Port efficiency and Brazilian exports: A quantitative ... · ORIGINAL ARTICLE Port efficiency and Brazilian exports: A quantitative assessment of the impact of turnaround time Vinıcios

OR I G I N A L AR T I C L E

Port efficiency and Brazilian exports: Aquantitative assessment of the impact ofturnaround time

Vin�ıcios P. Sant’ Anna1 | S�ergio Kannebley J�unior21Department of Economics, University of Illinois at Urbana-Champaign, Urbana, IL, USA2Department of Economics, Universidade de S~ao Paulo, Faculdade de Economia Administrac�~ao e Contabilidade deRibeirão Preto, Brazil

Funding informationResearch funded by Fundac�~ao de Amparo �a Pesquisa do Estado de S~ao Paulo, Grant/Award Number: (2013/22297-5)

1 | INTRODUCTION

Considering the significant growth in the volume of traded goods and services among countries, inaddition to the reductions in traditional barriers to trade, such as tariff rates, trade facilitationaspects as determinants of the costs associated with international trade have grown in importance.While the traditional barriers have dwindled, other trade obstacles, such as transportation costs,administrative hindrances and the efficiency of port procedures, have grown rapidly in importanceas far as the final cost of traded goods is concerned. As a result, trade facilitation has been high-lighted both in the WTO’s1 round of negotiations and in the planning guidelines for policies aimedat promoting international trade.

Among the various elements related to a country’s trade facilitation, port infrastructure, as wellas the efficiency of customs procedures, is among the most important for determining the final costof an exported good. In fact, ports are the main gateway to global trade. According to the Interna-tional Chamber of Shipping, ~90% of world trade is conducted by the international shipping indus-try. This importance is also highlighted in the academic literature. For instance, according toClark, Dollar, and Micco (2004), the activities required at the port level are sometimes crucial forinternational trade transactions. The same pattern is observed for Brazilian exports. According to

1At the WTO, the discussions about trade facilitation are relatively recent. According to WTO (2003), trade facilitationbecame an official discussion topic in the Singapore Ministerial Conference, in December 1996. On that occasion, analyticaland exploratory studies on the simplification of trade procedures were recommended to assess the role of the WTO in thenegotiation of agreements in the area. Afterwards, the topic was included in the discussion agenda at the Doha Round inNovember 2001, when the ministers agreed to hold the negotiations on trade facilitation after the Cancun Ministerial Con-ference in September 2003. In August 2004, the WTO’s General Council unanimously decided to commence the negotia-tions based on modalities agreed upon its members. These modalities established the basis for the work plan adopted in thefirst meeting of the Negotiating Group in November 2004. Over recent years, detailed negotiations have been conductedquite often, and the negotiating text was improved and simplified until it was finally approved unanimously during the Min-isterial Conference held in Bali in 2013.

DOI: 10.1111/twec.12654

World Econ. 2018;1–24. wileyonlinelibrary.com/journal/twec © 2018 John Wiley & Sons Ltd | 1

http://orcid.org/0000-0002-1070-2220

http://orcid.org/0000-0002-1070-2220

http://orcid.org/0000-0002-1070-2220

http://wileyonlinelibrary.com/journal/TWEC

data from the Brazilian Ministry of Development, Industry and Foreign Trade, between 2004 and2013, ~82.6% of the total value and 95.8% of the total weight of Brazilian exports were conductedthrough ports.

Despite this relatively great importance of Brazilian ports in trade flows, both port infrastructureand organisation are very far from the quality and efficiency observed in countries with high levelsof trade facilitation. According to the World Economic Forum indicator “Quality of Port Infrastruc-ture,”2 in 2013 Brazil ranked 130th among 148 countries in port quality and efficiency. Further-more, according to data from the National Agency for Waterway Transportation, in 2012 56.2% ofthe total time required for a ship to complete an average port procedure was spent on either wait-ing to dock or waiting to initiate the operations. These numbers suggest that Brazil suffers fromhigh inefficiency in port procedures. Therefore, considering that the time to complete port proce-dures is directly related to port efficiency, the purpose of this paper is to estimate the impacts ofturnaround time on both the volume of Brazilian exports and on the number of categories ofexported products. The decision to use Brazil for this study is justified by its geographic character-istics. In fact, Brazil is a country of continental dimensions and has a large number of ports withsignificant trade volumes. This ensures a data variability in terms of products and trade partnersthat cannot be found in many other countries.

We use a unique database with exports from Brazilian cities. Products are aggregated at the four-digit Harmonised System (HS) from the period between 2010 and 2012 to estimate the impacts of turn-around time in ports on Brazilian exports, assessing which stages of port procedures contribute most tothese impacts. We use information on turnaround time from the Brazilian National Agency for Water-way Transportation (ANTAQ). These variables are measured in hours and describe the mean timerequired for import or export procedures for standard container ships, from docking to undocking. Asfar as we know, there are no previous studies that analyse the particular effects of turnaround time oninternational trade. Therefore, this study contributes to the literature by presenting new empirical evi-dence regarding the implications of turnaround time for trade flows.

The database constructed for this study has several advantages over conventional data in the lit-erature. Most of these studies are performed with cross-sectional data. We use unique panel datawith detailed information about exporter cities, products, ports and destination countries that allowstime invariant components to be controlled by the fixed effect estimator.

By taking advantage of this highly detailed database, we adopt an innovative identificationstrategy inspired by Djankov, Freund, and Pham (2010) that enables us to control for several unob-servable factors that can affect export flows at the local level. The idea of this approach is to esti-mate a gravity in difference model where we use geographically close cities to build restrictedratios such that common trade facilitation components for a given group of cities, product and des-tination country can be controlled in the estimated equation. This approach may also reduce prob-lems arising from the possible endogeneity between time measures and exported volume.

The results indicate that each additional hour of delay in port procedures represents costs toBrazilian exporters, which may cause loss of competitiveness of domestic products abroad. Onaverage, the turnaround time in the sampled Brazilian ports averages 28 hours and 12 minutes.According to the estimates, each relative additional hour of delay in the average port is equivalentto a reduction of nearly 2% in relative local exports. Furthermore, a 10% reduction in relative turn-around time can increase the relative number of exported product categories by ~1%. Moreover,vessel downtime as a percentage of total port time averaged 45.17% in 2012. The estimated

2This indicator measures the country’s business people’s overall perception of their port facilities. This information is avail-able from the “Global Competitiveness Report” and the World Bank database.

2 | SANT’ ANNA AND KANNEBLEY J�UNIOR

coefficients indicate that a 10% reduction in relative downtime can lead to a 2.66% increase in rel-ative local exports.

This paper can be related to a growing empirical literature in trade facilitation that is specialisedin the estimation of the impact of time on trade flows. The idea of time as a trade barrier is rela-tively recent in the academic literature. Hummels and Schaur (2013) is the first to highlight theimportance of time to trade as a barrier to exports and imports, identifying the costs related to slowshipping time, compared with air transport, and its implications for trade patterns and for the glo-bal organisation of production. Using US trade data, he shows that each day of delay in the trans-portation of goods between two countries reduces the probability of trade by 1.5%.

Nord�as (2007) contributes to the literature by providing a wider notion of the impacts of time ontrade flows. According to the author, in addition to transportation, the time spent to perform all theadministrative requirements, inspections, purchase of licences and transportation of goods from the fac-tory gate to the port also has an important impact on trade flows. Using export panel data from Aus-tralia, Japan and the United Kingdom, she finds that trade volume can increase by 8%–40% in responseto a 10% reduction in trading time. In the analysis with cross-sectional data for the year 2004, the same10% reduction in export time would generate an increase in trade between 8% and 27%.

Djankov et al. (2010) also adopts this wider notion of time as a trade barrier. However, he con-tributes to the literature using a methodology for reducing endogeneity problems and obtaining morereliable coefficients. Based on the traditional gravity equation from Melitz (2003) and Anderson andVan Wincoop (2003), the authors use a difference gravity equation to minimise endogeneity prob-lems, as the effects of trade volume on time are probably similar between countries belonging to thesame geographic region. According to his findings, each additional day a product is delayed prior tobeing shipped reduces trade volume by more than 1%, which is equivalent to an increase of 70 kmbetween a country and its trade partners. The estimated coefficients are even greater when he analysesthe exports of time-sensitive products, such as perishable agricultural products.

