imf a.behar a comparison of alternative sources of trade ...thomas-orliac.net/conferences/isel/jour...

Do managers and experts agree? A Do managers and experts agree? A comparison of alternative sources comparison of alternative sources

of trade facilitation dataof trade facilitation dataof trade facilitation dataof trade facilitation data

Alberto Behar, IMF

ISEL Conference, Le Havre

November 2012

2 questions2 questions

1 How do indicators of trade facilitation 1. How do indicators of trade facilitation based on firm level data compare with multiple macro sources?multiple macro sources?

2. What is the relative importance of cross-country and cross firm / within country country and cross-firm / within-country variation?

A comparison of alternative sources of trade facilitation data

2Alberto Behar

Micro DataMicro Data1. World Bank Enterprise Surveys

– 39 734 firms in 87 developing countries (2006-2009)39, 734 firms in 87 developing countries (2006 2009)

– Firm-level manager surveys

– >10 employees

– Exporters and non-exportersExporters and non exporters

Average

days for Max days

Customs &

trade

Number of

% sample

firms

exporting

days for

exports

customs

clearance

Max days

for exports

customs

clearance

trade

constraint on

a 0 (best)-to-

4(worst)

firms directly (mean) (mean) scale (mean)

Chile 1,017 21% 6.5 10 0.6

Uganda 563 8% 5 10 0.7


3Alberto Behar

Macro Data W ld B k D i B i2. World Bank Doing Business

– Coordinated by international law firms in 183 countries + Information from local freight forwarders, shipping lines, customs brokers, port officialsofficials

– “Representative firm” the same for all countries

• 60 or more employees, exports more than 10% of sales

• Located in the country’s most populous city• Located in the country s most populous city

Trading Across Borders in Afghanistan (Country Overall Rank #183)

Nature of Export Procedures DaysUS$

CostExport Documents

Documents preparation 44 650

Bill of lading, Certificate of origin, Clean inspection report of

findings,

Customs clearance and

technical control 8 550

Commercial invoice, customs export declaration, customs

transit documenttechnical control 8 550 transit document

Ports and terminal handling 4 150

Duties exemption form, insurance certificate, packing list, tax

certificate,

Inland transportation and

h dli 18 2000

Terminal handling receipts, NOC/transit

ihandling 18 2000 permit

Totals: 74 3350


4Alberto Behar

Macro DataMacro Data3. World Bank Logistics Performance Index (LPI)

– 2007, 2010 and 2012 versions (2010 here)7

– International logistics professionals in 155 countries

– Each answers questions about a different set of 8 countries

– Subjective question on trade facilitation, rating from 1 to 5:j q , g 5

Country LPI Customs Infrastructure International

shipments

Logistics

competence

Tracking &

tracing

Timeliness

Germany 4.11 4 4.34 3.66 4.14 4.18 4.48

Singapore 4.09 4.02 4.22 3.86 4.12 4.15 4.23

Sweden 4.08 3.88 4.03 3.83 4.22 4.22 4.32

Correlation with LPI

Overall: 0 96 0 97 0 86 0 97 0 95 0 91Overall: 0.96 0.97 0.86 0.97 0.95 0.91


5Alberto Behar

Macro DataMacro Data4. World Economic Forum Enabling Trade Index (ETI)

– 121 countries

– 2008,2010,2012 (2010 here)

– External sources + Executive Opinion Survey

– Un-weighted average sub-indices, un-weighted overall averageg g , g g

Correlation with ETI

Overall: 0.20 0.96 0.92 0.89


6Alberto Behar

Export lead timeExport lead time

0

Correlation: 0.19

Enterprise Surveys vs. Doing Business (All Countries)

30

40

20

01

0

5.6 7.0

ES avg export clearance days (mean) DB customs clearance + port handling (days)

Mean: 6.8 Mean: 7.9Mean: 6.8

Std Dev: 4.4

Mean: 7.9

Std Dev: 5.1


7Alberto Behar

Micro-Macro CorrelationsMicro Macro CorrelationsES

Average

t

ES

Maximum

t

DoingBusiness

C t Doing

B i

Cross-CountryCorrelations

export

clearance

days

(mean)

export

clearance

days

(mean)

ES Customs Regulations

(mean)

Customs+ Ports

Handling(days)

Business Total

Export Time (days) LPI ETI

ES Average export clearance (mean days) 1.00

ES Maximum export clearance (mean days) 0.86 1.00

ES Customs regulations

obstacle (mean) 0.34 0.22 1.00

DB Customs + Ports

(days) 0 19 0 26 0 03 1 00(days) 0.19 0.26 0.03 1.00

DB Total Export Time (days) 0.45 0.32 0.25 0.55 1.00

LPI 0.15 0.07 0.25 0.22 0.56 1.00

ETI 0.52 0.39 0.55 0.24 0.64 0.85 1.00


Alberto Behar

Customs clearance correlationsCustoms clearance correlations

520

vg

30

vg

510

1

(mean)

expcla

v

10

20

(p 5

0)

expcla

v

0

0 10 20 30 40DB customs clearance + port handling (days)

0


40

30

020

30

mean)

expclm

ax

10

20

p 5

0)

expclm

ax

010

(m


0

(p



9Alberto Behar

Export lead timepExample countries, all firms

200

(days)

150

xp

ort

Cle

ara

nce

1

00

eys A

ve

rag

e E

x0

50

En

terp

rise S

urv

e

3 5.5 103 3 1

12.5

Argentina Bolivia Brazil Chile Colombia Uruguay Venezuela


10Alberto Behar

Between-country variation:l dclearance days

DBES

Mean Median Customs/ports

Interquartile range 5.2 3.5 4.0

St d d D i ti 4 6 4 3 5 0Standard Deviation 4.6 4.3 5.0

Minimum 1.3 1.0 2.0

Maximum 20 4 30 0 34 0Maximum 20.4 30.0 34.0

Mean 6.8 4.1 7.8

Std. Dev / Mean 0.7 1.1 0.6


11Alberto Behar

Within-country variation:l dclearance days

C t t t IQR SD SD ( i ht d) R tiCross-country stat IQR SD SD (weighted) Ratio

Mean 7.4 9.4 19.9 2.8

Median 5.0 8.2 13.7 2.5

Mi i 0 0 1 0 1 5 0 3Minimum 0.0 1.0 1.5 0.3

Maximum 29.0 33.4 97.7 8.6

Note: Ratio is SD (weighted) to within-country mean


12Alberto Behar

Within-country or between-country variationWithin country or between country variation

i h d i h d i h d i h d i h d i h d

Constant Country dummies Country & year dummies

unweighted weighted unweighted weighted unweighted weighted

RMSE 11.19 10.70 10.63 9.31 10.63 9.31

R2

0.00 0.00 0.11 0.26 0.11 0.26

p value (countries) . . 0.00 0.00 0.00 0.00

p value (years) . . . . 0.02 0.00

Table : Selected statistics from firm level regressions of export clearance daysTable : Selected statistics from firm-level regressions of export clearance days.


13Alberto Behar

Summary and implications (future k )work...)

1 Low correlation between country-level 1. Low correlation between country level indicators of trade facilitation: How reliable are country rankings?reliable are country rankings?

2. Relatively small role for country averages: Are cross country differences still Are cross-country differences still interesting or should we be directing more energy to understanding within country energy to understanding within-country variation?


14Alberto Behar

F.R.E.I.T WORKING PAPER

1

Alberto Behar*

August 2010 VERY PRELIMINARY: COMMENTS APPRECIATED!

This paper constructs country level aggregates of trade facilitation measures from firm level

responses in the Enterprise Surveys and compares them with the Doing Business indicators, the

Logistics Performance Index and the Enabling Trade Index. Correlations between the data

sources are low even for very specific and similar questions. We also use the Enterprise Surveys

to distinguish between within country inter firm variation and between country variation,

finding that the latter accounts for only a quarter of the total. For the purposes of identifying

where reform is needed and estimating the relationship between trade facilitation and exports,

these findings raise the issue of which form of variation is more informative and which data

source is more reliable.

1 INTRODUCTION

International trade has grown fast in recent years, helped by the signing of multilateral and other trade

agreements, but many countries remain relatively isolated. One reason is that transport costs remain

high in many parts of the world. While this is in part due to geography – many countries are landlocked

or far from attractive markets – man made and policy characteristics can help as well. Physical

infrastructure like roads, communications and ports have been found to be positively associated with

trade flows. As a result, large investments have been made by governments and multilateral institutions

to improve trade related infrastructure (Behar & Venables, 2010).

However, policymakers and researchers have recently turned their attention to the institutional and

administrative barriers to trade. Reforms aimed at removing this type of barrier are often referred to as

trade facilitating reforms. The extent to which such barriers exist can be very important because, given

recent investment, infrastructure may no longer be the binding constraint. Furthermore, unlike hard

infrastructure, it can be cheaper and easier to implement trade facilitating reforms. A number of studies

have concluded that countries with a higher degree of trade facilitation – lower

administrative/institutional obstacles to trade – tend to have higher trade flows (Wilson et al, 2005;

Clarke et al, 2004).


2

Much of the discussion is in the context of country level (or bilateral) characteristics that affect

transport costs, but a recent literature has considered the role of firms in international trade. Few firms

export from any given country, yet international trade appears to be dominated by a few multinationals

(Bernard et al, 2007). It then becomes pertinent to make the firm the unit of analysis rather than the

country. Why one country exports more than another remains an important question, but asking why

one firm exports and another doesn’t or why some countries have more exporting firms than others are

also good questions. It’s also important to establish whether the answers to all the questions have the

same implications for the importance of trade facilitation.

Much empirical work on trade facilitation has made use of macroeconomic data in gravity models.

Recent work inspired by Melitz (2003) recognizes that firms are not homogenous and uses this insight to

explain various features of international trade observed at both the firm and country level. Still using

macroeconomic data, gravity models have been modified to be able to distinguish between the effects

of trade costs on the proportion of firms exporting from a country as well as the quantity that each firm

exports (Helpman et al, 2008).

There are a number of macroeconomic sources of country level trade related indicators, including the

Doing Business indicators, the Logistics Performance Index and the Enabling Trade Index. These all

provide measures of trade facilitation that are candidate regressors in gravity models.1These differ in

their scope, methodology and coverage but one thing they have in common is that the information is

drawn from a number of experts but not the firms who are actually exporting. The Enterprise Surveys

are firm level surveys that include questions on trade facilitation and international trade and a handful

of studies have used this data for selected regions.2

Furthermore, the Enterprise Surveys have been consistently conducted across a number of countries for

the purpose of cross country comparability.3One objective of this paper is to present aggregate

summary trade facilitation statistics based on firm level responses, so we construct various country level

summary statistics of the firm level responses.

The second objective is to compare the different data sources, especially answers from the

microeconomic data with close analogues available in the macro sources. Focusing on measures that

facilitate exports (as opposed to imports), we find that various distinct indicators from the same source

are highly correlated, but measures of the same characteristic from different sources have a low

correlation. This happens even if we are quite precise about the type of question. For example, the

Enterprise Surveys and Doing Business both have information on the number of days exports take to

clear ports and customs yet the correlation is only 0.13.

This is important because the countries identified as being in need of reform can differ depending on

the source. From a measurement or econometric point of view, for example gravity models of trade,

1See for example Djankov et al (2010) for Doing Business, Behar et al (2009) for the LPI and Lawrence et al (2008)

for the ETI.2Li & Wilson (2009) do so for Asian firms while Balchin & Edwards (2008) do so with African firms.

3However, the World Bank notes that the cross country comparability characteristics of the Doing Business

dataset are superior. See at http://www.enterprisesurveys.org/Methodology/Compare.aspx


3

these indicators are usually interpreted as proxies for some underlying country characteristics.

Therefore, we discuss whether the methodologies produce alternative proxies for the same thing or

whether the issue being investigated becomes different. It also raises the issue of whether one question

is more relevant than another or whether one proxy is more reliable than another.

The third objective is to use the Enterprise Surveys to compare the variation occurring within countries

(between firms) with that occurring between countries. We find that cross country variation explains

only one quarter of the total. This suggests macroeconomic studies are ignoring most of the variation in

trade facilitation experience and raises the issue of whether a focus on countries is appropriate or

whether the within country variation is more interesting, useful or relevant for measurement and policy.

The answer to this depends in part on what the reason for the variation is. We therefore discuss four

interpretations of the cross firm variation in trade facilitation experience: (i) known firm specific random

draws from a distribution of ‘trade facility’, (ii) known firm specific but endogenous trade facility, (iii)

uncertain/stochastic trade facility common to all firms but varying for every shipment and, (iv)

noise/measurement error.

Section 2 introduces the various data sources, including their scope, coverage and methodology. This

includes the Enterprise Surveys, where we also explain our approach to producing country level

summary statistics and comparing them with the macroeconomic sources. Section 3 presents and

compares the descriptive statistics from each source, noting that the correlations between various

measures are low. In general, we present descriptive measures for the whole world but also illustrate

with examples from Central Asian countries. In addition, we decompose the variation into that between

countries and that within countries, observing that the latter is much bigger.

Section 4 discusses the findings. It provides alternative interpretations of the within firm variation,

attempts to reconcile the data sources and evaluates their relative merits for estimation and policy. We

remain unsure about which source is more appropriate, but hope that these empirical findings raise

awareness of the potential importance of the various data sources and how they are generated. Until

differences are properly reconciled or understood, empirical work should use more than one source in

the interests of robustness.


4

2 DATA CONSTRUCTION

This section discusses each of the original data sources. The core source of microeonomic data is the

Enterprise Surveys. The macroeconomic datasets are the Doing Business indicators, the Logistics

Performance Index and the Enabling Trade Index.

2.1 Enterprise Survey data description

The World Bank Enterprise Survey (ES) data is available from http://www.enterprisesurveys.org/. There

are two “core” or “comprehensive” data sets, which group countries together with comparable survey

questions. One set is for the years 2002 through 2006 and the other set is for the years 2006 through

2009. Because many survey questions are different between the two periods, the two datasets are

warehoused separately and we concentrate on the latter period. Therefore, we have access to

responses from about 40,000 firms across 87 developing countries, taken in various years from 2006 to

2009. This covers a very broad range of topics but we are particularly interested in answers to a number

of quantitative and qualitative questions regarding trade and trade facilitation. The following variables

are retained from the Enterprise Surveys:4

indirect exports as % total sales

direct exports as % total sales

Average days (over the past 2 years) it took to clear export customs from day of arrival at port

Maximum days (over the past 2 years) it took to clear export customs from day of arrival at port

Perception of customs and trade regulations as a constraint to business (index from 0 4)

While we will analyse some of the data at the firm level, a key component of our exercise is to calculate

summary statistics at the country level. Therefore, for each of these variables, we calculate the mean,

median, standard deviation and interquartile range. The measures of central tendency are in principle

comparable to the macro indicators. The dispersion measures can have a variety of uses, including

comparisons of within and between country variation and, depending on what one believes is

generating the dispersion, can be informative about the degree of uncertainty faced by firms.

We summarise the statistics across all firms who responded but prefer to use those summarizing

responses from exporters for a number of reasons. First, many trade analyses are by definition

conditioning on firms who export. Second, it is hard to interpret some answers from non exporters. For

example, if customs and trade regulations are not a constraint on the business, this can be because the

requirements are not onerous or simply because this constraint is never encountered by non exporters

4We also construct a variable for total exports as a percentage of sales and a dummy for whether or not the firm

was an exporter. For completeness, we have import analogues to the export measures as well as information on

waiting times for a licence and whether bribes were used for one.


