of more than 15 percent (column (1)). Or in logs (column (2)), a one percent increase in uncertainty leading to about a one percent reduction in exports. In contrast, uncertainty in the other variables is not robustly significant in reducing exports. The coefficient on documentation time uncertainty is negative and significant only when considered in levels and its magnitude is much smaller compared to inland transit uncertainty. Ports ˆ te d’Ivoire, Ghana, 24. Benin, Botswana, Burkina Faso, Cameroon, Republic of the Congo, Co Kenya, Madagascar, Malawi, Mali, Mauritania, Mozambique, Namibia, Nigeria, Rwanda, Sierra Leone, South Africa, Tanzania, Uganda and Zambia. 380 T A B L E 7 . The Effect of Time Uncertainty on Aggregate Exports, Full Sample OLS Regression Dependent Variable: Levels Logs Logs Levels Logs Levels Logs ln (Aggregate Exports) (1) (2) (3) (4) (5) (6) (7) Inland transit time uncertainty 2 0.154*** 2 0.978*** 2 0.104*** 2 0.467*** [0.021] [0.131] [0.022] [0.149] Ports and customs time uncertainty 0.021 0.399** 2 0.066 [0.018] [0.179] [0.115] Documentation time uncertainty 2 0.017** 2 0.174 [0.008] [0.115] GDP 1.655*** 1.623*** 1.482*** 1.377*** 1.319*** 1.147*** 1.116*** THE WORLD BANK ECONOMIC REVIEW [0.091] [0.091] [0.087] [0.091] [0.100] [0.071] [0.071] Population 2 0.736*** 2 0.662*** 2 0.410*** 2 0.286** 2 0.252** 2 0.058 2 0.125 [0.137] [0.136] [0.129] [0.121] [0.117] [0.108] [0.107] Distance 2 1.332*** 2 1.305*** 2 1.329*** 2 1.411*** 2 1.439*** 2 1.428*** 2 1.464*** [0.206] [0.204] [0.205] [0.189] [0.188] [0.177] [0.177] Inland transit Time 2 0.079*** 2 0.676*** 2 0.117*** 2 0.972*** [0.017] [0.137] [0.015] [0.104] Observations 1,602 1,602 1,668 1,713 1,713 1,713 1,713 R-squared 0.607 0.608 0.591 0.603 0.606 0.598 0.604 Notes: 1. Robust standard errors in brackets. ***p , 0.01, **p , 0.05, *p , 0.1. 2. Other control variables: partner FE, common language, common border, colony, remoteness, landlocked aid for trade. Data for uncertainty is available for 22 countries. See footnote 24. Source: Authors’ analysis based on data sources discussed in the text. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 Freund and Rocha 381 and customs time uncertainty shows a positive and significant coefficient only when entered in logs. This counterintuitive result might be due to the presence of multicollinearity with other time uncertainty variables. In fact, when the regression is performed including only customs and ports uncertainty, its coeffi- cient becomes negative and insignificant (column (3)). These results imply that high uncertainty in road transport times jeopardizes delivery targets. In addition, even if documentation requirements take more time than inland transit, they can either be done in advance or there may be learning by doing, such that exporters become more familiar with the procedures and uncertainty is limited. Finally, while exporters may be able to pay in the port to get things out more quickly, nothing can be done on the road. In columns (4) and (5) both inland travel times and inland travel uncertainty are included in the regression. The coefficients reflecting both variables are signifi- cant. When only inland transit is included (see columns (6) and (7)) in the same sample the coefficient is larger (as compared with columns (4) and (5)), implying that part of the effect of transit time on exports stems from uncertainty. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 V. G E O G R A P H Y V E R S U S I N S T I T U T I O N S The dominant role of transit suggests that geography is the main component of trade facilitation. But transit is about more than geography. It is also about road quality, fleet class and competition, border delays, road security and other institutional issues. Next it is investigated how much of the transit effect is pure geography and how much is institutional. Specifically, to control for dom- estic geography the GPS estimated distance and time based solely on geography and type of road is used. The regressions include the road distance in km from the principal city to the port of export (which is the relevant