Therefore, the literature indicates that time is an important determinant of trade flows. Most ofthe studies aim to estimate the effect of transportation time or total time to complete all trade pro-cedures on trade flows. The literature presents evidence that delays in the transportation of a pro-duct from the port to the importing country or in the completion of the whole export/importprocedure diminishes the volume of bilateral trade (intensive margin) and reduces the number offirms and production sectors that participate in international trade (extensive margin). This paper isin line with this literature, and it shows that turnaround time as a specific trade facilitation factor isan important determinant of local trade flows both at the extensive and intensive margins of trade.

The remainder of the paper is organised as follows: Section 2 describes the data and developsthe empirical strategy used in this study. Section 3 presents the descriptive analysis, the empiricalresults, the sensitivity tests and the final remarks.

2 | MATERIALS AND METHODS

2.1 | Data

We use an innovative database built for this study of local exports, taking into consideration theport used, the country of destination and products aggregated at the four-digit HS for the periodbetween 2010 and 2012. This period was chosen due the availability of the data.

Brazilian export data were obtained from the Alice Web System, developed by the NationalForeign Trade Department (SECEX), from the Ministry of Development, Industry and ForeignTrade (MDIC) in Brazil. The information was collected from 2010 to 2012 according to product

SANT’ ANNA AND KANNEBLEY J�UNIOR | 3

category (at the four-digit HS)3 and to the cities of origin of the goods.4 In total, export data wereobtained from 1,876 Brazilian cities, including 204 importing countries or regions and 16 Brazilianports. However, after applying the empirical strategy and imposing all the restrictions discussed inSection 2.2, we end up with a sample that includes 503 cities and 131 countries. Table 1 presentsthe descriptive statistics of the cities of the sample. Note that the information used comprises onlycities that exported products aggregated at the four-digit HS.

Is this paper, we selected data from 16 Brazilian ports: Bel�em—PA, Fortaleza—CE, Imbituba—SC, Itaja�ı—SC, Itaqui—MA, Paranagu�a—PR, Rio de Janeiro—RJ, Rio Grande—RS, Salvador—BA, Santar�em—PA, Santos—SP, S~ao Francisco do Sul—SC, Itagua�ı (Sepetiba)—RJ, Suape—PE, Vila do Conde—PA and Vit�oria—ES. Figure 1 presents the geographic locations of theseports. The selected ports account for most of the international trade in Brazil. According to datafrom the Alice Web system, between 2010 and 2012, these ports were responsible for ~77.6% oftotal Brazilian exports, and ~92.7% of exports by sea.

The gross domestic product data for Brazilian cities were obtained from the Brazilian Instituteof Geography and Statistics (IBGE). Information about the average income of workers from aux-iliary activities in the waterway transport sector in the cities where ports operate was collectedfrom Brazil’s comprehensive RAIS (Annual Report on Social Information) labour force records.

The distances between cities were calculated based on the latitude and longitude of each loca-tion. The data on the latitude and longitude of cities and ports were computed using the R software,with the geocode routine of the ggmap package.5 For each pair of locations, the geographic distancewas calculated in kilometres, based on the great-circle distances using the spherical law of cosines.6

We gathered information on turnaround time from the Brazilian National Agency for WaterwayTransportation (ANTAQ). These variables are measured in hours and present the average time nec-essary for a port to handle a complete import or export procedure for a standard container ship,from docking to undocking. Figure 2 illustrates the information collected.

The total port time in Figure 2 refers to the total time required to complete a trade procedure atthe port, from docking to undocking. This total port time can be broken down into four steps:waiting to dock, waiting to start operations, operation time and undocking time. Docked timestands for the average number of hours a ship has been docked at the port and includes the timespent waiting to start operations, operation time and undocking time. Downtime can be defined asthe sum of hours a ship spends waiting in the port.

3According to the Alice Web System, a port is defined as the last location in a country where goods are dispatched to thecountries of destination.4The Alice Web database provides information on exports by city of origin, which is defined as the city of the headquarters(fiscal domicile) of a company that carried out export operations. The database also provides port information that is deter-mined by the port or locality where the actual shipment of the merchandise took place, that is, the last authorised place ofthe Brazilian territory from where the merchandise exited the country. Therefore, we linked the city of exports to the port ofexports according to the information provided by this dataset.5Kahle, Wickham, and Kahle (2013).6To measure the distance more accurately, we applied the spherical law of cosines to calculate the great-circle distancebetween two points in the following manner:

d ¼ R arcsen sen i1ð Þsen i2ð Þ þ cos i1ð Þ cos i2ð Þ cos k1 � k2ð Þ½ �; (1)

where ι1 and ι2 are the latitudes and k1 and k2 are the longitudes of two points in a sphere with radius R. To calculate thedistance, we used = 6,371 km, which is accepted as the mean radius of the Earth.


2.2 | Analytical model and empirical strategy

Gravity models have been widely used in the literature to estimate aggregate effects of tariff andnon-tariff measures and costs reductions associated with trade facilitation. Among these studies,we can mention Wilson, Mann, and Otsuki (2005), Helble, Shepherd, and Wilson (2007) andShepherd and Wilson (2009). The standard gravity model used in the literature can be representedby the following equation:

Xji ¼ YiYjY

uji

QiQj

� �1�r

; (2)

where Xji, represents the exports from region j to region i; Yi is the importing country’s incomeand represents the demand for goods from the importing country; Yj is the exporting capacity ofcountry j; Y represents the world production; uji is a factor that represents all trade costs between i

TABLE 1 Descriptive statistics of cities in the final sample

Year Exports US$ million Lagged export coefficient Population

2010

Observations 480 480 451

Mean 159 0.0467424 163,144.7

Standard deviation 471 0.0933541 646,081.2

Minimum 0.0237 0 2,149

Maximum 4.79E+03 1.493204 1.10E+07

2011


Mean 206 0.0485627 157051.4

Standard deviation 619 0.0821695 632,639

Minimum 0.25E�05 0 2,495

Maximum 7.34E+03 0.9619763 1.13E+07

2012


Mean 192 0.0462561 157,616.4


Minimum 0.5E+02 0 2,446

Maximum 7.41E+03 0.5084792 1.14E+07

Total

Observations 1,461 1,461 1,432

Mean 186 0.0471895 159,164.2


Minimum 0.25E�05 0 2,149

Maximum 7.41E+03 1.493204 1.14E+07

Source: Data compiled by the authors.


and j; and r is the elasticity of substitution between products. Indexes Qi and Qj represent the so-called multilateral trade resistance. These indexes show that bilateral trade flows depend on boththe existing trade costs between two countries and the trade costs associated with other tradepartners.

This standard specification of the gravity model considers data at the national level. Consider-ing the database used for the analysis and the objective of estimating the impacts of turnaround

Belém

Fortaleza

ImbitubaItajaí

Itaqui

Rio Grande

Rio de Janeiro

Salvador

Santarém

SantosSepetiba

São Francisco do Sul

Suape

Vila do Conde

Vitória

Paranaguá

FIGURE 1 Ports assessed—geographic locations

Arrivalin port Docking

Start ofoperations

End ofoperations

End ofprocedure

Waitingto

dock

Waitingto start

operations

Downtime

Operationtime

Undockingtime

Dockedtime

Totalturnaround time

FIGURE 2 Turnaround time


time in ports on Brazilian exports, the model can be directly adapted to disaggregated data forexporting cities, for the sector of goods and for panel data:

Xkmit ¼

YkitY

kmt

Ykt

ukmit

Qki Qk

m

� �1�rk

; (3)

In this case, Xkmit represents the exports from city m to country i in product sector k in year t;

Ykit is country i’s production in sector k in year t; and Yk

mt is the exporting capacity of city m insector k in year t. Yk

t is the global production of industry k. ukmi is a factor that represents all trade

costs between city m and country i for a particular category of products k in year t. rk is the elas-ticity of substitution between products from the same category k. Indexes Qk

i and Qkm represent

multilateral resistance. Here, we assume that the multilateral resistance terms are fixed over time,7

and by applying the logarithm to this model,8 the log-linear gravity equation is obtained:

lnðXkmitÞ ¼ ln Yk

it

� �þ ln Ykmt

� �� ln Ykt

� �þ 1� rkð Þln ukmit

� �� 1� rkð Þln Qk

i

� �� 1� rkð Þln Qkm

� �:

(4)

From this equation, considering the purpose of the study, it is necessary to establish a specifica-tion for the cost termuk

mit affecting the trade flow of Brazilian cities and to consider all factorsaffecting the export procedure, including turnaround time. To satisfy this need, a decomposition ofthe cost term uk

mit was performed. Therefore, the disaggregated feature of the data set was essentialto allow the decomposition. Export data on Brazilian cities take into consideration the port usedfor the transport of traded goods. Note that trade costs can be represented by the new term uk

mpit

where p denotes the port used for export, and by the following decomposition into three compo-nents that incorporate all trade facilitation elements:

ln ukmpit

� �¼ ln lkmpt

� �þ ln xk

pt

� �þ ln skpit

� �; (5)

where lkmpt represents the set of variables that denote any cost factor between exporting city m and portp, such as the distance between the city and the port, the quality and availability of transport infrastruc-ture (roads and railways, for example), and comparative advantages in the production of some goods.xkpt is a set of variables that reflect the characteristics of the port used for export, such as the quality and

efficiency of customs procedures, including the time required to perform each step of the procedure.Finally, skpit is the set of variables that represent the costs during the trading process between port p and

µkm,p,t ωp,t ψk

p,i,t

City m Country iPort p

FIGURE 3 Decomposition of trade costs

7This is a common assumption in the gravity equation literature, and it allows us to control for these terms by using fixedeffects for the destination country and the city of origin. Anderson (2011) states that fixed effects estimation still seems thebetter and more cautious practice to follow in order to control for multilateral resistance. Therefore, we omit the time indexof these variables.8The specification log-log is widely used in the literature related to the gravity model, as in the works of Wilson et al.(2005), Helble et al. (2007) and Shepherd and Wilson (2009).


country of destination i in year t, such as the distance between the port and the country and other char-acteristics of the importing country. Figure 3 illustrates this decomposition.

Following Khan and Kalirajan’s (2011) definitions, the variables represented by lkmptconsist primarily of elements from “behind-the-border” trade facilitation, while skpit refers to “be-yond-the-border” elements. However, xk

pt denotes specifically all the aspects of the “frontier”because it captures trade facilitation characteristics that are inherent to ports. In this case, theempirical specification takes the following form:

lnðXkmpitÞ ¼ ln Yk

it

� �þ ln Ykmt

� �� ln Ykt

� �þ 1� rkð Þln lkmpt

� �þ 1� rkð Þln xk

pt

� �þ 1� rkð Þln skpit

� �� 1� rkð Þln Qk

i

� �� 1� rkð Þln Qkm

� �:

(6)

However, some of these terms, especially those related to “behind-the-border,” lkmpt, and “be-yond-the-border,” skpit, trade facilitation are unobservable or difficult to measure accurately for dif-ferent cities and countries in the sample. The omission of these factors could lead to biasedestimates of the coefficients of turnaround time. Given these potential problems, we elaborate anidentification strategy based on the development of assumptions about the trade facilitation ele-ments and on the construction of a difference gravity model.

Considering the “behind-the-border” facilitation elements, lkmpt, we use a similar approach toDjankov et al. (2010), in which a difference gravity equation is estimated where the variables arecalculated as the ratio of two observations regarding the exports of two cities located in the samegeographic region. The aim of this strategy is to eliminate the need to control for regional factorsthat could affect the volume of exports of a given city.

In this case, the difference in the gravity model is sustained between observations regardinggeographically close cities. The construction of these groups should be conducted so that the “be-hind-the-border” facilitation features observed for cities of the same group are similar. For the con-struction of these groups of cities, we used hierarchical cluster analysis, with geographic distanceas the dissimilarity variable between cities.9

By constructing ratios with observations of sufficiently close cities, so that “behind-the-border” tradefacilitation characteristics may be considered identical, we eliminate the need to control for these factorsthat could affect the exports of cities. Therefore, the first identifying assumption of the model is thatcities belonging to the same cluster share the same characteristics in terms of regional trade facilitation,such as business environment, general infrastructure and any other local factor.

However, this strategy alone is not sufficient to eliminate all unobservable components capturedby the term lkmpt. Therefore, this strategy should be complemented by a second identifying assump-tion, according to which, given two cities belonging to the same cluster, the components of lkmptthat differ for two distinct ports are invariant in the short term.10

We control for city-specific factors by clustering them, adding the assumption that aspectsbetween the port and cities of the same cluster do not vary from year to year. Therefore, with the

9The cluster analysis is a multivariate statistical procedure whose primary purpose is to look for patterns in the data to identify rel-atively homogeneous subgroups between observations called clusters. In this case, we used the hierarchical cluster, with the aver-age linkage clustering method. We also considered a limit of 40 kilometres as the criterion for the realisation of the cluster. Hence,the cluster analysis was used to group Brazilian municipalities according to geographical distance with a limit of 40 km (24.8miles) between at least one of them. Using this methodology, we were able to split the cities into 143 groups.10Here it is assumed that from one year to the next, “behind-the-border” trade facilitation involving the exporting city andthe port used does not change. This assumption is reasonable for the regional infrastructure. For instance, the constructionof a new railway linking a particular region and a specific port usually takes several years to complete.


estimation of a fixed-effects model, the term “behind-the-border” trade facilitation, lkmpt, can beeliminated from the equation.

To control for the elements of “beyond-the-border” trade facilitation, we made an additional restric-tion on the ratios. In fact, we calculated the ratios only with observations whose country of destinationis the same. This approach is based on the traditional gravity equation in Melitz (2003) and Andersonand Van Wincoop (2003) and used by Djankov et al. (2010). It eliminates the need to include “beyond-the-border” trade facilitation terms, skpi. Therefore, by incorporating the assumptions about the “behind-the-border” and “beyond-the-border” facilitation elements, by constructing ratios of local observationsfor the same region, and using the fixed-effects model, we obtain the following equation:

lnXkm1p1it

Xkm2p2it

!¼ b1ln

Ykm1t

Ykm2t

!þ b2ln

xkp1t

xkp2t

!; (7)

where Xkm1p1it are the exports from city m1, product k, using port p1, to country i, in year t, and

Xkm2p2it are the exports from city m2 belonging to the same cluster of m1, of product k, that were

shipped through port p2, to country i in year t. The other variables are defined as before.11

We also opted for the creation of ratios between cities that export the same product but thathave exported through different ports. This option allows the control of intrinsic characteristics ofeach category of products that are not observed, such as elasticity of substitution, eliminating omit-ted-variable bias.12

The strategy of building groups of local observations, although it may limit the examination of theeffects of turnaround time only for cities that already export, has several advantages from the economet-ric point of view. In addition to eliminating non-observable components, the approach allows the reduc-tion of endogeneity problems between the characteristics of trade facilitation and Brazilian exports.Under the assumption that investments in “behind-the-border” trade facilitation occur on a regionalscale, so that geographically close cities can equally benefit from it, the use of ratios calculated for localobservations belonging to the same cluster eliminates the endogeneity between the volume of local cityexports and “behind-the-border” trade. Additionally, because the ratio is calculated with data on portsand cities of the same country, in this case Brazil, any endogenous channel between time costs and vol-ume of trade that may affect the results on a national scale is also eliminated.