5

(or importers). Third, for more objective measures, many non exporting firms are likely to have no

experience of actual processes and are therefore likely to be guessing. Fourth, the structure of the

questionnaire does not appear to explicitly instruct enumerators to skip these questions for non

exporters, but response rates are much lower for this group.5Four countries do not have enough

exporters who answered the questions so we effectively have 83 countries.

Further, we provide the above statistics as sample summary statistics but also try to reflect the

population summary statistics by taking account of the survey design. When the surveys are conducted,

firms of all sizes and ownership structures were interviewed, but certain industries were concentrated

on for cross country comparability. The sampling methodology for Enterprise Surveys is stratified

random sampling with replacement. Stratified random sampling groups all population units into

homogenous groups and then selects simple random samples from each group. This includes an over

sampling of firms with over 100 employees (World Bank, 2009a). As a result, population estimates must

take account of population weights when calculating the means and must account for both the weights

and stratification6when calculating the standard deviation.

Because the Enterprise Surveys aim to visit most countries every three years, a handful got surveyed

twice in the 2006 9 period. Although exploiting time series variation within a panel is a fruitful line of

enquiry, our exercise only uses the latest survey for those countries.7

2.2 Macroeconomic datasets

2.2.1 Doing Business data description and survey years

The Doing Business (DB) report is produced annually by the World Bank and International Finance

Corporation. The 2010 edition (World Bank, 2009b) includes 183 countries, including developed and

developing nations. DB surveys are conducted in person with local experts, including lawyers,

consultants, accountants, freight forwarders and government officials, to verify the de jure requirements

for each step of the trading process, including each piece of paperwork, payment, and license necessary

5Overall, response rates can be low as a result. For example, less than 6,000 firms gave a number when asked

about the days it takes to clear customs.6For a handful of countries, there was no obvious stratification variable so we assumed one stratum. For those

with multiple strata of which some have only one sampling unit, we treat these as certainty equivalents with

scaling based on the variances of the other strata.7The timing of the Enterprise Surveys within years varies. In 2008 and 2009 surveys, enumerators record the time,

day, month, and year in which the survey is taken. Unfortunately, for 2005, 2006, and 2007 reports, the

enumerators do not record the survey date. Among the 2008 and 2009 surveys, there is no consistent pattern for

whether the year of the report and year of the survey correspond. That is, about half of the reports from 2009 had

surveys conducted in 2009 and the other half in 2008. Therefore, for 2008 and 2009, we use the year in which the

majority of surveys were taken as the year of the survey, whereas for 2005, 2006, and 2007, we use the reporting

year.


6

for a representative business to export or import. When surveying the experts, the firms in question are

assumed, inter alia, to have more than 60 employees, export at least 10% of their sales, be domestically

owned and located in the country’s most populous city (World Bank, 2009b).8

The World Bank “Trading Across Borders” section, located at

http://www.doingbusiness.org/ExploreTopics/TradingAcrossBorders/, reports six main measures of

trade facilitation:

number of documents to export or to import

days to export or to import

cost to export or import a standard shipping container in dollars

These six variables are available for all years, but the 2010 edition disaggregates the time

component for some countries. The breakdowns are:

days to clear ports (export or import)

days spent on document preparation (export or import)

days spent on in land transport (export or import)

The breakdown is available for exports and for imports and can be accessed through each country’s

profile. So, for document preparation, the 2010 edition has both the number of documents and the days

taken to process them.

We used the 2010 edition in order to have the time breakdowns but also use data from the edition that

corresponds to the year that the Enterprise Survey was conducted in each country. This is done with a

one year lag because Doing Business reports are typically released in the year preceding the report’s

label, so that data in the report corresponds to the previous year. For example, since the most recent

Enterprise Survey was conducted in Albania in 2007, data from the Doing Business 2008 report was used

for all Doing Business variables in Albania. For completeness, we also have the averages from the 2005

9 reports.

2.2.2 LPI

Starting in 2007, the World Bank Logistics Performance Index (LPI) will be based on surveys conducted

every two years. The two editions of the data currently available are discussed in reports named

Connecting to Compete: Trade Logistics in the Global Economy (Arvis et al, 2007, 2010). The reports and

data are available at www.worldbank.org/lpi. The 2010 report includes information for 155 countries,

including developed and developing countries.

8This applies to the “Trading Across Borders” component of the survey. Other components have different firm

characteristics.


7

The LPI reports six sub indexes and an overall index: (1) efficiency of customs clearance, (2) quality of

trade and transport related infrastructure, (3) ease of arranging competitively priced shipments, (4)

competence and quality of logistics services, (5) ability to track and trace consignments, and (6)

frequency with which shipments reach the consignee within the scheduled or expected delivery time.

The overall index is constructed from the 6 dimensions using principle components analysis. While both

reports share the same components, the 2010 report offers more detailed breakdowns and an

expanded emphasis on internal logistics. As a result, the 2010 edition makes a more explicit distinction

between local and international logistics although the international part is largely unchanged in content.

The LPI draws from a structured online survey receiving nearly 1,000 responses from logistics

professionals who are based in international logistics companies in 130 countries. Ten percent of

respondents are located in low income countries, 45 percent in middle income countries, and 45

percent in high income countries. Each respondent is asked to rate 8 overseas markets on logistics

performance using a qualitative assessment.9The 8 markets are different for each respondent, based

on the most important export and import markets of their location country, neighboring countries that

facilitate their goods transport to ports, and random selection (Arvis et al, 2010).

2.2.3 ETI

The World Economic Forum’s Enabling Trade Index (ETI) is a meta index that takes unweighted averages

of other indicators and whole indices. Of its roughly 55 components, it includes 15 from its own survey,

the Executive Opinion Survey, which it carries out annually to ask CEOs and other top business leaders to

rank country capacities from 1 to 7. The other 40 or so components include quantitative and qualitative

indicators and indices from publicly available sources: International Trade Centre (13 components),

World Bank Logistics Performance Index (5), World Bank Doing Business (6), International

Telecommunication Union (4), World Economic Forum Global Competitiveness Index (5), United Nations

Conference on Trade and Development (UNCTAD) (2), and International Air Transport Association (IATA)

(1).

Normalizing each indicator or index to a 1 to 7 scale to match its Executive Opinion Survey, the ETI

creates the unweighted average for each of its four sub indices and then the overall unweighted

average of the sub indices. The sub indices are

market access, which measures the extent to which the policy framework welcomes foreign

goods into the country and enables access to foreign markets for domestic exporters

border administration, which assesses the extent to which the administration at the border

facilitates the entry and exit of goods

9The domestic component of the 2010 LPI and some aspects of the 2007 index were also backed up by

quantitative information from respondents.


8

infrastructure, taking into account whether the country has transport and communications

infrastructure to facilitate the movement of goods within the country and across the border

business environment, which looks at governance, security and the regulatory environment

impacting importers and exporters.

In turn, these are based on nine pillars – the mapping to the indices is not explicitly clear – and these

pillars are based on a number of individual questions and components. While many of the trade related

variables include measures from sources we have discussed separately in this paper, which means there

is some duplication by construction, trade related components do also come from other sources. The

results have since 2008 been published annually in the Global Enabling Trade Report, of which the 2009

edition covers 121 developed and developing countries (Lawrence et al, 2009).10

2.3 Comparison of coverage

The following matrix shows the number of countries overlapping between the four data sources:

Enterprise Surveys

2006 9

Doing Business

2010LPI 2010 ETI 2009

Enterprise Surveys

2006 0987 87 76 61

Doing Business

201087 183 149 121

LPI 2010 76 149 155 115

ETI 2009 61 121 115 121

The main disparity is due to the fact that the Enterprise Surveys do not cover developed countries. Doing

Business has the most countries and its coverage by and large nests that of the other sources.

The obvious difference between the Enterprise Surveys and the others is that actual firms are surveyed

about their experiences. In particular, a business owner or top executive is interviewed. While

accountants or human resource officers may be interviewed for some sections, there is no indication

that a person responsible for logistics or operations is asked. The range of firms is wide although we rely

predominantly on summary statistics for exporters only. The answers they give can be both objective

and subjective. DB asks mostly objective questions of a number of local experts and restricts itself to

10This includes a 2010 version after our dataset was put together. This latest report is available at

http://www.weforum.org/en/initiatives/gcp/GlobalEnablingTradeReport/index.htm


9

large exporters in a particular city.11LPI asks logistics professionals in a number of countries about other

countries and is by and large a perceptions based index. ETI is a composite of other macroeconomic

sources, including those discussed here, but also includes information from its own survey of CEOs

opinions. We return to these issues in section 4.

2.4 Other macroeconomic data

For completeness, we have added in a number of other variables on trade and macroeconomic

characteristics that are regularly incorporated in gravity models and other analyses of trade

relationships.

To get the most up to date data possible such that it matches some of our trade facilitation data, we

sourced information on GDP and the population from the “Historical Data Files” section of the USDA

Economic Research Service’s International Macroeconomics datasets web site:

http://www.ers.usda.gov/data/macroeconomics/. GDP is real GDP is billions of 2005 US dollars, as

obtained directly from the USDA dataset and the population is the number of people. We have this for

the years 2005 9 but also an average over the 2006 9 period matching our Enterprise Survey coverage.

We have data on country area and whether or not it is landlocked. We also include dyadic data on

distance, of which there are various measures and we take the subset based only on the capital city. We

have dyadic information on whether countries share a border or a former common colonizer, and

whether they share the same language. This data can be found at

http://www.cepii.fr/anglaisgraph/bdd/distances.htm.

Trade statistics taken from the IMF Direction of Trade database. The data are available in US dollars but

we deflated these to 2005 dollars using the deflator from USDA Economic Research Service. We have

two forms of the data. One takes the average over the 2006 9 period to mitigate measurement error

and to account for the fact that many of our data sources are not available annually.12

The second

matches the trade year to that of the Enterprise Survey for that country. Recall that we also have some

indicators of trade activity derived from the Enterprise Surveys.

11Further useful comparisons between Doing Business and the Enterprise Surveys are available at

http://www.enterprisesurveys.org/Methodology/Compare.aspx12We use the annual data rather than higher frequency versions.


10

3 DESCRIPTION AND COMPARISON OF TRADE FACILITATION DATA

This section provides descriptive statistics from each of the sources and then compares them with other

sources. Our focus is on trade facilitation and, where there is a distinction, on measures related to

exports as opposed to imports. We leave such complementary analyses to future research.

3.1 Macroeconomic data sources

The Logistics Performance Index (LPI) comes in 2007 and 2010 editions. In the bottom left half of the

correlation matrix (Table 2), we can see that all sub components of the 2007 LPI are highly correlated

with the overall index (column 1) and with each other. Similarly, row 1 shows the 2010 components are

also highly correlated with the overall index and the top right half shows they are still highly correlated

with each other, although the shipments measure, which is the ease and expense with which one can

arrange the shipments of goods overseas, is less correlated with the others.

LPI 2010 correlations (top right)

Overall Customs Infrastructure Shipment Logistics Tracking Timeliness

LPI

2007

correlations

(bottom

left)

Overall . 0.96 0.97 0.85 0.97 0.95 0.91

Customs 0.97 . 0.95 0.77 0.93 0.88 0.85

Infrastructure 0.97 0.96 . 0.79 0.96 0.90 0.85

Shipment 0.96 0.91 0.92 . 0.78 0.78 0.69

Logistics 0.97 0.93 0.94 0.93 . 0.92 0.86

Tracking 0.96 0.91 0.92 0.90 0.94 . 0.83

Timeliness 0.92 0.86 0.86 0.85 0.87 0.88 .

Table 2: Correlations within various components of the Logistics Performance Index. Top right of matrix gives correlation

between components for 2010 while bottom left gives correlations for 2007 data.

Further, we report that the correlation between the 2007 and 2010 overall indices is 0.90, which

suggests some movement by countries over the period. This also is found for the correlation for the

ranks and as well as the Kendall Tau rank correlation statistic, which is 0.70. Further, the indications are

that, overall, the performance has improved over time. Across the 118 countries, the mean score has

increased from 2.75 (out of 5) to 2.9 and the difference between the two samples is significantly

different. Figure 1 shows that the 2010 density estimate lies to the right of the 2007 estimate. These

results are consistent with Arvis et al (2010), who also identify individual country improvers.


11

Figure 1: Density estimates of the two editions of the LPI.

Table 3 performs a similar analysis for the Enabling Trade Index (ETI). The three measures that describe

operating conditions – transport infrastructure, business environment and border processes – are highly

correlated with the overall index (column 1) and with each other. Market access, which is about tariff

and non tariff trade restrictions, is not closely related to the other measures.

Overall

Transp.

Infra. Bus. Env.

Border

Procs.

Market

Access

Overall 1.00

Transport Infrastructure 0.92 1.00

Business Environment 0.90 0.81 1.00

Border Processes 0.96 0.91 0.84 1.00

Market Access 0.21 0.09 0.00 0.04 1.00

Table 3: Correlations within ETI components.

0.2

.4.6

.8D

ensity

1 2 3 4 5Logistics Performance Index

2010

2007

kernel = epanechnikov, bandwidth = 0.1790

Kernel density estimate


12

Table 4 presents information on the 2010 edition Doing Business results.13

The first row of the table

presents the median number of days reported for all the countries. The median is 20 days overall. To

have an approximate indication of the breakdown, 7 of these are due to customs and ports delays

(further broken down roughly equally) and 12 are due to documentation preparation. We also mention

that transit delays last a median of 3 days.

Total (days)

customs/

ports (days) ports (days)

customs

(days)

documents

(days)

documents

(number)

cost

(dollars)

Median 20 7 4 3 12 6 1190

Total days to export .

customs/ ports (days) 0.57 .

ports (days) 0.44 0.87 .

customs (days) 0.53 0.78 0.37 .

documents (days) 0.84 0.40 0.29 0.38 .

documents (number) 0.56 0.36 0.20 0.43 0.43 .

cost (dollars) 0.77 0.29 0.20 0.30 0.64 0.37 .

Table 4: median and correlations between Doing Business export trade facilitation measures (all taken from 2010)

The rest of Table 4 presents correlations between various components. The first column suggests that

much of the overall correlation in total days is accounted for by documentation delays, while the

correlation with the customs/ports component is only 0.57. There is a fairly high correlation between

the total number of days and the financial cost of shipping a standard container.14The correlation of

0.43 between the number of documents required and the days it takes to complete them is far from

perfect.

The purpose of Table 5 is comparison between the three macroeconomic sources. The DB measures are

lower for better indicators and the others are higher if better. The first panel is for the overall indices.

Recall that the ETI is in part based on other indices including the LPI and Doing Business. The first column

indicates a fairly high correlation between ETI and LPI but it is lower with the two DBmeasures.

13Where there is overlap, the correlation between the 2010 measure and the measure taken from the year in

which that country’s Enterprise Survey was conducted is over 0.95.14For a discussion and comparison of the time and pecuniary costs of shipping goods, see Behar & Venables

(2010).


13

ETI (overall) LPI (overall) DB Cost

ETI (overall) 1.00

LPI (overall) 0.85 1.00

DB Cost 0.51 0.41 1.00

DB Days 0.64 0.56 0.77

ETI (trans. inf) LPI (inf) DB (inland days)

ETI (trans. inf) 1.00

LPI (inf) 0.90 1.00

DB (inland days) 0.37 0.21 1.00

ETI (border) LPI (customs) DB (customs days)

ETI (border) 1

LPI (customs) 0.88 1

DB (customs days) 0.40 0.19 1

Table 5: Comparison between macoreconomic sources for overall indices (top panel),

transport infrastructure (middle) and border/customs (bottom).