distance for which transport is calculated in the data). In addition, GPS-estimated total travel time is included. This variable is calculated as the total time it takes to get from the principal city to the port of exit by assuming a speed of 40 km per hour for unpaved roads and 80 km per hour for paved surfaces.25 If transit is primarily a geography effect then the GPS variables should pick up such effect. Indeed, both GPS variables are highly correlated with inland transport (0.71 for distance and 0.82 for time, Table 2). In addition, the difference between the Doing Business time and the GPS time is included, which should reflect institutions, since the GPS records how long it should take if there are no truck problems, road problems, borders, etc. The results using the full sample are reported in Table 8. The first column is for reference. The second and third columns include GPS distance and GPS time. Both coefficients are negative, but only distance is significant. Neither alters the effect of inland transit time on exports. 25. No information on road condition is used in the calculation of GPS travel time. Furthermore, delays at the border (or otherwise) are not included. 382 T A B L E 8 . Geography versus Institutions Dependent Variable: Levels Levels Levels Levels Logs ln (Aggregate Exports) (1) (2) (3) (4) (5) Inland transit time (levels) 2 0.089*** 2 0.087*** 2 0.080*** [0.012] [0.013] [0.014] GDP 1.079*** 1.111*** 1.121*** 1.121*** 1.071*** [0.054] [0.057] [0.056] [0.056] [0.056] POP 2 0.108** 2 0.120** 2 0.209*** 2 0.209*** 2 0.064 [0.055] [0.055] [0.057] [0.057] [0.060] Distance 2 1.166*** 2 1.157*** 2 1.226*** 2 1.226*** 2 1.226*** THE WORLD BANK ECONOMIC REVIEW [0.131] [0.131] [0.134] [0.134] [0.134] GPS distance principal city to port (km) 2 0.073*** [0.022] GPS travel times city to port (days) 2 0.110 2 0.190 2 0.465*** [0.133] [0.126] [0.126] DB times - GPS times 2 0.0800*** 2 0.803*** [0.014] [0.111] Observations 3,494 3,494 3,347 3,347 3,347 R-squared 0.514 0.515 0.518 0.518 0.523 Notes: 1. Robust standard errors in brackets. ***p , 0.01, **p , 0.05, *p , 0.1. 2. Other control variables: partner FE, common language, common border, colony, remoteness, landlocked aid for trade. Source: Authors’ analysis based on data sources discussed in the text. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 Freund and Rocha 383 Next, the geographical and institutional components of transit are explored separately. Specifically, the GPS transit time (geography) and the difference between Doing Business transit time and GPS transit time (institutions: DB times – GPS times) are introduced. The latter should reflect road delays that are not due to geography and distance. Column (4) shows the results. Only the institutional component is significant. Finally, Column (5) includes both vari- ables in logs. In this case both the geographical and institutional parts of transit times are significant, but the coefficient on the institutional component is significantly larger (F-Stat ¼ 3.46 of test that they are equal, p-value 0.06). In sum, the results imply that the distance from city to port and whether roads are paved are not the main reason for long delays in transit. There might be other factors such as the quality of the roads and vehicles, accidents, compe- tition in trucking, road blocks or border waiting times which affect the total time for an exporter to get his goods form the factory to the port of exit. This is good news in the sense that these institutional aspects of transit are likely to be more amenable to change than geographical ones. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 VI. CONCLUSION Export performance in Africa has been poor. It has been argued that this is a result of relatively slow income growth in the region. But that explanation is not entirely satisfactory, as export growth surely contributes to income growth. It could just as convincingly be argued that the failure of African exports to surge—as they did in the fast-growing developing regions over this period—is the cause of the lacklustre income growth. Moreover, even controlling for income and population, exports in Africa fall short of expectations. In this article an attempt is made to understand one of the important constraints to Africa’s exports. Detailed data on key components on the time it takes to move containerized products from the factory gate to the ship is used to estimate whether and how diverse trade costs affect export volumes in Sub-Saharan Africa. An augmented gravity equation is estimated by regressing aggregate bilateral exports on differ- ent time delay components such as inland transit, documentation, ports and customs, and other standard gravity variables. To control for the possibility that more trade leads to improved trade facili- tation, the effects that documentation, inland transport, customs and ports times respectively have on the exports of new products is analyzed. Exports in these products are unlikely to have an impact on the historical development of infrastructure. As a robustness check an instrumental variables approach is also used to examine the effect of time trade costs in transit countries on the exports of landlocked countries. Finally, a “difference-in-difference” regression is estimated on a sub-sample of agricultural products to determine whether trade costs affect exports of time-sensitive and time-insensitive goods, ranging 384 THE WORLD BANK ECONOMIC REVIEW from perishable products where time is most critical relative to preserved goods such as tinned food, differently. Our results imply that while inland transit delays have a robust negative impact on export values, higher times in other areas have much smaller effects in reducing Africa’s exports. A one day increase in inland transit time reduces exports by 7 percent on average. Put another way, a one day reduction in inland travel times translates into nearly a 2 percentage point decrease in all importing-country tariffs. In addition, this effect is higher for time-sensitive goods compared to time-insensitive goods. It is shown that long times are associated with high uncertainty in road transport, which jeopardizes expor- ters’ delivery targets. The empirical results have important policy implications. Export tariffs in Sub-Saharan African countries are already at a very low level. Furthermore these countries have preferential access to markets such as the United States and the European Union. Hence, while the benefits from a further decrease in tariffs among trading partners might be very small – or even negative in terms Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 of preference erosion if tariff reductions are MFN – reducing transport times will significantly increase their exports. Trade facilitation programs should therefore prioritize those programs directly affecting truck fleets and the infra- structure and security of Sub-Saharan Africa’s road systems. REFERENCES Acemoglu, D., S. Johnson, and J.A. Robinson (2001). “Reversal of Fortune: Geography and Institutions in the Making of the Modern World Income Distribution,” NBER Working Papers 8460. Anderson, J.E., and E. Van Wincoop (2003). “Gravity with Gravitas”. American Economic Review, vol. 93(1), pp. 170–192. Bernard, A., and B. Jensen (1995). “Exporters, jobs, and wages in U.S. manufacturing: 1976–1987” Brookings Papers on Economic Activity. Microeconomics, pp. 67– 119. Bernard, A., B. Jensen, S. Redding, and P. Schott (2007). “Firms in International Trade” Journal of Economic Perspectives, vol. 21(3), pp. 105– 130. Bhagwati, J. (1996). “The Demands to Reduce Domestic Diversity among Trading Nations” in J. BhagwatiR. Hudec (eds) Fair Trade and Harmonization. Prerequisites for free Trade. Cambridge, Massachusetts; London. The MIT press. Coe, D., and A. Hoffmaister (1999). “North-South Trade: Is Africa Unusual?” Journal of African Economies, vol. 8(20), pp. 228 –256. Djankov, S., C. Freund, and C.S. Pham (2010). “Trading on time,” Review of Economics and Statistics, vol. 92(1), pp. 166–173. Gast, K. (1991) “Postharvest Management of Commercial Horticultural Crops” Kansas State University Agricultural Experiment Station and Cooperative ExtensionService. Hall, R., and