From this model, the econometric specification adopted to estimate the impact of turnaroundtime on Brazilian exports involves the variables described in Section 2.1. Usually, empirical stud-ies use GDP as a measure of income of the countries involved in bilateral trade. However, in thecase of a disaggregated model for local data, GDP may not be the best variable for capturing theexport capacity of a city.13 To better capture this term, we used the export coefficient of cities,

11The exclusion of the multilateral resistance terms is implied by the use of our empirical strategy. First, by transformingthe model into ratios with the same importer country, we eliminate the need to control for multilateral resistance on theimporter side. Second, by using the fixed effects estimator, we also eliminate the need to control for the multilateral resis-tance ratio for the city of origin. An alternative way to deal with the multilateral resistance would be to assume that allBrazilian cities face the same multilateral resistance with respect to all the destinations. In that case, the ratios strategytogether with the inclusion of year dummies would be sufficient to control for the multilateral resistance.12Ratios were calculated only between observations referring to different cities (belonging to the same cluster) and differentports, but to the same importing country and related to the same product category.13This can be explained by the fact that cities usually have a much more limited range of production sectors than a country.In addition, growth in local GDP will be associated with greater trade flows only if the growth occurs in the export sector.If the growth of local income is strongly linked to the service sector, for example, the local income may not affect localexports.


calculated as the ratio of the total amount exported by each city in a given sector and the localGDP in a given year. Moreover, the export coefficient captures the ability of the city to export.Hence, it might also control for other non-observed city-specific characteristics that can affect thevolume of trade and the variety of its exports.

Finally, the “border” trade facilitation term should include the characteristics of Brazilian ports.Therefore, we include the number of dockings, the cost of port usage and turnaround time vari-ables in the equation.14 The empirical equation can be written as follows:

lnXkm1p1it

Xkm2p2it

!¼ akm1m2p1p2i þ at þ b1ln

Coef � Expkm1t

Coef � Expkm2t

!þ b2ln

Costp1tCostp2t

� �

þ b3lnTimep1tTimep2t

� �þ b4ln

Nr � Dockingp1tNr � Dockingp2t

� �þ ekm1m2p1p2i

; (8)

akm1m2p1p2i is the fixed-effects term for the pair of cities, pair of ports used, country of destinationand product category, while at represents a set of year dummy variables used to capture the intrin-sic effects of each year. Coef � Expkmt is the export coefficient of the city for product sector k attime t. Costpt is the cost of port p usage in year t. Timept refers to turnaround time variables inport p and year t. Nr � Dockingpt is the number of dockings for port p and year t, and ekm1m2p1p2i isa random error term. The estimates were based on four basic specifications, in which differentmeasures of time were inserted into the equation to better estimate the impact of the various stepsof port procedures.

According to Nord�as (2007), the time to transport goods, or to complete the trade procedure,may have effects on exports at the extensive and intensive margins of trade. Therefore, time repre-sents costs to exporters, which can affect both the volume of exported products (intensive margin)and the number of sectors and firms participating in international trade (extensive margin). In thissense, two measures of exports were used as dependent variables in the equations presented above.

The first measure was the volume exported in dollars. Equations with this dependent variable yieldestimates of the impact of turnaround time on the intensive margin of trade. In this case, we estimatethe effect of relative reductions in turnaround time on the intensity of relative trade volumes.

For the extensive margin of trade, we adopted a similar approach to that of Moreira, Blyde,Volpe, and Molina (2013), where the number of exported products is used as the dependent vari-able instead of the exported value. In this sense, the second measure used as the dependent vari-able was the number of categories of exported products. At this stage, the goal is to analyse towhat extent turnaround time can affect the diversification of exports. This variable was measuredas the total number of different product categories aggregated by the Common Nomenclature ofMERCOSUR—NCM at the 8-digit level.15 In this case, the ratios were calculated as the number

14We included in the equation the average income of workers from the auxiliary activities of the waterway transport sectorin the cities where the ports operate. This variable was used to control for the cost of port usage. In fact, the price a portcharges from the exporter can be an important determinant of which port will be used for the export procedure. However,this type of information is dispersed and difficult to obtain with precision for all the sampled ports. Thus, we used the aver-age income as a rough measure of this cost. Implicitly, we are assuming that the ports that have higher port charges arelocated in cities whose wages paid to employees in water transport auxiliary activities are also higher. The number of dock-ings is included as a measure of port size and allows for the control of endogenous channels between port size and effi-ciency.15Although we do not estimate the effect of time on the likelihood of new companies and production sectors entering theinternational market as exporters, the results demonstrate the effects of time on the diversification of exports. In this sense,the results can be interpreted as evidence of the effects of turnaround time in ports on the extensive margin of trade.


of product categories using the NCM at the 8-digit level in a given 2-digit HS for two cities in thesame cluster, for the same country and using two different ports.

3 | RESULTS

3.1 | Turnaround time

To better delineate the profile of each port of the sample, we present their relative importance to Brazil-ian trade. Figure 4 presents the total volume of exports of each port during the period 2010–12.

According to the figure, Santos is the Brazilian port with the highest export volume (approxi-mately US$64 billion in 2012). All the other ports can be divided into three groups. Vit�oria, Sepe-tiba, Paranagu�a, Itaqui and Rio Grande can be considered ports with a high volume of exports,ranging from 10 to 30 billion dollars. Itaja�ı, Rio de Janeiro, Vila do Conde, Salvador and S~aoFrancisco do Sul can be considered ports with a moderate volume of exports, ranging from 3 to 9billion dollars. Finally, Suape, Santar�em, Fortaleza, Bel�em and Imbituba are ports with a low vol-ume of exports, with less than 1 billion dollars. Despite this low volume compared with the othersampled ports, these ports are considered in the analysis due to their regional importance or to theproducts they usually export.

Considering turnaround time, it is possible to characterise each port of the sample according toits trade efficiency. The time each port takes to complete each stage of the trade process mayreflect its trade facilitation characteristics, such as the quality of the port’s infrastructure or portbureaucracy and other features that may affect the time needed to complete a standard procedure.

The first unit of information relevant to characterising ports is turnaround time (total port time),which consists of the time spent from the moment the ship arrives at the port to its undocking.Figure 5 presents the turnaround time for each port throughout the sampled years.16

According to the information from Figure 5, Imbituba and Rio de Janeiro may be consideredthe most agile ports, or those with the lowest turnaround time. On the other hand, S~ao Franciscodo Sul stands out as the most time-consuming port, where the entire port procedure can take morethan 70 hr. It is also worth noting that during the analysed period, only Itaqui, Paranagu�a, Suape,Itaja�ı and Imbituba reduced their total turnaround time. This result reveals a latent need for invest-ment in Brazilian ports, which, year after year, have increased the time necessary to carry outexport and import procedures.

Table 2 presents additional information related to the time variable in the sampled ports.According to the information, the sampled Brazilian ports present an average turnaround time of28 hr and 12 min.17 From this total, an average of 15 hr and 42 min is spent with the shipdocked, and the rest of the time is wasted in waiting steps.

Table 2 also provides information on the variation in turnaround time. While there are portswith no downtime, there are other ports in which the maximum waiting time is 60 hr and 12 min

16Three adjustments were made in the time database for the presence of outliers. By verifying the monthly data, ports inSalvador in April 2011, Paranagu�a in March 2011 and Fortaleza in August 2012 presented a turnaround time that farexceeded their monthly averages, affecting the annual value of this measure. Thus, the monthly observations affected by themonthly average of each port were replaced with the calculation of the annual time measure of these ports.17Ducruet, Itoh, and Merk (2014) shows that the average turnaround time (ATT) in 2011 for countries in Latin Americawas ~1.315 days, which is equivalent to 31.56 hr. Therefore, considering this information, Brazilian ports are more efficientby a few hours compared to the average for the whole of Latin America. Also, according to Ducruet et al. (2014) descrip-tive statistics of average turnaround time, the most efficient European countries in 2011 had an average turnaround of1.078 days or 25.872 hr. For the African countries with the highest ATT, it was ~2.535 days or 60.84 hr.


for docking or starting operations. Regarding operation time, the results also vary greatly acrossports. While operations in the most agile ports take 4 hours and 24 minutes, they take around44 hr in the slowest ones.

For a better understanding of turnaround time, Figure 6 presents information on each stage ofthe trading process over the analysed period. Each bar graph represents a year of the sample from2010 to 2012, and each stage of the procedure is added equally to the turnaround time.