The second panel compares measures of transit and infrastructure, where large correlation between the

ETI and LPI measures suggests the former is made up in large part by the latter. Otherwise, the

correlations are fairly low given that we think they are measuring similar things. The third panel

concerns customs/border processes. The correlation between LPI and ETI is again quite high but that

between the others is not. In particular, the correlation between the LPI perceptions measure and the

DB objective delays measure is only 0.192.

3.2 Microeconomic data (Enterprise Surveys)

Table 6 presents summary statistics of responses to the question on the degree to which customs

procedures and trade regulations are a constraint to the business.15

Recall we produced variables

representing the situation in each country. These are labeled in the columns, while the cross country

summary statistics are labeled in the rows. So, for example, the mean (across countries) of the survey

weighted means is 1.0. More generally, the averages in the first two rows are 1 or just over. The value of

1 corresponds to customs being a minor constraint. So, on average, customs are not perceived as a

major constraint. However, for the worst country, the sample median firm gave a value of 3.5, which

indicates the constraint is somewhere between major and severe. We will discuss measures of

dispersion at a later stage.

15Unless otherwise indicated, results are based on answers from exporting firms although the answers do not

differ materially for the broader set of firms.


14

ES Within country summary stats

Median Unweighted Mean Weighted Mean

Beteween

country

summary

stats

mean 1.3 1.4 1.0

median 1.0 1.4 1.0

min 0.0 0.3 0.2

max 3.5 3.1 2.4

Table 6: Summary statistics on perception of customs and trade regulations as a constraint to

business; note 0=no constraint, 1=minor,2=moderate,3=major,4=severe obstacle

Table 7 is analogous to Table 6 but presents an objective measure, namely the response to the question

on how many days it takes on average for exported goods to clear customs from the time they arrive at

port. The mean across countries of the survey weighted means is 6.81 days. The median across

countries is slightly lower at 5.55 days, which indicates a skewed distribution and/or possible outlier

countries. For example, the worst countries report averages of more than 20 days. Further, within

countries, the median (in column 1) is typically lower than the means (in the other columns). This may

be because many countries have a handful of firms reporting a large value.

ES Within country summary stats (days)

Median Unweighted Mean Weighted Mean

Beteween

country

summary

stats

mean 4.05 7.03 6.81

median 3.00 5.66 5.55

min 1.00 1.40 1.31

max 30.00 20.29 20.38

Table 7: Summary statistics on answer to question on the average number of days' for export

clearance

Firms were asked about the average number of days to clear customs, but were also asked about the

maximum number of days. A comparison between these two questions is presented in Table 8, where

the maximum clearance is as expected higher than the average clearance but typically less than twice as

high.

ES Within country average clearance ES Within country maximum clearance

Median Weighted Mean Median Weighted Mean

Beteween

country

summary

stats

mean 3.3 6.8 6.0 10.4

p50 3.0 5.0 5.3 9.2

min 1.0 1.0 1.3 2.4

max 12.5 30.0 16.5 33.2

Table 8: Comparison of answers to questions on average number of days' for export clearance with maximum number

of days; note 34 countries


15

We also note16

that the between country correlation between the mean responses to average and

maximum clearance days is 0.93 but the correlations between these two objective measures and the

subjective customs constraint measure were less than 0.3. Further, the correlations between statistics

calculated using only exporter responses and those using all firms were high. For the two objective

measures, these were in the mid to high nineties. For the perceptions question, some correlations were

as low as 0.8, but, as we noted, it is not obvious how to interpret a non exporter’s response to a

question on how export procedures constrain the business.

3.3 Comparing macroeconomic and microeconomic sources

This section compares the Enterprise Survey summary statistics with the appropriate macroeconomic

sources of data. Table 9 compares the responses to export clearance (from port arrival to customs

clearance) with the DB data. In principle, the most similar DB measure should be the one for clearance

of customs and ports. The cross country median (column 1) is 7 for Doing Business and 5.5 from the

micro data. The cross country means of 7.9 and 6.8 are insignificantly different at the 10% level. By

these measures, it appears that the two sources are comfortably close.

Cross country summary stats Pearson correl K Tau correl

median Mean~ Std. Err.~~ Mean* Median* Mean* Median*

ESMean* 5.5 6.8 0.5 . . . .

Median* 3 4.1 0.5 0.78 . 0.61 .

DoingBusiness Total (ES year) 24 30.2 1.9 0.33 0.31 0.19 0.21

Total (2010) 20 29.1 1.9 0.36 0.36 0.19 0.23

Ports 4 4.6 0.4 0.14 0.27 0.06 0.01

Customs 3 3.3 0.3 0.06 0.22 0.04 0.14

Customs/ ports 7 7.9 0.6 0.13 0.30 0.01 0.08

Table 9: Comparing microeconomic and macroeconomic measures of delays (in days) to clear exports. * refers to

question on the average days it takes for goods to clear customs from the point of arrival at the port and is either the

sample median or the population weighted mean. ~ mean is average across countries and ~~ std. error is of the

estimated mean. Total refers to the number of days it takes for goods to clear exports, including all steps, for 2010

(2010) and the year corresponding to the year of the Enterprise Survey for that country (ES year) respectively.

However, the correlations in the right half of the table indicate a different story. In the bottom row, we

see that the correlation between the within country means and the DB customs/ports delay is only 0.13.

The correlation between the within country medians and the same Doing Business customs/ports delay

is also low at 0.30. Because one is often concerned about the ranks of the countries, and for robustness,

16Results available on request


16

we also present the Kendal Tau rank correlation. By this measure, the correlation is even lower (or

negative). In fact, the correlation is insignificant.17

There are some higher Pearson correlations in the table, even though they in theory shouldn’t be. For

example, the 2010 total delay and the within country means/medians have a correlation of 0.36. While

it is not clear how high one might expect these to be, the comparison across answers to different

questions seen in the previous subsections yielded much higher correlations. On this basis, the

correlations in Table 9 are very low.

While Table 9 compared Doing Business with the ES responses to average clearance times, Table 10

compares the same DB data to the firms’ responses on maximum clearance times. The DB

customs/ports measure (last row) is still higher than the typical within country median responses (2nd

row), which is perhaps surprising given the latter is about maximum delays. However, it is

(insignificantly) lower than the typical within country mean responses (1strow). Although the within

country median responses to this question have a Pearson correlation as high as 0.61 with the DB

measures, this is not the case for the within country means or when using the rank correlation.

Cross country summary stats Pearson correl K Tau correl

median Mean Std. Err. Mean* Median* Mean* Median*

ESMean* 9.2 10.4 1.0 . . . .

Median* 5 6.8 0.9 0.77 . 0.64 .

DoingBusiness Total (ES year) 24 28.6 2.1 0.12 0.42 0.06 0.10

Total (2010) 20 26.2 2.1 0.19 0.54 0.07 0.09

Ports 4 5.1 0.7 0.23 0.61 0.05 0.07

Customs 3 3.4 0.2 0.12 0.10 0.11 0.01

Customs/ ports 7 8.5 0.9 0.16 0.54 0.12 0.02

Table 10: Comparing microeconomic and macroeconomic measures of delays (in days) to clear exports. * refers to

question on the maximum days it takes for goods to clear customs from the point of arrival at the port and is either

the sample median or the population weighted mean. ~ mean is average across countries and ~~ std. error is of the

estimated mean. Total refers to the number of days it takes for goods to clear exports, including all steps, for 2010

(2010) and the year corresponding to the year of the Enterprise Survey for that country (ES year) respectively.

To emphasise the point that the correlations are low and to gain insights into why the rank correlation

gives particularly low measures, Figure 2 presents four scatter plots. The x axis has the same DB

measure in all cases, but whether using the within country mean of the average clearance question (top

left), the within country median answer to that question (top right), the within country mean of the

maximum clearance question (bottom left) or the within country median answer to that question, there

is very little evidence of a systematic positive relationship.

What little positive correlation is picked up by the Pearson measure is being driven by a handful of

observations but the rank measure is not influenced in this way. The rank measure is not necessarily

17We also used the Spearman rank correlation and examined measures based on unweighted means, responses

from all firms (not just exporters), and so on.


17

superior; after all, it may be more important that both measures generally identify the countries with

extremely high delays. However, this is not done consistently either. In the bottom left panel, the

country with a microeconomic measure above 30 and a low DB measure is Bolivia. Two firms gave

answers in excess of 210 days but the summary statistic is based on a reasonable number (72) of firms.

In contrast, Angola has a low ES value (10.5) but a DB measure above 30 in the top right panel. Here, the

answer is based on only 5 responses and one more/less observation could have made the median 30.

Figure 2: Relationships between macroeconomic and microeconomic measures of customs/ports clearance.

For further comparison, we list and rank all the countries and the mean export clearance average days in

the appendix. By this measure, the best performing countries are Botswana and Namibia, who are both

part of the Southern African Customs Union. The worst performers are Mongolia and Tajikistan. The

appendix also includes the analogous Doing Business measures. Azerbaijan, Mongolia, Tajikistan and

Mauritius do well by one measure and poorly by the other. Micronesia, Angola, Venezuela, Republic of

Congo and Samoa are identified as being poor performers by both sources. The Baltic countries rank

well according to both measures.

Having compared objective measures (in days) across sources, Table 11 compares perceptions/index

measures in the Enterprise Surveys, ETI and the LPI. The bottom left presents Pearson correlations while

the rank correlations are in the top right. The microeconomic and ETI measures are relatively highly

05

10

15

20

(me

an

) expcla

vg


010

20

30

(p 5

0)

exp

cla

vg


010

20

30

40

(me

an

) expclm

ax


010

20

30

(p 5

0)

exp

clm

ax



18

correlated using either measure, especially if benchmarked against the correlation between the ESmean

and the ES median. The correlation with the LPI measure is lower. These should not be as comparable

with each other as those in the previous two tables, but one might still have expected a higher

correlation between measures of similar concepts. Overall, the overall picture for trade facilitation is

that the correlations are low.18

K Tau Correlations (top right)

Pearson

Correlations

(bottom

left)

ES mean* ES median* ETI LPI

ES mean* . 0.57 0.38 0.38

ES median* 0.75 . 0.37 0.37

ETI Border 0.56 0.60 . 0.53

LPI Customs 0.34 0.32 0.88 .

Table 11: Correlations between microeconomic and macroeconomic index

measures of customs/border clearance. * indicates within country summary

statistic of response to question on degree to which customs is a constraint to

the business (higher number implies greater constraint)..

3.4 Focus on Central Asia

To revisit some of the comparisons so far and with a view to the next subsection, we present the

statistics for individual countries. To compare within a region, we choose those in our dataset that are

located in Central Asia. This region is of particular interest to those working on trade facilitation. Its

geographical location means it tries to serve the European market but is sufficiently far for speed and

cost issues to be important. Further, because many countries are landlocked, border controls are

encountered often, so inefficiencies can accumulate.

Table 12 presents sample cross tabulations of firm level perceptions of the degree to which

customs/trade regulations constrain their business as well summary statistics. In total, the mean is 1.42

(bottom right), which places the average Central Asian firm between ‘minor’ and ‘moderate’ obstacle.

The equivalent measure across all non Central Asian firms is 1.43, so on average, this does not appear to

be a generally high constraint for the region.19

Armenia Azerb. Belarus Georgia Kazakh. Moldova Russia Tajik. Turkey Ukraine Uzbek. Total

no obstacle % 41 26 31 39 33 22 28 29 51 27 15 39

minor % 11 30 15 15 15 15 16 17 19 20 20 18

moderate % 31 22 23 17 30 23 23 13 14 22 46 20

18We briefly checked for other comparable criteria, like the import equivalent to the export measures and

measures of transit constraints or infrastructure quality, and found consistently low correlations.19The value of 1.43, which is a mean taken across all firms, is higher than the cross country mean of the firm level

means reported earlier, which had a value of 1.0.


19

major % 9 19 25 20 15 18 16 21 8 23 12 14

severe % 7 4 6 9 7 22 18 21 7 8 7 10

Firms # 54 27 65 46 27 60 193 24 583 173 41 1293

mean (0 4 range) 1.24 1.61 1.57 1.43 1.63 2.35 1.92 1.94 1.04 1.83 1.61 1.42

sd (0 4 range) 3.34 3.28 0.70 2.73 1.66 2.25 1.29 4.25 1.30 1.21 2.01 1.37

Table 12: Tabulations of firm perceptions of extent to which customs/trade regulations are an obstacle to their business. The (probability

weighted) total mean is across all firms in Central Asia, as is the standard deviation.

Moldova’s value of 2.35 places its average firm between ‘moderate’ and ‘major’, while Russia and

Tajikistan have values of close to 2. The standard deviation also hints at variation across firms within

countries. Given the bounded range of the answers, a cross tabular analysis is perhaps more

informative. In Moldova, Russia and Tajikistan, the firms are more or less uniformly distributed across

the levels of severity. These three countries distinguish themselves in being the only ones where more

than 10% of firms indicated the constraint is severe. Uzbekistan has a roughly symmetric distribution

with a clear mode at ‘moderate’. Turkey has a skewed distribution in which the proportion of firms

decreases at each level of severity. On average, Turkish firms are the least concerned about

customs/trade regulations.

Do the macro indicators lead to the same results? Table 13 presents measures from the Enabling Trade

Index and two editions of the Logistics Performance Index. Some indicators are missing but all three

measures place the Central Asian countries below the world average. While they are not supposed to be

measuring the exact same thing, this is at odds with the firm level summary. Perhaps the quality is

lower but many firms have adapted.

Armenia Azerb. Georgia Kazakh. Moldova Russia Tajik. Turkey Ukraine Uzbek. Asia World

ETI Border 3.25 2.91 . 2.27 3.59 2.82 2.4 4.05 3.07 . 3.05 4.03

LPI Customs '10 2.1 2.14 2.37 2.38 2.11 2.15 1.9 2.82 2.02 2.2 2.22 2.60

LPI Customs '07 2.1 2.23 . 1.91 2.14 1.94 1.91 3 2.22 1.94 2.15 2.55

Table 13: Customs / border quality as measured by ETI (1 7 index) and LPI (1 5 index).

Unlike Table 12, Moldova has a better ETI than the Central Asian average and the LPI measures put it in

the middle. For Russia, ETI puts it at third worst but this value is close to that for the middle countries,

while the LPI measure shows it was among the worst but improved. Tajikistan continues to be the worst

or among the worst in Asia, but recall its worldwide position depends on the data source. As for the

Enterprise Survey, Turkey is the best performer.

We turn our attention to export clearance days. Display 1 presents the answers given by exporting firms

to the question on how many days exports take to clear customs together with the summary statistics in


20

0 10 20 30Average Export Clearance(days)

Uzbekistan

Ukraine

Turkey

Tajikistan

Russia

Moldova

Kazakhstan

Georgia

Belarus

Azerbaijan

Armenia

excludes outside values

the accompanying table. Most countries have a number of statistical outliers, which may have an impact

on the country level means, but these are not presented in the box plots.20

The table identifies Kazakhstan as requiring over a month for goods to clear ports/customs while

Uzbekistan and Azerbaijan need almost two weeks. For reference, we note the world wide median is 7

days. The days are much lower according to the summary statistics of the firm level responses; the

means for these countries are 8.5, 5.1 and 1.9 days respectively. In reverse, the Tajikistan firm responses

have a mean of 20 days and a median of 15, while Doing Business reports only 4. Both Tajikistan and

Kazakhstan have median responses that lie well above the cross country median of this summary

statistic (3 days). These two countries also have mean responses that lie above the cross country mean

of this summary statistic.