C. Jones (1999). “Why Do Some Countries Product so Much More Output per Worker than Others?” Quarterly Journal of Economics, vol.114 (1), pp. 83– 116. Head, K. (2003). “Gravity for Beginners” Mimeo University of British Columbia, Vancouver, B.C. Hummels, D. (2001). “Time as a Trade Barrier”. GTAP Working Papers 1152, Center for Global Trade Analysis, Department of Agricultural Economics, Purdue University. Freund and Rocha 385 Hummels, D., and A. Skiba (2004). “A Virtuous Circle: Regional Trade Liberalization and Scale Economies in Transport” in A. Estevandeordal, and D. Rodrik (eds) FTAA and Beyond: Prospects for Integration in the America. Harvard University Press. Krueger, A.O. (1998). “Why Trade Liberalization is Good for Growth” in The Economic Journal, vol.108 (450), pp. 1513–1522. McArthur, J.W., and J.D. Sachs (2001). “Institutions and Geography: Comment on Acemoglu, Johnson and Robinson (2000)” NBER Working Papers N. 8114. OECD and World Trade Organization (2009). Aid for Trade at a Glance 2009: Maintaining the Momentum. Paris: OECD and Geneva: WTO. Portugal, A., and J. Wilson (2009). “Why Trade Facilitation Matters to Africa” World Bank, Policy Research Working Paper, 4719. Rodrik, D. (1997). “Trade Policy and Economic Performance in Sub-Saharan Africa” Mimeo, Harvard University. Schank, T., C. Schnabel, and J. Wagner (2007). “Do exporters really pay higher wages? First evidence from German linked employer–employee data” Journal of International Economics, vol. 72(1), pp. 52– 74. Winters, L.A. (2004). “Trade Liberalization and Economic Performance: An Overview” Economic Journal, vol. 114 (493), pp. F4– F21. World Bank (2007). Doing Business. World Bank, Washington DC, www.doingbusiness.org. Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 APPENDIX T A B L E A 1 . Summary Results for First Stage Regressions Partial R2 F statistic p-value Inland transit time (levels) 0.4123 90.98 0.000 Customs and ports time (levels) 0.3529 70.70 0.000 Documents time (levels) 0.3741 77.50 0.000 Inland transit time (logs) 0.3007 55.75 0.000 Customs and ports time (logs) 0.4607 110.78 0.000 Documents time (logs) 0.3895 82.72 0.000 Source: Authors’ analysis based on data sources discussed in the text. T A B L E A 2 . IV Landlocked Sample Regressions using Customs and Ports as Separate Instruments Dependent variable: Levelsa Levelsa Logsa Logsa Levelsb Logsb ln (Aggregate exports) (1) (2) (3) (4) (5) (6) Inland transit time 2 0.096*** 2 0.096*** 2 1.789*** 2 1.640** 2 0.124*** -2.176*** [0.018] [0.028] [0.484] [0.713] [0.030] [0.478] Customs and ports 0.084 0.229 2 0.007 1.374 0.177 0.637 [0.104] [0.148] [1.372] [1.867] [0.152] [1.288] Docs time 0.005 0.055 2 0.434 0.698 0.043 0.444 [0.025] [0.033] [0.845] [1.080] [0.029] [0.529] GDP 0.153 0.468* 0.042 0.566* 0.392 0.439* (Continued ) 386 THE WORLD BANK ECONOMIC REVIEW TABLE A2. Continued Dependent variable: Levelsa Levelsa Logsa Logsa Levelsb Logsb ln (Aggregate exports) (1) (2) (3) (4) (5) (6) [0.177] [0.278] [0.212] [0.323] [0.264] [0.247] POP 2 0.251 2 0.907* 2 0.044 2 0.587 2 0.757 2 0.388 [0.328] [0.489] [0.361] [0.545] [0.495] [0.425] Distance 2 0.934*** 2 1.396*** 2 0.994*** 2 1.420*** 2 1.439*** 2 1.457*** [0.281] [0.362] [0.285] [0.354] [0.358] [0.348] Tariffs (simple av.) 2 0.062** 2 0.063** 2 0.070** 2 0.069** [0.030] [0.029] [0.029] [0.028] Observations 991 479 991 479 479 479 R-squared 0.546 0.520 0.564 0.540 0.531 0.554 p-value of Sargan 0.859 0.368 0.977 0.188 0.820 0.864 statistic Notes: 1. Robust standard errors in brackets. ***p , 0.01, **p , 0.05, *p , 0.1. 2. Other control variables: partner FE, common language, common border, colony, remoteness, land- Downloaded from wber.oxfordjournals.org by guest on October 18, 2011 locked aid for trade and GSP (whenever import tariffs are included in the regressions). a. Instruments: documents, customs, ports and inland transit times in transit countries. b. Instruments: documents customs and ports times and GPS distance from border between the landlocked and the transit country and the port of the latter. Source: Authors’ analysis based on data sources discussed in the text.