Each stage varies greatly across the sampled ports and over the sampled years. In fact, the oper-ation time captures the efficiency in loading and unloading activities in each port. Santar�em and

0

10

20

30

40

50

60

70

80

Tim

e in

hou

rs

2010 2011 2012

Santos

Vitória

Sepeti

ba

Parana

guá

Itaqu

i

Rio Gran

deIta

jaí

Rio de

Jane

iro

Vila do

Con

de

Salvad

or

S. F. d

o Sul

Suape

Santar

ém

Fortal

eza

Belém

Imbit

uba

FIGURE 5 Brazilian ports—turnaround timeSource: ANTAQ (2015).

0

10

20

30

40

50

60

70

Tota

l exp

orts

(U

S$ B

illio

ns)

2010 2011 2012

SANTOS - SP

VITÓRIA

- ES

SEPETIBA -

RJ

PARANAGUÁ -

PR

ITAQUI -

MA

RIO G

RANDE - RS

ITAJA

Í - S

C

RIO D

E JANEIR

O - RJ

VILA D

O CONDE -

PA

SALVADOR -

BA

SÃO FRANCIS

CO DO S

UL - SC

SUAPE - PE

SANTARÉM

- PA

FORTALEZA -

CE

BELÉM -

PA

IMBIT

UBA - SC

FIGURE 4 Brazilian ports—total exported in US$ billions from 2010 to 2012Source: Alice Web (2015).


Itaqui are examples of ports whose operation times higher than those in the other sampled ports.On the other hand, Rio Grande and Imbituba are ports with relatively more agile operations.According to Figure 6, the undocking time represents the smaller portion of the total turnaroundtime. Santar�em had the greatest average undocking time in 2012. Nevertheless, the sampled portstake around 2 hr for undocking, on average.

According to the information provided, in most of the ports, waiting to dock is the stage thatcontributes the most to total turnaround time. S~ao Francisco do Sul, for example, has the highestwaiting time for docking, and that has considerably increased throughout the sampled years.

TABLE 2 Time measures

Variable Obs Mean Median SD Min Max

Port time 48 28.2 25.5 13.6 10.3 70.3

Docked time 48 15.7 12.7 9.7 5.7 48.9

Downtime 48 13.6 11.8 11.8 0 60.2

Waiting to dock 48 12.5 11.4 11.9 0 57.6

Waiting to start 48 1.1 0.8 1.6 0 10.6

Operations

Operation time 48 12.5 9.6 8.2 4.4 44.1

Undocking time 48 2.1 1.3 3.5 0 24.4

Source: ANTAQ (2015).

0

10

20

30

40

50

60

70

80

Tim

e in

hou

rs

Time to undock Operation time

Waiting time to initiate the operation Waiting time to dock

Santos

Vitória

Sepeti

ba

Parana

guá

Itaqu

i

Rio Gran

deIta

jaí

Rio de

Jane

iro

Vila do

Con

de

Salvad

or

S. F. d

o Sul

Suape

Santar

ém

Fortal

eza

Belém

Imbit

uba

FIGURE 6 Brazilian ports—average time for each step of port procedures


Salvador, Fortaleza and Vit�oria also showed an increase in the time needed for docking. Otherports, such as Itaqui, Paranagu�a, Itaja�ı and Suape, reduced the time necessary for docking, but thisstage still represents a relevant part of port procedures. Ports such as Santar�em, Vila do Conde andBel�em stand out, as incoming ships do not have to wait to dock there. However, the waiting timefor initiating operations is rather long in these ports. Indeed, the sum of both waiting times iscalled downtime and represents the time during which the ship is prepared to load and unload, butis kept waiting, either to dock or to initiate operations. Table 3 demonstrates the participation ofwaiting time in the total turnaround time in each port over the sampled years.

According to Table 3, the least ineffective ports are those in Rio de Janeiro, Bel�em and Vila doConde, in which the waiting time is below 10% of the total turnaround time. On the other hand, ofthe 16 sampled ports, 11 present a waiting time longer than 40% of the total turnaround time.Among them, Fortaleza, Vit�oria, Itaja�ı, Santos, Rio Grande, Paranagu�a, Salvador and S~ao Fran-cisco do Sul show percentage values greater than 50%. In these ports, ships are inoperative duringmost of the waiting time. In many cases, the waiting time has increased over the sampled years. Infact, the average waiting time has increased for the sampled ports. These results indicate the pres-ence of a bottleneck in the Brazilian production flow in the international market. Because of thegrowth in Brazilian trade flows over the last decade, these limitations have become even more evi-dent, thus highlighting a need for investments to increase port capacity and reduce the inefficien-cies that generate these elevated rates of downtime in Brazilian ports.

With regard the size of the ports of the sample, we can use the number of dockings as a simplemeasure for port size. Figure 7 presents the annual number of dockings for the ports in our sam-ple. The port of Santos is the largest in our sample, with between 2,500 and 3,000 dockings peryear. The Santar�em, Itaqui, Imbituba, Bel�em, Vila do Conde and Fortaleza ports are classified as

TABLE 3 Waiting time as a percentage of total turnaround time

Port 2010 (%) 2011 (%) 2012 (%)

Rio de Janeiro 16.80 14.56 3.47

Bel�em 8.59 8.70 4.76

Vila do Conde 5.88 11.81 8.95

Santar�em 8.37 0.00 21.68

Itaqui 32.31 35.47 27.30

Itagua�ı (Sepetiba) 47.60 47.98 42.24

Imbituba 37.95 58.25 47.57

Suape 62.95 46.04 47.78

Fortaleza 26.56 37.55 52.45

Vit�oria 60.79 59.93 56.75

Itaja�ı 71.58 60.00 57.64

Santos 63.44 50.54 58.00

Rio Grande 72.65 75.11 62.64

Paranagu�a 66.67 56.90 72.35

Salvador 56.67 58.44 73.48

S~ao Francisco do Sul 59.52 80.91 85.63

Mean 43.65 43.89 45.17

Source: ANTAQ (2015).


small, with less than 200 dockings per year. The rest of the ports in the sample are more homoge-neous regarding the number of dockings (between 200 and 1,000 dockings per year). It is not sur-prising that the size of the port might be correlated with its efficiency. Larger ports have a greateroutflow of products, which might lead to more time delays in the port procedures. On the otherhand, large ports might also receive greater amounts of investment to increase capacity and effi-ciency. To consider this possible endogeneity channel, we conduct an exercise in Section 3.3where we restrict the sample for ports that have between 200 and 1,000 dockings per year.

3.2 | Estimated results

Table 4 presents the estimations of Equation (8) made by the fixed-effects model.18 In this case,the dependent variable is the ratio of exported value from two different cities located in the samegeographic region (cluster), having the same product category and the same importing country,and using two distinct ports. All coefficients of the time variables should be interpreted in relativeterms to the port of reference.

According to the estimated coefficients, the local export coefficient has a positive and statisti-cally significant effect on the exported volume in all estimated specifications. These results showthe importance of export capacity to determine Brazilian exports at the local level. These resultssupport the theory of the gravity model and are similar to the results found in the traditional grav-ity model literature.

The average number of dockings has a positive and statistically significant coefficient in all ofthe estimated equations. This result is intuitive and supports the idea that ports with a higher num-ber of dockings also have a higher volume of trade. On the other hand, the variable that measuresport costs presents negative and statistically significant coefficients in all equations. This result

0

500

1,000

1,500

2,000

2,500

3,000

2010 2011 2012

Santos

Vitória

Parana

guá

Itagu

ai (S

epeti

ba)

Itaqu

i

Rio Gran

deIta

jaí

Rio de

Jane

iro

Vila do

Con

de

Salvad

or

São F

rancis

co do

sul

Suape

Santar

ém

Fortal

eza

Belém

Imbit

uba

FIGURE 7 Annual number of dockings by port

18In addition to eliminating the need for the inclusion of many unobservable characteristics, the choice to use fixed-effectsmodels was confirmed by running Hausman statistics, which corroborate this option instead of the random-effects model.


indicates that port cost is an important determinant of trade volumes. Moreover, the results high-light the importance of the inclusion of these two variables in all equations to control for port char-acteristics that may affect trade volume.

The first specification shows the effect of total turnaround time on exports for Brazilian cities.In fact, assuming that a reduction in turnaround time affects only the exports of cities that use thatport, a 10% reduction in total turnaround time, or any stage of port procedures, would generate anincrease of ~5.14% in relative local exports. For a more detailed analysis, specifications (2) and (3)estimate the effects of turnaround time disaggregated by procedure steps.