The box plots indicate substantial variation within Kazakhstan and Tajikistan, which suggests that some

firms have very different customs experiences to others. The variation is quite large in a number of

countries. The interquartile range and standard deviation quantifies this. For example, Armenia’s

standard deviation is more than four times its mean while the multiple is less than unity for Belarus. This

20The width of the rectangular boxes gives the interquartile range, with the left and right sides giving the 25

thand

75thpercentile of answers. The vertical line inside the box gives the median (in Azerbaijan and Moldova, the

median is also the 25thpercentile). The length of the lines outside the rectangular boxes is determined by the

largest data point that is lower than 1.5 times the interquartile range beyond the 75thpercentile (to the right) or by

the smallest data point that is higher than 1.5 times the interquartile range beyond the 25thpercentile (to the left).

DB* Mean Median IQR SD

3 4.0 2 3 18

11 1.9 1 2 4

4 2.8 2 2 2

4 3.8 3 6 8

34 8.5 10.5 18.5 14

8 2.6 1 2 5

6 6.4 3 5 5

4 20.4 15 22 66

6 5.3 3 6 7

5 3.6 2 3 6

12 5.1 4 4 10

Display 1: Firm responses to average

clearance question (unweighted) and

summary statistics. * customs/ports from

Doing Business.


21

variation within countries motivates a comparison of the sources of variation within countries with the

variation between countries.

3.5 Within country variation vs between country variation

Table 14 focuses on the dispersion of various measures across countries. In the panel with the ‘days’

heading, we present two country level summary statistics based on the Enterprise Surveys, namely the

sample median and the population weighted mean across firms. While we have already reported the

cross country means of these summary statistics are 6.8 and 4.1, the cross country standard deviation

of the within country mean is 4.6. Relative to the cross country mean, the ratio is 0.7. For the closest

corresponding DB measure, customs/ports, the standard deviation is similar. Thus, while there is

variation across countries, it is relatively small compared to the mean. The interquartile ranges are also

similar and narrow.

Days Indices

Clearance ave. Doing Business Customs constraint ETI LPI

Mean Median Total Customs/ports Mean Median Border Customs

Interquartile range 5.2 3.5 16.0 4.0 0.6 1.0 1.5 0.9

Standard Deviation 4.6 4.3 16.8 5.0 0.4 0.9 1.1 0.6

Minimum 1.3 1.0 5.0 2.0 0.2 0.0 2.0 1.5

Maximum 20.4 30.0 102.0 34.0 2.4 3.5 6.5 4.0

Mean 6.8 4.1 24.4 7.8 1.0 1.3 4.0 2.6

Std. Dev / Mean 0.7 1.1 0.7 0.6 0.4 0.7 0.3 0.2

Table 14: Measures of between country variation in clearance days and perceptions. Note columns refer to variables, eg

mean is the country level summary of the firm level responses, while rows refer to calculated summary statistics across

countries, so standard deviation is the variation between countries. ES customs constraint range is 0 4 while ETI and LPI

ranges are 1 7 and 1 5 respectively. Std. Dev / Mean is calculated as ratio of the two rows of summary statistics.

The measures of dispersion for indices are not directly interpretable , but they can be compared to the

within country statistics based exclusively on the Enterprise Surveys. Table 15 presents various

measures of within country dispersion in columns and summarises these across the world in the rows.

For example, the country with the highest interquartile range has a range of 29 days while the median

country has a range of 5 days. We also note that the survey design leads to a large distinction between

the weighted and unweighted standard deviations, despite the latter also accounting for stratification.

On average, the within country standard deviation is substantially higher than the between country

standard deviation. The median country has an unweighted within country standard deviation that is

twice as high as the standard deviation of the median in the previous table. By other measures, the

discrepancy is even higher. For example, the mean ratio of the (weighted) standard deviation to the

mean is 2.8, which is four times as high as 0.7.


22

Similarly, for the perceptions responses, the within country (weighted) standard deviation is about ten

times the mean (Table 15) while the analogous between country ratio is less than one (Table 14). This

strongly suggests that within country variation is bigger than between country variation.

Days Customs constraint perception

IQR SD SD (weighted) Ratio IQR SD SD (weighted) Ratio

Mean 7.4 9.4 19.9 2.8 2.0 1.2 11.1 11.6

Median 5.0 8.2 13.7 2.5 2.0 1.3 9.4 11.1

Minimum 0.0 1.0 1.5 0.3 0.0 0.5 1.6 1.4

Maximum 29.0 33.4 97.7 8.6 4.0 1.8 37.6 37.0

Table 15: Measures of within country variation in clearance days and perceptions. Note columns refer to variables eg SD

(weighted) is the measure of the variation across firms within a country after accounting for survey design, while rows refer to

summary statistics of these variables across countries, so mean is the mean across countries. Ratio is of SD (weighted) to

mean calculated for each country.

For an alternative comparison between the two sources of variation, we ran a number of regressions to

see how much of the total variation across firms world wide can be explained by between country

variation. Table 16 presents the results for the average days it takes for exports to clear.

Constant Country dummies Country & year dummies

unweighted weighted unweighted weighted unweighted weighted

Root mean square error 11.186 10.702 10.634 9.305 10.634 9.3053

R2

0 0 0.1101 0.2556 0.1101 0.2556

p value (countries) . . 0.000 0.000 0.000 0.000

p value (years) . . . . 0.020 0.000

Table 16: Selected statistics from firm level regressions of export clearance days.

For reference, the first two columns present the root mean square error from pooled regressions of all

firm responses on a constant (unweighted and population weighted). Our main indicator of the

importance of between country variation is the R2, which explains how much variation is explained by

country dummies. The R2of 0.11 suggests that between country variation explains a small part of the

sample variation. Adjusting this for population weights raises the statistic to 0.2556. The country

dummies are jointly significant. In the final two specifications, we see that dummies included to

represent the different years in which the survey was conducted are also jointly significant. However,

they do not make any meaningful contribution to explaining the variation across firms.21

Therefore,

three quarters of the firm variation remains unexplained. This is consistent with the comparison of

standard deviations given earlier and implies that the within country variation is more substantial than

between country variation.

21Although significant, the coefficients did not indicate a trend relative to 2006. The macroeconomic data did

produce evidence of improvement over time.


23

Returning briefly to table 15, we note that there is variation across countries in the extent to which

there is within country variation. The country with the lowest dispersion in export clearance days has a

ratio of 0.3 while the maximum is 8.6.

4 DISCUSSION

Section 2 discussed and compared the construction of the various data sources including the firm level

responses from the Enterprise Surveys and the country level information from Doing Business, the

Logistics Performance Index and the Enabling Trade Index. Focusing on objective and perception or

index measures of trade facilitation, section 3 presented summary statistics from these sources and

section revealed that within country variation across firms is bigger than variation between countries.

Further, the correlation between sources is quite low – even for what appear to be responses to very

similar objective questions. The purpose of this section is to offer potential interpretations and

explanations for these related findings as well as their implications.

4.1 Sources of variation: interpretation and usefulness

With a view to expanding a country’s international trade, policy makers are typically interested in

reforming country wide trade facilitation (and broader institutions). However, we have seen the

variation in firm experience within countries is large. Further, some authors have demonstrated that a

relative low number of firms export (Bernard et al, 2007; Rankin, Soderbom & Teal, 2006).

Gravity models of trade typically focus on cross country or bilateral variation in trade and trade

facilitation.22

Some disaggregate between products but the question in mind is still typically at the

country level. For example, Djankov, Freund & Pham (2010) find that country level time delays (taken

from Doing Business) affect the ratio of time sensitive to time insensitive exports. A few use firm or

even firm product destination data within a country but exploit the cross importing country variation

experienced by each firm (Bernard et al, 2009; Lawless, 2008,9) to say something about macroeconomic

phenomena. Others examine across firm, within country variation in trade and trade facilitation.

Examples include Dollar et al (2006), Balchin & Edwards (2008) and Li & Wilson (2009).

Conceptually, cross firm variation is not necessarily the result of firm specific characteristics. For

example, that fact that one firm is close to the coast while another is far is clearly an attribute of the

firm. However, the difference in experience between firms can be small if the roads to the coast are

22Those studying trade facilitation include Clark et al (2004), Hoekman & Nicita (2009) and Wilson, Mann & Otsuki

(2005). Most forms are found to have a positive effect on trade.


24

good or if the main source of delay is getting off the dock onto a ship rather than getting to the dock.

So, reforms at the country level can have an impact on differences between firms.

Melitz (2003) builds a model where firms vary in the efficiency with which they produce things such that

only some are sufficiently productive to cover the additional costs of shipping goods overseas, so only

some export. This cost is the same for all firms, but a reduction in this cost means some firms

subsequently find it profitable to export. This introduces a distinction between the intensive and

extensive margin of trade, where the former refers to firms exporting more and the latter refers to more

firms becoming exporters.23Based on this framework, gravity models with macroeconomic data have

been used to show that improved logistics quality is associated with an increase along both margins

(Behar et al, 2009).

This framework does not explicitly accommodate variation in international transport costs across firms.

While the large variation observed within the Enterprise Surveys can be interesting, it is by no means

clear that the source is econometrically useful. Papers exploiting cross firm legislation are explicitly or

implicitly assuming that the ease with which they export goods is the result of an exogenous random

draw. Many would find this assumption difficult to accept.24After all, firms may choose to locate close

to the coast because they want to trade internationally and the experience of exporting can make them

better at dealing with the additional procedures required. As noted, the literature inspired by Melitz

(2003) has firms draw their productivity randomly from a distribution. If this applies to the efficiency

with which firms make things, why not the efficiency with which they move them? Leaving this potential

inconsistency aside, it is important for both the policy question and for the measurement exercises

conducted to inform that question that we understand what the variation actually means. We offer

some potentially complementary interpretations.

One interpretation, which we have already touched upon, is that the answers given in the firm level

surveys are due to different firm specific draws from the same distribution. Putting it crudely, firms who

drew too low a transport cost productivity do not find it sufficiently profitable to export while those who

drew a high productivity do. By this interpretation, variations between firms in their export levels and

their trade facilitation responses would be used to measure the extent to which easier export clearance

would raise exports. Of course, if the variation in trade facilitation is genuinely random, then policy

cannot help some firms be like others. However, national policy can affect the moments of the

distribution such that more firms can have sufficiently favourable draws. Certain types of firms could

benefit through help; speculative examples include information packs on how to deal with customs or

incentives to relocate.

23To be more precise, this refers to bilateral exports; that is, exports to a new destination. Additional margins can

operate within firms. For example, firms can expand the range of products they export to a destination as well as

the quantity of each product exported there (see Bernard et al, 2009).24Country level variation in trade facilitation is potentially endogenous to trade flows too. On the one hand, more

trade raises congestion. On the other, some trade infrastructure projects and reforms are only worth investing in

at already high trade levels. Nonetheless, the within country endogeneity problem is arguably more serious than

for macroeconomic data because firm characteristics are more malleable. For example, they can choose their

location more easily within a country than the country itself.


25

A second interpretation is that the differences are firm specific but almost entirely endogenous to

observable and unobservable firm characteristics. A refinement has firms draw from different

distributions depending on their features. Some may be observable, for example age, size or the

education level of the manager. Further within country analysis could investigate to what extent

variation in answers to trade facilitation correlates with other observables and thus produce conditional

distribution analogues to the unconditional means and standard deviations described here.

To the extent that some characteristics remain unobserved and that these characteristics are also

related to export success, econometric estimates using within country variation can be compromised.

This endogeneity problem has been noted earlier. Furthermore, without knowing what sources of

variation across firms are under the policy maker’s control, it is not clear what steps he could take to

improve firms’ trade facilitation experiences.

A third interpretation is that the variation within countries reflects the stochastic nature of the process.

An extreme form says that a draw from the distribution take place every time a set of goods is shipped,

that it is not specific to firms, and that the variation across firms merely reflects a limited number of

recent experiences. This source of variation is not very useful for measuring relationships between firm

level responses and firm level trade outcomes.

However, the interpretation alerts one to the possibility that there is uncertainty faced by firms. This

manifests itself as ex post uncertainty – how long will the shipment take this time? – but the first

interpretation produces uncertainty because firms may only discover their firm specific draw after they

have attempted to export.

If we view within country variation as an indicator of the uncertainty faced by firms in that country, then

the variation is of direct policy relevance. For example, Freund & Rocha (2010) find that African exports

are more responsive to transit delays than other sources of delay. Even though transit makes a small

contribution to the total on average, they argue this is because of the unpredictability of this

contribution. More generally, it is entirely plausible to expect uncertainty to have a negative impact on

exports. Uncertainty in delivery times requires delivery to be earlier, which uses up working capital of

both customer and seller, while early arrival of goods can place burdens on the buyer’s storage space.

To examine uncertainty more systematically, one could compare the relationship between each

country’s standard deviation of response times and its exports by adding this measure to a gravity

specification.25An arguably more direct measure of within firm uncertainty is the gap or ratio between

the answer to the question on average export clearance days and the answer to the question on

maximum export clearance days. Under certain exogeneity assumptions, this analysis could be

conducted across firms within countries as well as between countries.

25As suggested, not all this variation is uncertainty faced ex post by firms. One may then wish to more accurately

capture the uncertainty faced by firms by calculating the residual standard deviation (or standard error) after

conditioning on a number of firm level covariates.


26

If it’s true that uncertainty negatively affects exports, then steps could be taken to reduce it. Further

investigation might reveal the uncertainty is to do with random departures from scheduled opening

hours at the border post, which has obvious remedies, or to periodic road closures due to heavy rainfall,

which can be mitigated with tarred roads.

A fourth interpretation is that the variation across firms is noise or error. In other words, the accuracy

given by respondents is doubtful. While this might be uncertainty in the sense discussed above, we are

in this case referring to the respondent giving genuinely inaccurate answers of their experience and/or

not even knowing what their distribution looks like.26Insufficient knowledge may be of direct relevance

for export policy and may reveal ignorance to the prospect of exporting. From an econometric

viewpoint, measurement error attenuates estimates of any genuine within country relationship that

might exist between trade facilitation and exports. It can also affect the reliability of the summary

statistic for the country.

The next section includes a discussion of how the source of the variation affects the reliability of the

country level summary statistic and hence the correlations with the macro data sources.

4.2 Why do the sources have a low correlation and is one superior?

Our comparison between the macroeconomic and microeconomic measures revealed some low cross

country correlations between data sources. In part, this reflects the nature of the question. For example,

conditions may be bad in an objective sense, which is partly captured by indices like the LPI and its

components, but firms may have coped such that this does not affect their business, as noted in the ES

perceptions measure. Ironically, if prohibitive export processes cause a firm to focus on the domestic

market, then it could well say such processes are no constraint to the current operations of the firm.

Similarly, the low correlations could reflect the fact that different issues are being investigated, but this

is not entirely satisfactory for at least two reasons. First, while this may provide grounds for legitimate

variation, it is not useful when interpreting econometric results. For example, one typically includes an

index as an explanatory variable to proxy the truth about how easy it is to export. Various indices are

supposed to be alternative proxies of this truth. At a minimum, if is important to see if these underlying

proxies give the same message (or at least the same econometric results). Further, these alternative

proxies could be combined with common factor analysis to provide a potentially more accurate proxy.27

Similarly, the econometrician may include the number of documents required to clear exports as an

explanatory variable but should not necessarily interpret this literally as the effect of document

numbers on exports. A significant export documentation variable may have an obvious policy

26Purists may prefer the use of the label “risk” as opposed to “uncertainty” for the third interpretation.