In specification (2), the operation time and the remaining procedure time, which is the total turn-around time excluding operating time, were included in the equation. According to the estimatedcoefficients, by keeping the time of the other steps constant, a reduction in the relative operation

TABLE 4 Impacts of turnaround time on volume of exports in Brazilian cities

Dependent variable: ln(Ratio Exported Value FOB)

Fixed-effects models

(1) (2) (3) (4)

ln(Ratio Export Coefficient) t�1 0.170*** 0.174*** 0.171*** 0.176***

(0.054) (0.054) (0.054) (0.054)

ln(Ratio Number of Dockings) 1.170*** 1.167*** 0.757*** 0.904***

(0.209) (0.210) (0.199) (0.195)

ln(Ratio Port Cost) �1.098*** �1.202*** �1.229*** �1.335***

(0.233) (0.267) (0.241) (0.227)

ln(Ratio Turnaround time) �0.514***

(0.136)

ln(Ratio Operation Time) �0.050

(0.147)

ln(Ratio Waiting Time + Undock) �0.353***

(0.086)

ln(Ratio Waiting to Dock) �0.100***

(0.020)

ln(Ratio Docked Time) �0.029

(0.167)

ln(Ratio Waiting Time Rate) �0.266***

(0.070)

Constant �0.554*** �0.563*** �0.426*** �0.470***

(0.106) (0.109) (0.104) (0.101)

Observations 11,377 11,377 11,377 11,365

Year dummies Yes Yes Yes Yes

R2 .040 .040 .041 .038

Notes: Standard errors robust to heteroscedasticity are in parentheses.The coefficients with *** are significant at the 1% level; with ** are significant at the 5% level; and with * are significant at the10% level.Source: Data compiled by the authors.


time does not affect relative exports, while a 10% reduction in the remaining procedure time wouldgenerate a 3.53% increase in exports. These results suggest that differences in the operation step donot matter for the exported volume, while at least one of the other steps (waiting to dock, waiting toinitiate operations or to undock) presents a negative and statistically significant effect.

Specification (3) estimates the impact of the length of the procedure disaggregated by theimpact of waiting time to dock and docked time. According to the estimated coefficients, a 10%reduction in the waiting time to dock is reflected in a 1% increase in relative exports, while the dif-ferences across ports as to the time the ship remains docked would generate no relevant impact onlocal exports. This result corroborates the coefficients presented in specification (2) by highlightingthe importance of negative effects on trade volumes when the waiting time increases. In fact, mostof the docked time is determined by the operation step.

The last specification estimates the impact of downtime on the total turnaround time. Therefore,we created a waiting time measure as a percentage of the total turnaround time. The objective is toestimate the impact over trade from reductions in downtime as a percentage of total turnaroundtime. The estimated coefficients indicate that a 10% reduction in downtime is reflected in a 2.66%increase in exports.

By putting the results into context, the median number of hours of stay, as shown in Table 2, is25.5. Therefore, each additional hour of a ship’s stay in the average port is equivalent to a reduction of~2% (0.514 9 1/25.5) in local exports. More specifically, each hour of delay in time to dock in theaverage port, ceteris paribus, is equivalent to a reduction of 0.88% (0.1 9 1/11.4) in local exports.

In addition to the estimation of the impact of turnaround time on the volume of trade, we alsoevaluate the effects of time on the number of exported products. At this stage, the goal is to exam-ine how turnaround time can affect the diversification of exports in Brazilian cities. Table 5 pre-sents the estimation results of the proposed models.

According to this second group of estimations, the export coefficient has a positive and statisti-cally significant association with export diversification. This result highlights the idea of the grav-ity model theory, in which regions with higher export capability not only export a greater volumebut also have a greater variety of products exported. Like the results presented before, the numberof dockings shows a positive and statistically significant relation to the number of exported prod-ucts. However, the coefficients of the costs of port usage are not statistically significant. This indi-cates that although the costs may be important to determine the volume of trade, they do not seemto be relevant to explain the variety of trade.

Turnaround time has a negative and statistically significant relationship with the number ofexported products. A 10% reduction in total turnaround time increases the number of exportedproducts by 1%. Regarding the steps of port procedures, the results indicate that operation time isnot relevant when the effects on export diversification are analysed. According to specification (2)in Table 5, a 10% reduction in the waiting time to dock, to initiate operations or to undock couldincrease the number of exported products by 0.71%. The other two specifications present negativecoefficients for the time variables, but none of them is statistically significant.

In sum, the results suggest that turnaround time represents an important determinant of the vol-ume of exports and of export diversification. Moreover, when we estimate the time effect whiletaking into account procedure steps, we find that downtime or waiting time has a relevant impact.

3.3 | Sensitivity analysis

Although the empirical strategy adopted controls for many unobserved variables, there might stillexist some endogeneity with respect to the time measures. In face of the challenge of finding


suitable instrumental variables for each of our time variables, our strategy to evaluate the presenceof endogeneity is to re-estimate our models while imposing sample restrictions. The goal is to ver-ify how the results change as the sample varies in relation to the coverage of ports and products.

Table 6 presents the estimates only with observations in which the two ports considered in theratio are in the same geographic region. This additional restriction reduces the number of observa-tions, but it also strengthens the assumption made to control for “behind-the-border” trade facilita-tion. The first four columns show the results for the volume models, and the last four columnsshow the results for the diversification models.

Another sensitivity exercise was performed with respect to port sizes. The size of the port mightbe correlated with other port characteristics that affect the efficiency of the port, specially the timevariables. Therefore, we conducted an exercise in which we re-estimated the models without the

TABLE 5 Impacts of turnaround time on export diversification

Dependent variable: ln(Ratio Number of Exported Products)

Fixed-Effects Models

(1) (2) (3) (4)

ln(Ratio Export Coefficient) t � 1 0.071*** 0.072*** 0.073*** 0.073***

(0.019) (0.019) (0.019) (0.018)

ln(Ratio Number of Dockings) 0.340*** 0.327*** 0.248*** 0.276***

(0.078) (0.078) (0.076) (0.075)

ln(Ratio Port Cost) �0.093 �0.054 �0.090 �0.132

(0.097) (0.107) (0.098) (0.093)

ln(Ratio Turnaround time) �0.099**

(0.047)

ln(Ratio Operation Time) �0.075

(0.052)

ln(Ratio Waiting Time + Undock) �0.071**

(0.030)

ln(Ratio Waiting to Dock) �0.012

(0.009)

ln(Ratio Docked Time) �0.082

(0.052)

ln(Ratio Waiting Time Rate) �0.051

(0.036)

Constant �0.235*** �0.228*** �0.187*** �0.210***

(0.048) (0.048) (0.045) (0.045)

Observations 16,062 16,062 16,062 16,047

Year dummies Yes Yes Yes Yes

R2 .004 .004 .004 .004



TABLE

6Resultsrestrictingratio

sforpo

rtslocatedin

thesameregion

Restrictio

n:same-region

ports

ln(R

atio

Exp

ortedva

lueFO

B)

ln(R

atio

Num

berof

Exp

ortedProdu

cts)

ln(Ratio

ExportCoefficient)t�

10.101

0.116*

0.10

60.12

1*0.05

4***

0.05

8***

0.05

9***

0.05

3***

(0.068

)(0.069

)(0.066

)(0.066

)(0.019

)(0.020

)(0.021

)(0.020

)

ln(Ratio

Num

berof

Docking

s)1.815*

**1.871*

**0.58

6**

1.01

2***

0.30

7***

0.29

2***

0.14

50.22

4**

(0.306

)(0.320

)(0.279

)(0.259

)(0.093

)(0.097

)(0.100

)(0.089

)

ln(Ratio

PortCost)

�1.059**

*�1

.262

***

�0.672

*�1

.176

***

�0.158

�0.104

�0.058

�0.150

(0.322

)(0.356

)(0.366

)(0.344

)(0.132

)(0.143

)(0.144

)(0.134

)

ln(Ratio

Turnaroundtim

e)�0

.994**

*�0

.087

(0.172

)(0.060

)

ln(Ratio

Operatio

nTim

e)�0

.074

�0.096

(0.204

)(0.082

)

ln(Ratio

Waitin

gTim

e+Und

ock)

�0.733

***

�0.069

*

(0.123

)(0.039

)

ln(Ratio

Waitin

gTim

eforDocking)

�0.115

***

�0.014

(0.021

)(0.009

)

ln(Ratio

DockedTim

e)�0

.446

*�0

.174

*

(0.245

)(0.099

)

ln(Ratio

Waitin

gTim

eRate)

�0.304

***

�0.043

(0.073

)(0.038

)

Con

stant

1.353*

**1.359*

**0.71

6***

0.86

2***

0.11

3***

0.10

7**

0.05

10.06

5

(0.150

)(0.154

)(0.129

)(0.127

)(0.042

)(0.043

)(0.045

)(0.043

)

Observatio

ns4,408

4,408

4,40

84,39

66,08

76,08

76,08

76,07

4

Yeardu

mmies

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

R2

.099

.104

.088

.077

.004

.005

.005

.004

Notes:Standard

errors

robustto

heteroscedasticity

arein

parentheses.