27The LPI uses principal component analysis to summarise the variation of various distinct but correlated

components. The ETI aggregates across various measures to create potentially more accurate proxies but uses

simple averages rather than letting the implicit weightings be determined statistically.


27

implication – reduce the number of documents – but this may not be legitimate if it is approximating the

general ease with which one can export in the econometrics. Moreover, these literal interpretations can

be dangerous, with countries becoming liable to “reforming to the test”. That is, reducing the number of

documents required and advertising this on CNN while still leaving the underlying environment

unchanged.28

Second, correlations between separate components of the same data source – for example

infrastructure and customs indices within the LPI – were high. Tellingly, we were able to compare two

sources of answers to a precise question, namely average (or maximum) days it takes for goods to clear

customs from arrival at port from the Enterprise Surveys and the days it takes for goods to clear ports

and customs from Doing Business. These correlations were still low and certainly lower than those

between conceptually distinct components. The latter may be too high due to some form of halo effect

experienced by respondents, but the low correlations in absolute terms demand further interrogation.

Earlier, we noted that Doing Business is restricted to the largest city and large firms exporting more than

10% of exports. Although the Enterprise survey is broader in geographical coverage and we include all

exporters in our summary statistics, the sector coverage was limited for survey design reasons. To the

extent that sectors and geographical heterogeneity varies across countries, this may introduce

additional variation in the country level summary statistics. Further, as noted in the World Bank’s own

comparison between sources,29

the Enterprise Surveys are supposed to yield answers that are

representative of experiences of actual firms in that country, while Doing Business uses a case study to

investigate a hypothetical firm in a theoretical situation. This affects the inference drawn from the data.

For example, one country taking more days than another could reflect the fact that its goods are more

complicated to move or inspect. Therefore, some of the low correlation could be attributable to

differences across countries in the direction in and extent to which actual experience varies from

hypothetical.

For the firm level data, senior firm executives were interviewed. For Doing Business, a number of local

experts including lawyers in the country were asked. Although the case studies permitted the

hypothetical firm to avail itself of any means of speeding up the process and assumed it did so,

responses are to the de jure trade facilitation situation. However, individual firms may know “short

cuts”, for example bribes, that can make de facto delays shorter. To the extent that these short cuts are

reflected in the managers’ responses, this explains the generally higher country level averages found in

Doing Business than in the Enterprise Surveys. Furthermore, the low correlation can be due to

differences across countries in the availability of short cuts.

With varying degrees of precision, we have entertained the possibility that the issue, context or person

asked affects the actual question and hence answer. Putting it differently, the different data sources are

drawing from different distributions (or, loosely speaking, different parts of the distribution). In contrast,

they could be drawing from the same distribution but summarizing it with more or less reliability. In

28Moving from significant indices in a regression has the opposite problem because it is not clear what the actual

policy response should be.29Found at http://www.enterprisesurveys.org/Methodology/Compare.aspx


28

other words, who you ask can affect the quality of the answer. De jure answers might be unreliable

guides to de facto experiences. Moreover, logistics professionals may know more about logistics than

CEOs. However, the CEO is talking about his own firm’s actual experience while the logistics professional

is evaluating a number of foreign countries.

The firm level data includes some implausibly large values, which does not allay fears that the within

firm variation is due to noise and therefore unreliable. The Law of large Numbers implies that, given

enough firm responses, the mean will converge on the true expected value. So, even if the variation is

driven by ignorance (or uncertainty or a few lucky/unlucky instances), the average for the country will

be accurate if many firms respond. For some countries, the number of respondents to trade related

questions is small.30

While this need not imply a systematic discrepancy in the cross country mean

between sources, it might account for the low correlation between them.

Low numbers are by no means exclusive to selected Enterprise Survey data points. The macroeconomic

sources strive to consult as many people as possible, but the data is still based on the responses of

relatively few people. Each person’s response is implicitly an aggregation of a few or many experiences

(but recall Doing Business is based on a case study). However, we generally do not know whether the

answer for a country would be much different if a different set of experts had been asked.31Therefore,

especially when the Enterprise Survey yields responses from enough firms, it’s hard to know which data

source is more reliable or appropriate.

It is therefore imperative to check for robustness of results to different data sources. To the extent that

there are differences in the nature of the question and who answers it, this is important for interpreting

the answers. To the extent that this is due to unknown differences or reliability issues, this is important

from a pure statistical / data quality perspective.

30Earlier, we mentioned Angola, which had only 5 responses to a question and where there was a large

discrepancy between the data sources.31The Logistics Performance Index explicitly notes the stochastic nature of the responses and takes standard errors

into account (Arvis et al, 2010).


29

REFERENCESArvis, J, M Mustra, J Panzer, L Ojala & T Naula (2007), ‘Connecting to Compete: Trade logistics in the global

economy’, The World Bank

Arvis, J, M Mustra, L Ojala, B Shepherd & D Saslavsky (2010), ‘Connecting to Compete: Trade logistics in the global

economy’, The World Bank

Balchin, N & L Edwards (2008), ‘TRADE RELATED BUSINESS CLIMATE AND MANUFACTURING EXPORT

PERFORMANCE IN AFRICA: A Firm level Analysis’, Journal of Development Perspectives

Behar, A, P Manners & B Nelson (2009), ‘Exports and Logistics’, Oxford Department of Economics Discussion Paper

Behar & Venables (2010), ‘Transport Costs and International Trade’, Oxford Department of Economics Discussion

Paper

Bernard, A, J Jensen, S Redding & P Schott (2007), ‘Firms in International Trade’, Journal of Economic Perspectives,

21 (3), 105 130

Bernard, A, J Jensen, S Redding & P Schott (2009), ‘The Margins of International Trade: Long Version’

Clarke, X, D Dollar & A Micco (2004), ‘Port efficiency, maritime transport costs, and bilateral trade’, Journal of

Development Economics, 75, 417– 450

Djankov, S, C Freund & C Pham (2010), ‘Trading on Time’, Review of Economics and Statistics

Dollar, D, M Hallward Driemeier & T Mengistae (2006), ‘Investment Climate and International Integration’, World

Development

Freund, C & N Rocha (2010), ‘What Constrains Africa's Exports?’, World Bank Policy Research Working Paper

Helpman, E, M Melitz & Y Rubinstein (2008), ‘Estimating Trade Flows: Trading Partners and Trading Volumes’,

Quarterly Journal of Economics, 123 (2), 441 487

Hoekman, B & A Nicita (2008), ‘Trade Policy, Trade Costs, and Developing Country Trade’, World Bank Policy

Research Working Paper

Lawless, M (2008), Lawless, Martina (2008). ‘Deconstructing Gravity: Trade Costs and Extensive and Intensive

Margins’, CBFSAI Technical Paper

Lawless, M (2009), ‘Destinations of Irish Exports: A Gravity Model Approach’, Central Bank and Financial Services

Authority of Ireland Research Technical Paper

Lawrence, Blanke, Hanouz & Moavenzadeh (2008), ‘The Global Enabling Trade Report 2008’, World Economic

Forum

Lawrence, Hanouz & Moavenzadeh (2009), ‘The Global Enabling Trade Report 2009’, World Economic Forum

Li & Wilson (2009), ‘Trade Facilitation and Expanding the Benefits of Trade: Evidence from Firm Level Data’,

ARTNeT Working Paper

Melitz (2003), ‘The Impact of Trade on Intra industry Reallocations and Aggregate Industry Productivity’,

Econometrica

Rankin, N, M Söderbom & F Teal (2006), ‘Exporting from Manufacturing Firms in Sub Saharan Africa’, Journal of

African Economies

Wilson, J, C Mann & T Otsuki (2005), ‘Assessing the Benefits of Trade Facilitation: A Global Perspective’, The World

Economy, 841 871

World Bank (2009a), ‘ENTERPRISE SURVEY AND INDICATOR SURVEYS SAMPLING METHODOLGY’, August 29th, 2009

World Bank (2009b), ‘Doing Business 2010: Reforming Through Difficult Times’


30

APPENDIX: CUSTOMS/PORT CLEARANCE BY COUNTRY

This table presents the probability weighted means of responses to the Enterprise Survey question on

how many days exports take to clear customs from the point of arrival at port from best to worst. It also

presents the Doing Business statistics and the country ranking.

ES

rank Country

ES

days

DB

days

DB

rank

ES

rank Country

ES

days

DB

days

DB

rank

1 Botswana 1 7 41 43 GuineaBissau 6 8 49

2 Namibia 1 9 61 44 Nepal 6 8 49

3 Bosnia 1 7 41 45 Peru 6 8 49

4 Latvia 2 4 7 46 Panama 6 2 1

5 Azerbaijan 2 11 69 47 Czech Rep. 6 5 19

6 Serbia 2 7 41 48 Chile 6 6 28

7 Bhutan 2 9 61 49 Sierra Leone 6 8 49

8 Estonia 2 3 3 50 Honduras 6 5 19

9 Romania 2 4 7 51 Poland 6 2 1

10 Lithuania 3 4 7 52 Argentina 6 4 7

11 Niger 3 10 66 53 Russia 6 6 28

12 Moldova 3 8 49 54 Rwanda 7 8 49

13 Slovakia 3 6 28 55 Togo 7 5 19

14 Belarus 3 4 7 56 Ecuador 7 6 28

15 Albania 3 5 19 57 Benin 7 11 69

16 FYMOR 3 6 28 58 Colombia 7 5 19

17 Croatia 3 9 61 59 BurkinaFaso 7 6 28

18 ElSalvador 3 7 41 60 Ghana 7 7 41

19 Uruguay 3 6 28 61 LaoPDR 8 7 41

20 Ukraine 4 5 19 62 Kazakhstan 8 34 83

21 DRC 4 19 81 63 Senegal 9 5 19

22 Swaziland 4 8 49 64 Eritrea 10 14 77

23 Georgia 4 4 7 65 Malawi 10 6 28

24 Gabon 4 9 61 66 Philippines 10 5 19

25 Mauritania 4 15 79 67 Mozambique 10 6 28

26 Armenia 4 3 3 68 CapeVerde 10 10 66

27 Guinea 4 9 61 69 Mauritius 10 3 3

28 Bulgaria 4 6 28 70 Samoa 10 17 80

29 Uganda 4 10 66 71 Chad 12 6 28

30 Hungary 4 7 41 72 Congo 14 12 71

31 Burundi 4 8 49 73 Venezuela 14 12 71

32 SouthAfrica 5 13 74 74 Madagascar 14 4 7

33 Guatemala 5 4 7 75 Cameroon 15 7 41

34 Gambia 5 13 74 76 Bolivia 15 4 7

35 Slovenia 5 3 3 77 Kyrgyz Rep. 16 6 28

36 Nicaragua 5 13 74 78 Brazil 16 5 19

37 Uzbekistan 5 12 71 79 Angola 16 29 82

38 Tanzania 5 8 49 80 Cote d'Ivoire 17 8 49

39 Turkey 5 6 28 81 Micronesia 18 14 77

40 Lesotho 5 8 49 82 Mongolia 19 4 7

41 Paraguay 5 8 49 83 Tajikistan 20 4 7

42 Mexico 6 4 7

What can gravity models tell us about What can gravity models tell us about

logistics and exports?logistics and exports?

Alberto BeharInternational Monetary Fund

[email protected]

Alberto Behar 1What can gravity models tell us about

logistics and exports?

OutlineOutline

• Gravity vs tradeGravity vs trade

• Sources of transport costs

Ph i it i it• Physics gravity vs economics gravity

• Logistics – typical gravity interpretation

• Extensions: 3rd country effects, macro vs

micro

• Conclusion



GravityGravity

Force 1

Mass 1 Mass 2

Force 2



ExportsExports

Exports 1

GDP 1 GDP 2

Exports 2



Gravity equationGravity equation

Exports12 = 1*GDP1 + 2*GDP2 –“distance”Exports12 1 GDP1 2 GDP2 distance

Exports12 = 1*GDP1 + 2*GDP2 –“trade costs”

Exports12 = 1*GDP1 + 2*GDP2 –“transportExports12 1 GDP1 2 GDP2 transport costs; other trade costs”

{channels: money, time, complexity, { y, , p y,uncertainty}

Exports12 = 1*GDP1 + 2*GDP2 – *distance1212 1 1 2 2 12

- *signatures1 + *infrastructure1+ *FTA12 ...

{“reduced form”}{ }



Estimation (“econometrics”)Estimation ( econometrics )



Variations in transport costsVariations in transport costs

US$ per container Days

Region or

Economy Import Export Import Export

OECD Average 1146 1090 11 11

World Average 1625 1404 27 25

Si 439 456 3 5Singapore 439 456 3 5

Chad 6150 5497 100 75

Table 1, the costs of transporting goods - source: World Bank Doing Business

d h // d b / l / d d /Indicators; http://www.doingbusiness.org/ExploreTopics/TradingAcrossBorders/



Impediments to exportsImpediments to exports

• Natural/geographicalNatural/geographical

• Political (borders, free trade agreements)

T h l i l ( i f t t l i ti )• Technological (eg infrastructure, logistics)

Behar & Venables (2011), Anderson & van Wincoop (2004)



DENMARK BURUNDI

DAYS 5 67DAYS 5 67

DOCUMENTS 3 11

SIGNATURES 2 29

Source: World Bank Doing BusinessSource: World Bank Doing Business



Exports are not PhysicsExports are not Physics

• Firms choose to sell and people choose toFirms choose to sell and people choose to

buy– Exports = f(demand)= f(p/P)-1 =f(distance/remoteness,Exports f(demand) f(p/P) f(distance/remoteness,

relative logistics quality)-1

• Countries choose to improve logisticsp g

– Exports12 =

1*GDP1 + 2*GDP2 + *Logistics2 +(unobserved factor)

• “Endogeneity” / “reverse causality”

• “Does better infrastructure lead to more exports or do

higher trade volumes lead to more infrastructure?”higher trade volumes lead to more infrastructure?



Effects usually assumed (log) linearEffects usually assumed (log) linear

Exports12 = 1*GDP1 + 2*GDP2 –Exports12 = 1 GDP1 + 2 GDP2

*distance12 - *signatures1

+ *infrastructure + *FTA+ *infrastructure1+ *FTA12 ...

• What about “binding” constraints, supply-

h i l i ti i ti i t tichain analysis, optimization, interactions

between factors?



Exports and logisticsExports and logistics

Behar, Manners and Nelson (forthcoming)

• Gravity model using Logistics Performance Index (LPI) based on 6 sub-indicators (Arvis et al, 2007) - specifically 4 referring to international logistics “International Logistics Index”.

i Efficiency of the clearance process by customs and otheri. Efficiency of the clearance process by customs and other border agencies

ii. Transport and information technology infrastructure

iii Local logistics industry competenceiii. Local logistics industry competence

iv. Ease and affordability of international shipments

v. The facility to track and trace shipments

vi The timeliness with which shipments reach their destinationvi. The timeliness with which shipments reach their destination

(New and improved versions now available)



Log GDP 0.922***

Dependent variable: Log

bilateral exports

2

12

logistics

)exportsln(

Log GDP 0.922

[0.0157]

Logistics (ILI) 0.597***

[0.0556]

Log Distance -1.467***

1 standard deviation improvement in logistics (0.4)

would increase a country’s exports by:

exp(0.4*0.597)-1=27%

[0.0496]

Border 1.071***

[0.176]

Colony 0.501***

[0.133]

(typical interpretation).[0.133]

Language 0.646***

[0.0956]

Samecountry 0.398

[0.243]

l ***Religion 0.826***

[0.143]

Landlocked -0.648***

[0.0653]

Island -0.410***

[0.0694]

Constant -33.18***

[0.602]

N 6939

Si ifi l l * 0 1 **Significance levels: * p<0.1, **

p<0.05, *** p<0.01. Std errors in

brackets.



Third country effectsThird country effects• Most gravity models are used to assess what happens to exports

after a change in logistics (transport costs) between two after a change in logistics (transport costs) between two countries assuming no change by anybody else.

• But the trade between two countries depends not only on logistics of one country but on how this compares to logistics of one country but on how this compares to everybody else, so the effect of improved logistics must take this into account.

( ) d• Anderson & van Wincoop (2003): ignoring 3rd country effects can lead to a 20-fold overestimate of the effects of international border!

• Behar and Nelson (2012): if everybody reduces trade costs, trade falls between most country pairs. 3rd country effects reduce the impact on global trade by two thirds.pact o g oba t ade by two t ds.



The algebra of gravityThe algebra of gravity

1

1

12TYYM

11

1,

21122112

21

2112

pptyym

PPYYM

12

12

2

12

Logistics

t

Logistics

m

12

21

2

12

2

12

Logistics

pp

Logistics

t

dLogistics

dm

{...}

222

Alberto BeharWhat can gravity models tell us about

logistics and exports?15

Third country effects and logisticsThird country effects and logistics

,share GDPlogistics

)exportsln(2

2

12

countriesofnumber t coefficiengravity but

logistics2

Exporter Share Homogeneous Heterogeneous Exporter Share Homogeneous Heterogeneous

China 5.34% 3.50 3.93 Djibouti 0.00% 0.001 0.002

Brazil 1.90% 1.29 1.53 Gambia 0.00% 0.001 0.002

India 1.82% 1.24 1.45 Liberia 0.00% 0.001 0.001

Mexico 1.81% 1.23 1.35 Solomon 0.00% 0.001 0.001

Russia 0.99% 0.68 0.81 Guinea-Bissau 0.00% 0.000 0.001

Argentina 0.89% 0.61 0.72 Comoros 0.00% 0.000 0.001

Source: Behar, Manners & Nelson, Exports and International Logistics , Oxford Bulletin of Economics and Statistics



Trade costs and the margins of tradeTrade costs and the margins of trade

• Intensive margin: quantity of a product te s ve a g : qua t ty o a p oductexported (to a particular destination) by a firm

• Extensive margin: new products / firms g pexporting (to a particular destination)

• Empirical studies: – Gravity/macro data + theory (eg logistics)

– Firm-level data (Bernard et al)

l l d ( h )• Transactions level data (customs authorities): http://econ.worldbank.org/exporter-dynamics-

database (many countries not France )database (many countries, not France...)



ReferencesReferences

• Anderson, J. & van Wincoop, E. (2003), “Gravity with Gravitas: A Solution to

the Border Puzzle”, American Economic Review

• Anderson, J. & van Wincoop, E. (2004), “Trade Costs”, Journal of Economic

Literature

Alb t B h Phil M d B N l "E t d I t ti l• Alberto Behar, Phil Manners, and Ben Nelson. "Exports and International

Logistics" Oxford Bulletin of Economics and Statistics (forthcoming)

• Alberto Behar and Ben Nelson. 2012. "Trade Flows, Multilateral Resistance

and Firm Heterogeneity" F R E I T working paperand Firm Heterogeneity F.R.E.I.T working paper

• Alberto Behar and Anthony J. Venables. "Transport Costs and International

Trade" Handbook of Transport Economics. Ed. André de Palma, Robin

Lindsey, Emile Quinet & Roger Vickerman. Edward Elgar, 2011.

• Bernard, A., Jensen, J., Redding, S. & Schott, P. (2007), “Firms in

International Trade”, Journal of Economic Perspectives



ConclusionConclusion

• Economists study the impact of transportEconomists study the impact of transport

costs on trade using gravity models

• Countries with better logistics export more• Countries with better logistics export more

• But:

– This varies between countries

– Further work needed on micro data

• Log-linear specification is a major

simplificationp

Alberto BeharWhat can gravity models tell us about

logistics and exports?19

1 Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2012. Published by Blackwell Publishing Ltd,

9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 0305-9049doi: 10.1111/j.1468-0084.2012.00715.x

Exports and International LogisticsÅ

Alberto Behar†, Philip Manners‡ and Benjamin D. Nelson§

†International Monetary Fund, 700 19th Street, Washington, DC 20431, USA

(email: [email protected])

‡The Centre for International Economics, GPO Box 397, Sydney 2001, Australia


§Bank of England, Threadneedle Street, EC2R 8AH London, UK


Abstract

Better international logistics raise a developing country’s exports, but the magnitude of

the effect depends on the country’s size. We apply a gravity model that accounts for

firm heterogeneity and multilateral resistance to an international logistics index. A one-

standard deviation improvement in logistics is equivalent to a 14% reduction in distance.

An average-sized developing country would raise exports by approximately 36%. Most of

the countries are much smaller than average, so the typical effect is 8%. This difference

is chiefly due to the dampening effect of multilateral resistance, which is more important

for small countries.

I. Introduction

Integration into the world economy is widely viewed as one of the key factors under

lying the success of the fastest growing economies (Growth Commission, 2008), yet many

developing countries remain isolated. This manifests itself in the form of low international

trade, and high trade costs can be an important factor.Although tariffs on industrial products

have generally declined, non-tariff barriers remain. One example is the cost of transport-

ing products to foreign markets, both in pecuniary terms and in the form of delays (Behar

and Venables, 2011). For this reason, multilateral and donor organizations have sought to

view ‘aid-for-trade’ packages as a promising new developmental tool (Huchet-Bourdon,

Lipchitz and Rousson, 2009).

Does such assistance work? Economists have studied the potential impact of transport

and other costs on trade using gravity since at least Tinbergen (1962).1 This paper makes a

ÅPart of this work was conducted when the authors were at the Department of Economics, University of Oxford,and while Behar was at the Development Economics Research Group at the World Bank. The authors thank PeterNeary, Caroline Freund, Tony Venables, Alberto Portugal-Perez, Luis Serven and Adrian Wood. Please note: theviews expressed in this paper are those of the authors and not necessarily those of the Bank of England, the MonetaryPolicy Committee, or the Financial Policy Committee, and should not be attributed to the IMF, its Executive Board,or its management.

JEL Classification numbers: F10, F13, F14, F17, O24.1See Clark , Dollar and Micco (2004); Limão and Venables (2001); Wilson, Mann and Otsuki (2005); Djankov,

Freund and Pham (2010).

2 Bulletin

substantive contribution to our understanding of the importance of logistics for developing

countries because we use a newWorld Bank index, which draws on a wide range of crite-

ria, has broad country coverage from a single source, and is based on detailed evaluations

provided by logistics professionals (Arvis et al., 2007).

This paper also makes a methodological contribution to the estimation and interpreta-

tion of gravity models and hence our understanding of the importance of transport costs for

trade.We derive and estimate a newmodel that shows standard approaches would produce

an almost three-fold exaggeration of the typical impact of such factors for developing

countries. We uncover this dramatic exaggeration because our novel approach accounts

for two issues, namely multilateral resistance and firm heterogeneity.

Regarding the first issue, Anderson and vanWincoop (2003) (AvW) show that it is not

just bilateral trade costs, but those costs relative to multilateral trade costs that are relevant

for predicting bilateral trade flows. In particular, imports by country i from country j, Mij,

are an increasing function f (·) of, inter alia, bilateral trade costs tij relative to the product

of the two countries’ aggregate price terms Pi×Pj, such that

Mij = f

(Pi×Pj

tij

). (1)

AvW (2003) call the price terms multilateral resistance (MR) because they work to

aggregate trade costs across the two countries’multiple trading partners. Omitting controls

forMR can lead to biased coefficient estimates andmisleading comparative static estimates

of the impact of changes in trade barriers on trade flows. This is because changes in trade

costs affect both the denominator and the numerator of the argument on the right-hand side

of equation (1); empirical studies typically ignore the latter. Economically, for the exporter

j, it is the trade cost associated with exporting to i relative to those trade costs incurred

when trading with all other traders that matters for its exports to i.

Trade responses to trade costs are approximately proportional to country size because

bigger countries are less affected by MR. Intuitively, since larger countries typically trade

a smaller fraction of their total output internationally, a change in international trade costs

affects a proportionately smaller subset of their total production. Accordingly, their MR

changes by less and has a smaller dampening effect, such that the overall elasticity net

of MR is larger. Conversely, smaller countries will have smaller elasticities net of MR.

Given the skewness in the world’s distribution of country size, most countries are small.

Therefore, standard estimates overstate the impact of changes in logistics on trade for

most countries. Conversely, for a handful of large developing countries, the impact is

underestimated.

Firm heterogeneity is addressed by Helpman, Melitz and Rubinstein (2008) (HMR),

who develop a method to account for the consequences of heterogeneous firm produc-

tivity in gravity models. Firm heterogeneity gives rise to two margins of adjustment to

changes in trade barriers: the intensive margin, which captures exports per firm, and the

extensive margin, which captures the number of exporting firms. Ignoring the effects of

trade costs on firm entry results in misleading estimation and country-level comparative

statics.

Behar and Nelson (2012) develop a model which accommodates both MR and firm

heterogeneity. They demonstrate the importance of these effects for comparative statics

Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2012

Exports and international logistics 3

when trade costs are captured by bilateral distance. In this paper, we adapt that

approach to the case of logistics. Unlike distance, logistics might not be exogenous, so

we take potential endogeneity seriously. Further, unlike bilateral distance, the logistics

index is a country-specific variable which precludes the use of fixed effects to control for

MR in estimation. Instead, we proxy MR terms using an adaptation of Baier and Berg-

strand’s (2009) Taylor approximation method. Our approach allows us to implement this

method together with HMR’s procedure to account for firm heterogeneity.While the appli-

cation in this paper is to logistics, the implications fall on a wide class of country-specific

international trade costs.

Section II provides an overview of the existing literature on logistics and international

trade before expanding on the importance of MR and the selection issues associated with

firm heterogeneity. Section III describes the data. The World Bank constructed its Logis-

tics Performance Index (LPI) using six indicators. We describe those indicators and how

we extract only those that are relevant to international trade to produce what we call the

International Logistics Index (ILI).

Section IV formalizes our gravity modelling framework, which accounts for both firm

heterogeneity and MR, and derives the Taylor approximation for country-specific trade

costs in this context. In deriving the full comparative statics, this section illuminates how

the trade impact varies with country size and reveals that the coefficient on logistics can

be interpreted as the effect for a country of average size. Section V discusses estimation

issues. Because trade flows may affect investments in logistics, it may be an endoge-

nous regressor, so we propose our instrumentation strategy based on business start-up

procedures. We recast the issues of MR and firm heterogeneity as omitted variable bias.

While the homogeneous firms model can be estimated by OLS as in Baier and Bergstrand

(2009, 2010) or two-stage-least-squares, we also adapt the HMR two-step procedure to

estimate the heterogeneous firm model.

In section VI, the benchmark linear specification suggests a one-standard deviation

improvement in logistics quality, which would put Rwanda on a par with Tanzania, raises

exports 27%. We use our homogeneous firms model to suggest that neither endogeneity

nor MR are materially biasing the logistics coefficient. Consistent with HMR, account-

ing for firm heterogeneity using the two-stage procedure produces bigger country-level

comparative statics for an average-size country than does the homogeneous goods model

estimated by OLS.

The estimates imply a one-standard deviation rise in the index is equivalent to a reduc-

tion in distance of about 14%, while our simulations in section VII indicate it would raise

exports by about 36% for an average-size country. Since the impact of MR varies by

country size, we compute the elasticity for each of our exporters. Rwanda’s trade response

would be 1% of the response implied by the benchmark specification while Brazil’s would

be three times as big. Averaging over all exporters, the typical effect is only 8%, which is

about one-fifth of the average-size country effect because most countries are small. The

linear benchmark would exaggerate this almost three-fold. So, while the impact of MR

on estimation may be small, it has a first-order effect on comparative statics. Section VIII

concludes that small countries have much smaller trade responses than the average, but

cautions against interpreting these results as a weak case for logistics upgrades in those

countries.


4 Bulletin

II. Literature

This section reviews two important methodological advances and briefly discuss empirical

work on transport costs in this context.

Methodological concepts

Our approach accounts for two important insights provided by the recent gravity literature

on trade flow estimation. In particular, our estimation and comparative static exercises

account for both MR and firm heterogeneity.

Multilateral resistance

AvW (2003) show that it is essential to account for the general equilibrium effects of

changes in trade costs if the trade elasticity ∂ lnMij/∂ ln tij is to be calculated correctly.

General equilibrium effects work through the MR terms Pi ×Pj that enter the bilateral

gravity equation, as illustrated in equation (1). The bilateral trade flow between two coun-

tries depends not only on the bilateral trade barrier between them, but the severity of this

barrier relative to those confronted when the two countries trade with others (including

domestic trade),tij

Pi×Pj. It follows that the overall impact of a change in a trade barrier must

account for these potentially significant ‘third party’effects. The impact on Brazil’s exports

to Peru of signing a trade agreement depends also on whether other countries are party

to that agreement. Were the agreement bilateral only, a reduction in the Brazil–Peru trade

barrier would stimulate Brazilian exports to Peru, potentially reducing exports to third

parties (e.g. Uruguay) and to itself. Were the agreement to include Uruguay, the relevant

cost for Brazilian exports to Peru is the new cost of exporting to Peru relative to that of

exporting to Uruguay; in relative terms, these costs have not changed. In this case, the

only change in relative trade costs is that between domestic and international trade, with

the costs associated with the latter falling relative to the former.

These effects are shown to be quantitatively important byAvW(2003) in explaining the

so-called US-Canada ‘border puzzle’ of McCallum (1995), and Behar and Nelson (2012)

show that the effects of MR are large for multilateral changes in trade costs. Changes in

a given country’s logistics quality share some of this multilateral characteristic: if Kenya

were to achieve an improvement in its logistics, its relative trade barrier across all export

destinations would be affected, and our comparative statics on exports to a particular desti-

nation must reflect this. Put differently, since the gravity equation for bilateral trade flows

is derived from a general equilibrium system, any statement about the likely impact of

this change on bilateral trade flows must take general equilibrium effects through MR into

account. Without doing so, comparative statics exercises will generally overestimate the

true magnitude of the response of bilateral trade flows to a change in trade costs.

Furthermore, larger countries typically trade a larger fraction of their output domesti-

cally for a given international trade cost. For large countries, a smaller proportion of their

total (domestic plus international) trade is affected by changes in international trade costs

with all destinations, which are captured by changes inMR.As a result, MR effects are less



important and hence trade elasticities are greater for larger countries. This is ‘Implication 1’

in AvW (2003).2

There are a number of empirical approaches to controlling for MR in estimation. First,

since MR terms are country-specific, they can be controlled for by including country fixed

effects. This is not appropriate for our purposes as we want to identify the effect of coun-

try-specific logistics quality. Second, some attempts have been made to control for MR

by using price data to construct the appropriate price indices. However, published price

indices are typically stated relative to an arbitrary base period, which makes comparison of

levels impossible (Feenstra, 2004). Furthermore, they tend to include too many non-trade-

able goods and thus fail to capture the additional costs embedded in internationally traded

goods (Anderson and van Wincoop, 2004). Third, AvW (2003) propose solving for the

system of prices together with the gravity equation, but this involves a potentially prob-

lematic customized nonlinear program. The research community has shown that, while

gravity models are popular in empirical trade, estimating the entire nonlinear system is

not.3 Fourth, Baier and Bergstrand (2009) introduce a method by which the MR terms

are approximated using a first-order Taylor expansion, yielding a log-linear expression for

MR which contains exogenous variables only. The MR terms can then be included in a

single linear equation. This approach has the advantage of yielding tractable comparative

statics; in particular, the role of country size in determining the appropriate comparative

static effect is made explicit, while Baier and Bergstrand show that the approximation error

associated with this method is small for the majority of country pairs.