The

coefficientswith

***aresignificantat

the1%

level;with

**aresignificantat

the5%

level;andwith

*aresignificantat

the10%

level.

Source:Datacompiledby

theauthors.


largest and smallest ports. After eliminating the observations from Santos (the largest port) andSantar�em, Itaqui, Bel�em, Imbituba, Vila do Conde and Fortaleza (the smallest ports),19 our samplebecomes more homogeneous with respect to port size.

Table 8 present the estimates without the top 20 products (2-digit HS) exported through theports included in our sample.20 In Brazil, many of the sampled ports specialise in the export ofspecific products. Thus, ports can invest more in loading and unloading capacity or even in spe-cial access for the port to specific products. By eliminating the main exported products, the esti-mates of the impact of time procedures become less susceptible to possible endogeneity that mayexist between the exported amount of specific goods and the time for each step of the portprocedure.

In general, the control variables maintain their signals and statistical significance. Consideringthe time variables, the pattern shown in Table 4 was maintained, that is, turnaround time continuesto be an important determinant of trade volume, and the most important steps of the procedure arethose related to waiting. Furthermore, the magnitude of the coefficients of the time variablesincreased compared with those presented in Table 4. In fact, according to this new set of estimates,a 10% decrease in relative turnaround time is reflected in an increase of 6.37% to 14.94% in localexports. Reductions of 10% in waiting or in time to undock are associated with increases of 4.13%to 10.34% in relative export volume.

The results show that the coefficient of waiting time to dock is the most insensitive to thechanges in the sample for volume models. According to the results, a 10% reduction in waitingtime to dock would increase the relative exports of Brazilian cities by 1.13% to 1.38%. Moreover,when the sample considers same-region ports only, the docked time becomes statistically signifi-cant at the 10% level. Specification (4) in all three tables also yields relatively higher coefficientsfor waiting time when compared to the results presented earlier.

Regarding the models estimated with the ratio of the number of products as a dependent vari-able, the coefficients of the control variables maintain the same general pattern in terms of sig-nals and statistical significance as those presented in Table 5. Considering the estimations forsame-region ports (Table 6) or excluding the largest and smallest ports (Table 7), the coefficientof total turnaround time becomes statistically non-significant. According to the results shown inTables 6 and 7, the undocking time is the only measure with significant impacts on the numberof products.

On the other hand, considering the estimations without the most-traded products (Table 8),the results are closer to those presented in Table 5. In this case, a 10% reduction in total turn-around time would increase the number of exported goods by ~1%. Moreover, a 10% reduc-tion in downtime or in undocking time would increase the number of exported products by0.96%.

Based on these exercises, we can conclude that each additional hour a ship takes to com-plete export procedures while in a port implies negative impacts on Brazilian exports, regard-less of the sample analysed and the restrictions on the constructed ratios. While some variationmay occur in the magnitude of the coefficients and in the relative importance of each step, theoverall framework that the ineffective steps are the most important determinants does notchange.

19As an exclusion criterion, using the number of dockings as a measure of port size, we eliminate all ports that have <200or >1,000 dockings per year.20The product categories excluded from the sample in this exercise were the following: 02, 09, 10, 12, 15, 17, 20, 23, 24,26, 27, 28, 39, 40, 47, 72, 84, 85, 87 and 99.


4 | CONCLUSION

In this paper, we built an innovative database with local export data considering the portused, country of destination and products aggregated at the four-digit HS for the period

TABLE 7 Results restricting the sample by eliminating the largest and smallest ports

Without the largest and smallest ports

ln(Ratio Exported value FOB) ln(Ratio Number of Exported Products)

ln(RatioExportCoefficient)t�1

0.161* 0.164* 0.105 0.103 0.082*** 0.089*** 0.087*** 0.078***

(0.088) (0.090) (0.086) (0.084) (0.020) (0.021) (0.021) (0.021)

ln(RatioNumber ofDockings)

2.418*** 2.310*** 0.725** 1.113*** 0.389*** 0.262** 0.206** 0.293***

(0.363) (0.378) (0.324) (0.308) (0.109) (0.121) (0.105) (0.096)

ln(Ratio PortCost)

�0.476 �0.509 �0.438 �0.924** �0.078 0.066 �0.001 �0.083

(0.387) (0.450) (0.430) (0.425) (0.141) (0.162) (0.151) (0.141)

ln(RatioTurnaroundtime)

�1.494*** �0.108

(0.209) (0.068)

ln(RatioOperationTime)

�0.367 �0.204*

(0.273) (0.106)

ln(RatioWaitingTime +Undock)

�1.034*** �0.042

(0.145) (0.048)

ln(RatioWaitingTime forDocking)

�0.138*** �0.012

(0.031) (0.009)

ln(RatioDockedTime)

�0.471 �0.182*

(0.303) (0.103)

ln(RatioWaitingTime Rate)

�0.431*** �0.048

(0.134) (0.037)

Constant 0.608*** 0.583*** 0.276** 0.196** �0.035 �0.003 �0.026 �0.060**

(0.097) (0.110) (0.110) (0.090) (0.026) (0.031) (0.029) (0.025)

Observations 4,117 4,117 4,117 4,117 5,362 5,362 5,362 5,362

Yeardummies

Yes Yes Yes Yes Yes Yes Yes Yes

R2 .113 .113 .081 .065 .006 .007 .007 .006



TABLE

8Resultsrestrictingsampleby

elim

inatingthemost-traded

prod

ucts

With

outmost-trad

edgo

ods

ln(R

atio

Exp

ortedva

lueFO

B)

ln(R

atio

Num

berof

Exp

ortedProdu

cts)

ln(Ratio

ExportCoefficient)t�

10.274*

**0.279*

**0.276*

**0.283*

**0.072*

**0.07

4***

0.07

4***

0.07

2***

(0.062

)(0.062

)(0.061

)(0.061

)(0.018

)(0.018

)(0.019

)(0.018

)

ln(Ratio

Num

berof

Docking

s)1.016*

**0.945*

**0.473*

*0.699*

**0.283*

**0.30

2***

0.19

1**

0.22

3***

(0.217

)(0.225

)(0.231

)(0.220

)(0.083

)(0.085

)(0.085

)(0.081

)

ln(Ratio

PortCost)

�1.452**

*�1

.469**

*�1

.564**

*�1

.794**

*�0

.174*

�0.160

�0.170

�0.219

**

(0.253

)(0.282

)(0.261

)(0.263

)(0.104

)(0.114

)(0.106

)(0.100

)

ln(Ratio

Turnaroundtim

e)�0

.637**

*�0

.099**

(0.144

)(0.050

)

ln(Ratio

Operatio

nTim

e)�0

.210

�0.040

(0.158

)(0.057

)

ln(Ratio

Waitin

gTim

e+Und

ock)

�0.413**

*�0

.096**

*

(0.091

)(0.031

)

ln(Ratio

Waitin

gTim

eforDocking)

�0.113**

*�0

.010

(0.021

)(0.008

)

ln(Ratio

DockedTim

e)�0

.193

�0.089

(0.185

)(0.064

)

ln(Ratio

Waitin

gTim

eRate)

�0.275**

*�0

.053

**

(0.073

)(0.026

)

Constant

�0.460**

*�0

.433**

*�0

.291**

�0.363**

*�0

.214**

*�0

.224**

*�0

.173

***

�0.197

***

(0.105

)(0.110

)(0.115

)(0.108

)(0.045

)(0.045

)(0.043

)(0.042

)

Observatio

ns8,104

8,104

8,104

8,092

9,978

9,97

89,97

89,96

3

Yeardu

mmies

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

R2

.060

.061

.063

.057

.006

.006

.006

.006

Notes:Standard

errors

robustto

heteroscedasticity

arein

parentheses.