Firm heterogeneity

Models of monopolistic competition with firms of heterogeneous productivity predict

selection into export markets in the presence of fixed costs of trade. The reason is that

the least productive firms do not generate sufficient profits to cover the fixed costs. As

illustrated by HMR, a change in a trade barrier affects both the amount a given firm

exports, the intensive margin, and the number of firms that export, the extensive margin.

Traditional gravity equations conflate these two effects, whereas HMR’smethod allows for

their decomposition. Moreover, for high fixed costs, no firms in a given country may find it

profitable to export to a particular destination. This provides an explanation for the ‘zeros’

observed in bilateral trade data indicating that many country-pairs do not appear to trade.

This is a country selection effect, and also induces bias in traditional gravity estimates.

HMR propose a two-stage estimation procedure to construct controls for both firm and

country selection. In the procedure described below, we use the elements of their approach

that allow us to control for the effects of firm heterogeneity and simultaneously account

for MR using the Baier and Bergstrand (2009, 2010) approximation in both estimation and

comparative statics.

2

AvW’s (2003) Implication 1 states that ‘trade barriers reduce size-adjusted trade between large countries morethan between small countries’. So small countries experience smaller trade elasticitieswith respect to uniform changesin trade barriers. The reason, as AvW (2003) state, is that ‘a uniform increase in trade barriers raises multilateralresistance more for a small country than a large country’ (p.177).

3

Furthermore, this method is especially demanding when considering about 100 countries and allowing for asym-metries in trade frictions. Bergstrand, Egger and Larch (2007) have shown that the system solution can yield complexnumbers.


6 Bulletin

Studies on transport costs

Behar and Venables (2011) summarize the literature on the determinants of transport costs

and their consequent impact on trade. These determinants include geography, hard infra-

structure and procedural/institutional characteristics of a country. For example, Limao and

Venables (2001) map information on road, rail and phone infrastructure to shipping cost

information and calculate that variation in infrastructure accounts for 40% of variation in

transport costs. In a gravity framework, they find that a country improving its infrastructure

from the median to the 75th percentile would increase its trade by 68%.

Clark et al. (2004) find that a deterioration in port facilities and general infrastruc-

ture from the 25th to 75th percentile is associated with a 12% rise in ocean freight costs.

They find that these costs, which are based on containerization, the regulatory environ-

ment, seaport infrastructure and other variables also materially impact trade. Nordas and

Piermartini (2004) adopt a similar approach to Limao and Venables but use more infra-

structure measures. They have separate specifications for a number of indicators – airports,

roads, telephone lines, port efficiency and the median port clearance time – which are esti-

mated separately. They find all components are significant determinants of trade, with port

efficiency being the most influential.

Moving beyond infrastructure, Djankov et al. (2010) calculate that a transit delay of

one day reduced trade by 1%, which is equivalent to an additional bilateral distance of

about 70 km. Hummels (2001) finds that improvements in customs clearance sufficient

to reduce waiting times by a day would be equivalent to a 0.8% reduction in ad valorem

tariffs.

These papers make important contributions to our understanding of the relationship

between transport costs and trade flows. A number control for MR in estimation indirectly

through the use of fixed effects. However, they do not explicitly control for MR or firm

heterogeneity4 in estimation and comparative statics. As suggested by the methodologi-

cal discussion in this section, this means the effects of reforms on trade can be severely

miscalculated.

III. Data

The 2007 Logistics Performance Index (LPI) is sourced from the World Bank and is con-

structed on a scale from 1–5. The LPI is calculated by the World Bank using a Principal

Components Analysis of six different sub-indicators drawn from the same source. The

indicators (with their weights in the LPI in brackets5) are: (i) Efficiency of the clearance

process by customs and other border agencies (0.18), (ii) ease and affordability of inter-

national shipments (0.20), (iii) the facility to track and trace shipments (0.16), (iv) the

4

Drawing on an earlier version of our paper, Portugal-Perez and Wilson (2010) follow our estimation approachbut do not fully explore the comparative static implications.

5

The weights that are used are not reported by the World Bank but can be backed out. For each country i, wehave Li =wCi where Li is i’s LPI score, Ci is a [6x1] vector of i’s component scores, and w is a [1x6] vector of theweights. We can then use six different country LPI scores to form a [6x1] matrix L in which row i corresponds tocountry i’s LPI score, Li , together with each country’s component scores to form a [6x6] matrix C, in which columni corresponds to country i’s vector of scores Ci . Then we solve L= wC for w= C−1L to obtain the weights. We dothis for a number of different sets of countries to ensure that the weights we calculate are unaffected by roundingerrors.



TABLE 1

Summary statistics for logistics indices

Statistics LPI ILI

High income Mean 3.74 3.84

SD 0.34 0.32

Middle and low income Mean 2.49 2.61

SD 0.39 0.40

Notes: LPI isWorld Bank Index. ILI is our own sum-marymeasure of the components affecting internationallogistics.

Source: World Bank and authors’ calculations.

timeliness with which shipments reach their destination (0.15), transport and information

technology infrastructure (0.15), and (vi) local logistics industry competence (0.16).

Further details of the construction of each indicator are available inArvis et al. (2007).

In summary, the index is based on more than 5,000 country evaluations by logistics pro-

fessionals. The perceptions-based measure is corroborated with a variety of qualitative

and quantitative indicators. They calculate that, on average, a one-point rise in the LPI

corresponds to exports taking three more days to travel from the warehouse to port.

Efficiency of the clearance process by customs and other border agencies, the ease and

affordability of arranging international shipments, the ability to track and trace those ship-

ments as well as the speed with which they reach their destinations are directly relevant

to international trade. Transport and IT infrastructure are relevant to all trade, whether

international or domestic, as is the competence of the local logistics industry.

Costs given by tij (i /= j) reflect international trade cost factors relative to trading within

borders. As a result, it is conceptually correct to measure those aspects of logistics which

affect international trade costs and are relevant for cross-border trade. Strictly speaking,

measures that affect both internal and international costs equally have no impact on exports

in fully-specified gravity models. We could use each of the relevant components listed,

but separate treatment would lead to multicollinearity because they are highly correlated.

Therefore, we construct our own International Logistics Index (ILI) based on components

i–iv using the weights listed. A regression of the LPI on the ILI has an R2 of more than

0.99, so the international logistics index explains a large proportion of the overall index.

Therefore, while the ILI is conceptually more appropriate, using it makes little practical

difference.6

Table 1 presents summary statistics for the full LPI and for our new International Logis-

tics Index (ILI). High income countries have measures of quality that are of the order of

three standard deviations higher. Approximately 70% of developing countries have an ILI

of between 2 and 3, but there is still considerable variation. Appendix Table A.1 lists the

values for all countries.

Our interest is exports from developing countries. Based on the availability of logistics

data and possible instruments and controls, our sample consists of 88 low- and middle-

income exporters. We have 116 importers regardless of income classification.7 We use

6

The 2009 version of this paper uses the full LPI.


8 Bulletin

merchandise exports data for 2005 from the IMF Direction of Trade Statistics and we

observe 7,246 positive cross-border trade flows, 2,548 zeros and 826 missing values. We

therefore need to account for potential sample selection issues in our model and econo-

metric methodology. We use 2005 GDP measured in constant (2000) US Dollars from

the World Development Indicators. As will become clear, each country’s share of world

GDP is an important component of our analysis, so we calculate each country’s share by

dividing its GDP by the sum of the 116 GDPs in our sample.

Themeasure of bilateral distance that we use captures the internal distance in a country,

accounts for the distance from a number of major cities and is constructed by the Centre

d’Etudes Prospectives et d’Informations (CEPII). 8 Our control variables, including dum-

mies for whether or not two countries share a border or language,9 a common colonizer or

were once the same country, as well as a dummy for being landlocked, are also fromCEPII.

IV. Theory

We model the relationship between exports and logistics using a gravity equation. This

equation has a successful history in explaining bilateral trade patterns while theory has

subsequently provided grounding (Anderson, 1979; Bergstrand, 1985). The importance of

MR was highlighted in AvW (2003, 2004) and that of firm heterogeneity by HMR. We

present the full heterogeneous firms model before showing how it can be understood as a

generalization of its homogeneous firms counterpart.

The model

There are J countries, indexed by =1, ..., J . Within each country are monopolistically

competitive firms which produce a continuum of differentiated products. Consumers have

CES preferences given by

uj =

[∫

i∈B

xj(i)a di

] 1a

, (2)

where x(i) is consumption of variety i, contained in the set of varieties available in j,

B. Let r≡ 1/ (1− a) be the elasticity of substitution. In this endowment10 economy with

exogenous income in j of Yj, firms face demand of

xj(i)=Yj

P1−rj

pj(i)−r, (3)

where pj(i) is the price of variety i in j and Pj is j’s ideal price index, given by

Pj =[∫

i∈Bpj(i)

1−r di] 1

1−r .

7

Further discussion of why we drop rich exporters is postponed until Appendix B.8

The distance measure used is distw and is described at http://www.cepii.fr/anglaisgraph/bdd/distances.htm9

We construct a dummy that is equal to one if two countries share either a common official or common ethniclanguage.

10

As inAvW (2003, 2004) and most of the gravity literature, we do not allow for endogenous or excess capacity inany economy. This precludes the use of previously idle resources for exports and precludes vent-for-surplus exports,where goods are exported as a result of insufficient domestic demand.



Each country produces a number of varieties of measure one, with one variety per firm.

The firm-specific unit cost of production is a as in Melitz (2003). Firms draw a indepen-

dently from the identical distribution function G(a) with support [aL,aH ] such that aL is the

lower bound on possible unit input requirement draws and aH is the upper bound. We can

identify a firm’s variety with its cost draw a: though there may be a measure of varieties

with the same cost, each variety with a given cost draw behaves symmetrically, such that

they can be indexed by a alone.

Firms face factor costs cj and two cost of exporting. The first is an ‘iceberg’ variable

trade cost tij >1, which we will specify further later. The second is a fixed cost of exporting

fij >0, fii =0. Taken together, a firm in j exporting to i producing qij units of output has a

cost function given by

Cij(a)=acjtijqij + fij. (4)

Given demand and costs, each firm chooses price so as to maximize its profits. This

gives the price, demand and profit function for a firm exporting from j to i as

pij(a)=cjtija

a, (5)

xij(a)=Yi

P1−ri

pij(a)−r (6)

pij(a) =(1−a)

[cjtija

aPi

]1−r

Yi− fij. (7)

Sales by firms located in country j are only profitable in country i if pij(a)>0. Define

a productivity cut-off aij by pij(aij)=0, which is the cost level (or inverse productivity

level) below which it is profitable to export. Firms with a > aij do not generate profits high

enough to cover the fixed costs of exporting fij. Using an exporting firm’s profit function

above then gives the cut-off as

aij =

[Yi(1−a)

fij

] 1r−1 aPi

cjtij

. (8)

When aij is higher, the extensivemargin is greater, implying a larger subset of firms exports.

It rises when fixed or variable trade costs fall. Whenever aij < aH , there will be firm selec-

tion into exporting. The total value of imports by importer i from exporter j is given by

Mij =∫ aij

aLpijxij dG(a). Substituting in for prices and quantities, we obtain

Mij =

[cjtij

aPi

]1−r

Yi

∫ aij

aL

a1−r dG(a). (9)

We then define Vij ≡∫ aij

aLa1−r dG(a) as a term capturing the firm selection effect. As aij

rises, Vij rises. In other words, as the cost level above which firms find it unprofitable to

export rises, more firms export. Using this, we have bilateral exports from j to i given by

Mij =

[cjtij

aPi

]1−r

YiVij. (10)


10 Bulletin

Vij in equation (10), which is the same as in HMR, forms the basis for accounting for firm

heterogeneity. In equation (10), P1−ri =

∑j

aij∫aL

pij(a)1−r dG(a) and can be written as:

P1−ri

=∑

j

(cjtij/a)1−rVij. (11)

General equilibrium

Assuming trade balance will allow us to write an AvW (2003) style gravity equation for

bilateral exports. Trade balance requires Yj =∑

i Mij, so summing both sides of equation

(10) yieldsYj

∑i

[tij

Pi

]1−r

YiVij

=(

cj

a

)1−r. Using this in equation (10) and defining Y ≡

∑h Yh as

total income allows one to write11

Mij =YiYj

Y

(tij

PiPj

)1−r

Vij (12)

P1−ri

=∑

j

(tij

Pj

)1−r

sjVij (13)

P1−rj

=∑

i

(tij

Pi

)1−r

siVij. (14)

In this system, the sk terms represent country k’s GDP as a share of the total income

of all the countries. That is, si≡Yi/Y is country i’s GDP as a share of total income. The

system (12)–(14) resembles that of AvW (2003), but with the crucial difference that it

allows for firm heterogeneity. Reductions in trade costs affect both the numerator and the

denominator of the gravity equation. Because a reduction in tij affects the MR terms, the

resulting increase in bilateral trade will be smaller than in the absence of changes in MR,

all else being equal.

A further implication of imposing trade balance is that equation (8) for the extensive

margin also takes a gravity-like form. Drawing on Behar and Nelson (2012),

ar−1ij

=(1−a

) YiYj

Y

1

fij

(tij

PiPj

)1−r

, (15)

which is a gravity equation for the cost cut-off defining the extent of the extensive margin.

Just as for bilateral exports, it responds positively to the product of the trading countries’

GDPs, negatively to bilateral trade costs, and positively to MR, captured by the PiPj term.

Note also that the fixed trade cost fij enters equation (15), such that higher fixed costs

reduce the cost level below which exporting is profitable. In this way, fixed costs affect

the number of exporting firms, but not how much each exports. In other words, fixed costs

only affect the volume of bilateral exports through their impact on aij, which determines

Vij. This is important for the identification strategy in the empirical section.

11

We have abstracted from differences in the sets of active traders across countries. See Behar and Nelson (2012)for a discussion.



Further, equation (15) makes explicit the role of MR on the extensive margin. Just as

for bilateral trade flows, a multilateral increase in trade costs increases both the numera-

tor and the denominator of equation (15); the effect of trade costs on the price terms in

the denominator therefore acts to mitigate the direct effect in the numerator. Just as AvW

(2003) show for bilateral trade flows, comparative statics on the extensive margin will be

misleading when ignoring MR.