The

coefficientswith

***aresignificantat

the1%

level;with

**aresignificantat

the5%

level;andwith

*aresignificantat

the10%

level.

Source:Datacompiledby

theauthors.


2010–12. A difference gravity equation for disaggregated data was estimated by the fixed-effects model.

According to the measures analysed, Brazilian ports in the sample present an average turnaroundtime of 28 hr and 12 min. From this total, an average of 15 hr and 42 min is spent with the shipdocked, and the rest is wasted in waiting. The ports from the sample presented an average downtime(time waiting to dock and time to the beginning of operations) of 45.17% in 2012. These resultsindicate the presence of a bottleneck in the flow of Brazilian production to the international market.Because of the growth in Brazilian trade flows in the last decade, these limitations have becomeeven more evident and point to the necessity of improvements in port capacity in order to reducethe inefficiency that generates such high percentages of downtime in Brazilian ports.

In general, the results suggest that the reduction in turnaround time can provide Brazilian expor-ters with better access to the international market. In fact, reducing turnaround time ultimatelyimplies a reduction in export costs, ensuring that the product will arrive at the destination marketwith higher quality and at a lower final cost, improving its competitiveness abroad. According tothe estimates, a 10% reduction in total turnaround time may increase local exports by ~5.1%.

The study also presents empirical evidence that turnaround time also affects the variety of prod-ucts that can be exported by a city. According to the results, a 10% reduction in the relative time aship stays in port can increase the number of product categories by ~1%. This result, althoughmore modest than the coefficients estimated for the trade volume models, is statistically significantand leads to the conclusion that the time spent specifically on port procedures is also an importantdeterminant of the variety of sectors involved in international trade.

The results also suggest that the steps involved in downtime are the most important determinants oftrade volume and diversification. A 10% reduction in waiting time or in undocking time could increasethe export volume by 3.5%. An equivalent reduction in waiting time to dock increases the volume oflocal exports by ~1%. However, reductions in operation time or in docked time seem to have no signifi-cant impact on the volume of trade. For the time effects on trade diversification, a 10% reduction indowntime or in undocking time can increase the number of product categories by ~0.7%.

In a world scenario where tariff barriers have been falling gradually as trade facilitation elementsgain importance in trade negotiations and in public policy agendas, this study provides evidence ofthe high and statistically significant impact of turnaround time on Brazilian exports. Moreover, ifwe think of the waiting time as being directly related to port capacity, the results suggest that invest-ments in the expansion and modernisation of port infrastructure, which would ensure increasedaccess to ports. This could reduce waiting time for ships and expedite loading and unloading opera-tions, which would significantly impact exports at both the intensive and extensive margins.

ORCID

Vin�ıcios P. Sant’ Anna http://orcid.org/0000-0002-1070-2220

REFERENCES

Anderson, J. E. (2011). The gravity model. Annual Review of Economics, 3(1), 133–160. https://www.annualreviews.org/doi/abs/10.1146/annurev-economics-111809-125114

Anderson, J. E., & Van Wincoop, E. (2003). Gravity with gravitas: A solution to the border puzzle. The AmericanEconomic Review, 93(1), 170–192. https://doi.org/10.1257/000282803321455214

Clark, X., Dollar, D., & Micco, A. (2004). Port efficiency, maritime transport costs, and bilateral trade. Journal ofDevelopment Economics, 75(2), 417–450. https://doi.org/10.1016/j.jdeveco.2004.06.005


http://orcid.org/0000-0002-1070-2220

http://orcid.org/0000-0002-1070-2220

http://orcid.org/0000-0002-1070-2220

https://www.annualreviews.org/doi/abs/10.1146/annurev-economics-111809-125114

https://www.annualreviews.org/doi/abs/10.1146/annurev-economics-111809-125114

https://doi.org/10.1257/000282803321455214

https://doi.org/10.1016/j.jdeveco.2004.06.005

Djankov, S., Freund, C., & Pham, C. S. (2010). Trading on time. The Review of Economics and Statistics, 92(2),166–173. https://doi.org/10.1162/rest.2009.11498

Ducruet, C., Itoh, H., & Merk, O. (2014). Time efficiency at world container ports (Discussion Paper No. 2014-08),THE INTERNATIONAL TRANSPORT FORUM (ITF). Paris, France: OECD Publishing. Retrieved from:https://www.itf-oecd.org/sites/default/files/docs/dp201408.pdf

Helble, M., Shepherd, B., & Wilson, J. S. (2007). Transparency and trade facilitation in the Asia Pacific: Estimat-ing the gains from reform. Canberra, ACT: Department of Foreign Affairs and Trade, Australia.

Hummels, D. L., & Schaur, G. (2013). Time as a trade barrier. American Economic Review, 103(7), 2935–2959.https://doi.org/10.1257/aer.103.7.2935. Latest version available at http://www.nber.org/papers/w17758

Kahle, D., Wickham, H., & Kahle, M. D. (2013). Package “ggmap”.Khan, I. U., & Kalirajan, K. (2011). The impact of trade costs on exports: An empirical modelling. Economic Mod-

elling, 28(3), 1341–1347. https://doi.org/10.1016/j.econmod.2011.02.031Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate industry productivity. Econ-

ometrica, 71(6), 1695–1725. https://doi.org/10.1111/1468-0262.00467Mesquita Moreira, M., Blyde, J. S., Volpe Martincus, C., & Molina, D. (2013). Too far to export: Domestic trans-

port costs and regional export disparities in Latin America and the Caribbean. In M. Mesquita Moreira (Ed.),The elusive obvious: Transport costs and Brazil’s regional export disparities. Special report on integration andtrade (pp. 25–72). Washington, DC: Inter-American Development Bank.

Nord�as, H. K. (2007). Time as a trade barrier: Implications for low-income countries. OECD Economic Studies,2006(1), 137–167. http://doi.org/10.1787/eco_studies-v2006-art4-en

Shepherd, B., & Wilson, J. S. (2009). Trade facilitation in ASEAN member countries: Measuring progress andassessing priorities. Journal of Asian Economics, 20(4), 367–383. https://doi.org/10.1016/j.asieco.2009.03.001

Wilson, J. S., Mann, C. L., & Otsuki, T. (2005). Assessing the benefits of trade facilitation: A global perspective.The World Economy, 28(6), 841–871. https://doi.org/10.1111/j.1467-9701.2005.00709.x

WTO (2013). Briefing note – trade facilitation: Cutting “red tape” at the border. 9th WTO Ministerial Conference,Bali.

How to cite this article: Sant’ Anna VP, Kannebley J�unior S. Port efficiency and Brazilianexports: A quantitative assessment of the impact of turnaround time. World Econ.2018;00:1–24. https://doi.org/10.1111/twec.12654


https://doi.org/10.1162/rest.2009.11498

https://www.itf-oecd.org/sites/default/files/docs/dp201408.pdf

https://doi.org/10.1257/aer.103.7.2935

http://www.nber.org/papers/w17758

https://doi.org/10.1016/j.econmod.2011.02.031

https://doi.org/10.1111/1468-0262.00467

http://doi.org/10.1787/eco_studies-v2006-art4-en

https://doi.org/10.1016/j.asieco.2009.03.001

https://doi.org/10.1111/j.1467-9701.2005.00709.x

https://doi.org/10.1111/twec.12654

Documents

Port efficiency and Brazilian exports: A quantitative ... · ORIGINAL ARTICLE Port efficiency and Brazilian exports: A quantitative assessment of the impact of turnaround time Vinıcios