Imposing an assumption about the distribution of productivities grants us further ana-

lytical tractability. Following much of the recent trade literature, we impose a Pareto

distribution on firm specific variable costs12 such that

1

a∼Pareto(k), a∈ [aL, aH ],

where k is the shape parameter. This implies G(a)=ak−ak

L

akH−ak

L

, and g(a)= k ak

akH−ak

L

1

aso that we

can write the extensive margin as

Vij =max

{k

akH −ak

L

[(aij

aL

)k−r+1

−1

], 0

}, (16)

such thatwhenever aij < aL,Vij =0. Fromequation (12), this generates zero bilateral exports

from j to i when aij < aL. Following HMR, one way to operationalize equation (16) is to

consider the profits of the firm in j with the lowest variable costs aL. If this firm does not

find it profitable to export to i, then no firm in j will. For firm aL, the ratio of variable profits

to fixed export costs can be written as

Zij≡Yi(1−a)

fij

(tijaL

aPi

)1−r

, (17)

and it follows that

Vij >0 iff Zij >1. (18)

Zij is HMR’s latent variable, which under the trade balance assumption can be written as

(Behar and Nelson, 2012)

Zij = Zij

(PiPj

)r−1, (19)

where

Zij≡YiYj

Y

(1−a)(tijaL)1−r

fij

. (20)

That is, the latent variable can be decomposed into two components: MR, and a component

Zij independent of prices. Next, using equations (15), (19) and (20) allows us to relate the

extensive margin cut off and the latent variable according to

12

This distribution is commonly used in theoretical and empirical applications, for example Chaney (2008) andHelpman, Melitz and Yeaple (2004). Many phenomena observed in nature follow the discrete analogue of the Paretodistribution, namely Zipf’s Law. Firm size and efficiency, which are highly correlated, fall in this category and arewell approximated by the Pareto distribution over much of the relevant range (Luttmer, 2007; Eaton, Kortum andKramarz, 2011).


12 Bulletin

Zij =

(aij

aL

)r−1

, (21)

which gives equation (16) according to

Vij =max{

k[Zd

ij

(PiPj

)d(r−1)−1

], 0}, (22)

where d≡ k−r+1

r−1and where k≡ k

akH−ak

L

. Increases in trade costs will decrease Zij, but will

increase PiPj, mitigating the net effect on the latent variable Zij and hence on the extensive

margin Vij.

Taking logs of equation (12) yields the equation we work with for estimation and

comparative statics. Specifying trade costs (r−1) ln tij = cdij−k

2(Li +Lj), where dij is log

bilateral distance andLk is ameasure of country k’s export logistics (such that trade frictions

fall with better logistics) gives

mij =w+ yi + yj− cdij +k

2(Li +Lj)+xij + ln

(PiPj

)r−1, (23)

where w is a constant. xij is given by

xij≡ ln{

ed[zij + ln(PiPj)r−1]−1

},

which is the term containing the extensive margin. It contains zij = ln Zij, which since

Zij = Zij

(PiPj

)r−1, is

zij =χ+ yi + yj− cdij +k

2(Li +Lj)− ln fij +(r−1) ln

(PiPj

), (24)

in which χ is a constant. As in equation (23), better international logistics increase the

extensive margin of bilateral trade flows.

Accounting for MR requires a way of dealing with price index terms. We will use

an extension of the approach taken by Baier and Bergstrand (2009), which is a work-

horse in macroeconomics: Taylor’s method. In particular, Behar and Nelson (2012) show

that, when the extensive margin terms Vij entering the price indices are approximately

Vij≃ Zdij

(PiPj

)d(r−1), the MR term ln

(PiPj

)r−1is well approximated by

1

1+d×

−

World Trade Resistance︷︸︸︷∑

l

sl

∑

h

sh [(r−1) ln tlh−dzlh]

+∑

h

sh [(r−1) ln tih−dzih]

︸︷︷︸Importer’s MR

+∑

l

sl

[(r−1) ln tlj−dzlj

]

︸︷︷︸Exporter’s MR

. (25)

Equation (25) extends Baier and Bergstrand’s (2009) approximated MR term to the

case of firm heterogeneity. The expression includes not only an intensive margin effect

(r− 1) ln tlh in each component of the MR term, but also an extensive margin compo-



nent through dzlh. This approximated MR term shares with BB’s method the advantage of

yielding analytical tractability and a clear intuition for the comparative statics effects we

subsequently compute. It also preserves a role for potential asymmetries in trade costs.

Equation (25) shows that MR can be conveniently decomposed into three terms. The

first captures world trade resistance, which averages the importing MR of all importers

from j.When this world resistance term is higher, world trade in general is subject to higher

trade frictions, reducing bilateral trade all else being equal. The other two terms in equation

(25) are i’s importing MR and j’s exporting MR respectively. When either of these two

terms is high, trading with other countries in the world trade system is subject to high trade

costs, encouraging i and j to tradewith each other instead. This clearly captures the idea that

it is relative trade costs that matter in determining bilateral trade flows. For example, when∑h sh [(r−1) ln tih−dzih] is large, all exporters to i incur high trade costs in trading with i.

Country i therefore incurs relatively small trade costs in importing from j, raising exports

from j to i.

Substituting our expression for trade frictions into equation (25) allows us to write MR

terms for distance, logistics and fixed costs as

ln(PiPj

)r−1= cMRdist

ij −k

2MR

logisticsij +jMR

fij , (26)

where j=1/ (1+d), and where

MRdistij ≡−

∑

l

sl

∑

h

shdlh +∑

h

shdih +∑

l

sldlj, (27)

MRlogisticsij ≡

∑

l

sl

∑

h /= l

sh(Ll +Lh)−∑

h /= i

sh(Li +Lh)−∑

l /= j

sl(Ll +Lj), (28)

MRf

ij ≡−∑

l

sl

∑

h

sh ln flh +∑

h

sh ln fih +∑

l

sl ln flj. (29)

Then

mij =w+ yi + yj− c(dij−MRdist

ij

)+

k

2

(Li +Lj−MR

logisticsij

)+jMR

fij +xij, (30)

gives our heterogeneous firms gravity equation accounting for MR.

Special case: homogeneous firms

As heterogeneity disappears, this set-up reduces to a simple homogeneous firms model. To

see this, note that, as (i) all firms export and (ii) the support of the distribution of firm pro-

ductivities collapses, such that aij→aH and aH →aL, then, by algebraic manipulation and

L’Hopital’s rule, we have limaH→aLVij =max

{a1−r

L , 0}. In that case, the gravity equation is

simply

mij =w′+ yi + yj− c(dij−MRdist

ij

)+

k

2

(Li +Lj−MR

logisticsij

), (31)

where the constant w′ now contains an additional term reflecting the magnitude of aL.


14 Bulletin

Comparative statics

To emphasize the role of MR, we consider comparative statics in a homogeneous firms

model. Thereafter, we also allow for firm heterogeneity and hence firm entry.

Homogeneous firms model

When the exporter’s logistics quality improves, the partial equilibrium change is∂mij

∂Lj= k

2.13

Accounting for MR, the general equilibrium effect is

∂mij

∂Lj

∣∣∣∣homog.

=k

2

{1+2sj

(1− sj

)− sj−

(1− sj

)}(32a)

=ksj

(1− sj

)(32b)

≈ sjk. (32c)

The first term in the {·} brackets in equation (32a) gives the partial equilibrium ef-

fect in the absence of MR. The third term is the effect operating through the importer’s

MR, which falls by the exporter’s GDP share, dampening the partial equilibrium effect.

The fourth term is the exporter’s MR, which falls across all export destinations relative

to domestic trade. The proportion of j’s export demand this covers is 1− sj. The sec-

ond term is the effect operating through ‘world resistance’, which captures how costly

international trade is relative to domestic trade for all countries. sj

(1− sj

)enters twice

because, on the one hand, it makes j export more directly. On the other hand, it also

makes it import more, which through trade balance makes it export more.14 The net

general equilibrium effect after one allows for terms to cancel is ksj

(1− sj

). It illustrates

the diversion away from domestic trade and towards international trade as international

trade costs fall.

As a result, the net comparative static effect is not k

2, but something much smaller. The

comparative static effect (32c) is increasing in country size, which is consistent withAvW

(2003). As discussed in section II, the reason for this is that smaller countries consume a

smaller proportion of their produce domestically and export a larger proportion of their

products. More of their trade is international trade, so more is subject to international trade

costs, so MR has more of an effect.

It is informative to consider the special case when all countries are the same size so

that sk = 1

n. Then

MRlogistics

ij =−2

nL−

(Li +Lj

)(n−2

n

), (33)

13

Analogous to bilateral variables, the effect would be k if both the importer and exporter were to improve logistics.14

If we do not specify importer logistics as part of the bilateral trade cost function, then we end up with the samecomparative static effect, except we attribute it entirely to the first world resistance effect. It is impossible to identifyhow much of the effect is due to ‘exports’ and how much is due to ‘imports’, but the AvW (2003) cost symmetryassumption implies it is half each.



where L= 1

n

∑l Ll such that the first term on the right hand side is a constant. We can

rewrite equation (31) as

mij =w′+ yi + yj− c(dij−MR dist

ij

)+

k

n

(Li +Lj

), (34)

which allows one to interpret the coefficient on(Li +Lj

)as the effect of an improvement

in international logistics quality on trade for an average-size exporter with sj =1

n. With

reference to equation (32c), the comparative static effect for an average-size country is

∂mij

∂Lj

∣∣∣∣homog.

≈k

n≡ k. (35)

Heterogeneous firms model

In the presence of firmheterogeneity, the comparative statics of a change in logistics quality

must capture three effects:

(i) The effect at the intensive margin,−(r−1)∂ ln tij

∂Lj= k

2;

(ii) The MR effect occurring at the intensive margin,(r−1)∂ lnPiPj

∂Lj=− k

2(1−2sj +2s2j );

(iii) The effect at the extensive margin, which also has bilateral and multilateral com-

ponents, where∂zij

∂Lj=ksj

(1− sj

)and

∂wij

∂Lj=

(de

dxij

edxij−1

)ksj

(1− sj

).

Combining the effects yields

∂mij

∂Lj

∣∣∣∣hetero.

=k

2

[2sj

(1− sj

)] [1+

dedxij

edxij −1

]. (36)

The gravity parameter k

2is the effect at the intensive margin not accounting for MR. The

first square bracket is the adjustment for MR. The second square bracket is the amplifica-

tion brought about by allowing for the effect at the extensive margin. Note that ignoring

this term gives the intensive margin provided we have controlled for wij in estimation in

our heterogeneous firm model. For the case of symmetric countries,

mij =w+ yi + yj− c(dij−MR dist

ij

)+

k

n

(Li +Lj

)+jMR

fij +xij. (37)

In the heterogeneous model, k

nis the approximate intensive margin change for an aver-

age-size country, while k

n

dedxij

edxij−1

gives the approximate extensive margin change for an

average-size country. The overall bilateral country-level effect for an average-size country

is:

∂mij

∂Lj

∣∣∣∣hetero.

≈ k

[1+

dedxij

edxij −1

]. (38)

V. Estimation

To place our model in a stochastic framework, we allow for measurement error in the

reporting/recording of trade flows and unobserved trade costs. A necessary condition for


16 Bulletin

consistent estimates is that the error term is Independently and Identically Distributed

(IID). This section will discuss potential reasons why the IID assumption might not hold.

It also discusses the dropping of high-income exporters from our sample.

Endogeneity bias

The IID assumption rules out reverse causation, but higher trade volumes may stimulate

the construction of new infrastructure and the introduction ofmore efficient clearance tech-

nologies: the marginal value of investments in trade facilitating measures may be higher

if exports are high, while some aspects of the logistics technology are subject to scale

economies and thus only worthwhile at high volume. This could cause an upward bias in

the estimated coefficient. In contrast, high trade volumes may increase the strain on the

system, leading to queues at the border and longer customs processing times (Djankov

et al., 2010), and causing downward bias in the estimates. The instrumental variables (IV)

specifications we include provide a check for robustness. Drawing on the World Bank’s

Doing Business database, we use the sum of business start-up procedures in both exporter

and importer as one instrument and their product as another.We expect suchmeasures to be

correlated with logistics. At the same time, trade volumes should not influence the bureau-

cracy associated with filling out forms. While this should account for the main source of

endogeneity, we acknowledge the instrumentsmay be invalid if there are unobserved coun-

try characteristics that are common to both logistics and procedures. As a result, we use

start-up procedures observed in the importing country. Unlike the exporter’s procedures,

it is much harder to argue that the importer’s procedures should be directly related to the

exporter’s exports.

Omitted MR terms

While we have emphasized the importance of MR for comparative statics, it can have

an effect on estimation. If we have the bilateral trade cost variables but omit their MR

analogues, these terms would be in the error term. By construction, this would make the

error term correlated with the regressors, would invalidate IID and lead to biased estimates

of the coefficients. We construct MR terms for all bilateral variables and include importer-

equivalents for all country-specific variables. We follow Baier and Bergstrand (2010) and

Melitz (2008)15 and construct the MR terms by taking simple averages. We also follow

Baier and Bergstrand (2009, 2010) and perform estimation with the equality restrictions

implied by equation (31) imposed. For example, we include the sum of exporter and

importer logistics as a single variable and include dij −MR distij as a single variable. As

suggested by equation (34) or equation (37), this means we can interpret the logistics

regression coefficient as the relevant trade effect for a country of average size.

15

We thank an anonymous referee for the reference. Melitz’s derivation of MR terms in the country-specific caseis similar to ours (cf Melitz’s equation 7 and footnote 10) but not exactly the same. This is in part due to differencesin treatment of the country-specific variable when i = j.



Firm heterogeneity

The heterogeneous firms model also indicates potential violation of the IID assumption.

Leaving out the control for the proportion of firms exporting would lead to omitted vari-

ables bias. Furthermore, the model suggests how country-selection into trade is a function

of the variables of interest and a potential source of sample selection bias. To address these

issues and implement the heterogeneous firms model, a two-step procedure is needed.

In the first stage, we estimate a probit model for the probability that country j exports to

i, denoted qij. Letting Tij be unity when exports from j to i are observed and zero otherwise,

we write

qij =Pr (Tij =1|observables, unobservables). (39)

Predicted values of qij are used to generate zij =U−1(qij) (cf. equation (24)), where U(.) is

the standard cumulative normal distribution function. To control for the country

selection effect, we predict the Inverse Mills Ratio (IMR) gij=

/ij

U(zij), where /(.) is the

standard normal density function. Drawing on HMR, zij and gij can be used to account for

firm selection as follows: define the propensity to export xij≡ zij + gij. This is an estimate

of the latent variable zij as a function of both observable and (an estimate of) unobservable

trade frictions. As above, attaching a Pareto distribution to firm productivities allows us

to map xij to a consistent estimate of the number of firms profitable enough to export.

We include this xij = ln(edxij −1), together with gij, in the second stage of our regression,

modifying equation (30) to yield

mij =w+by

(yi + yj

)− c

(dij−MR dist

ij

)+

k

n

(Li +Lj

)+jMR

fij + ln

(edxij −1

)+bggij. (40)

We estimate this second stage using nonlinear least squares and use bootstrapped standard

errors to allow for the fact that we have generated regressors in the second stage.16 For

reliable identification, we need to have one variable in the probit equation that is not in the

second stage. Our theoretical framework suggests variables that affect a firm’s fixed costs

(fij) of exporting but not its variable costs. For this purpose, we use a dummy for whether

two countries were once the same country (Samecountry). This is a good measure of fixed

costs because countries that were formerly the same should already have established net-

works, for example trade contacts or even members of the same family, which reduces

fixed costs of seeking new markets for products. Conditioning on standard variables like

distance and common borders, there is limited scope for the dummy to affect variable costs.

Our first-stage probit only differentiates between zeros and ones and ignores themissing

observations. Baranga (2009) suggests this may induce further selection bias. Addressing

these concerns, we also ran specifications in which we estimated a preliminary probit

for missing variables. Including an analogous IMR in the ‘first-stage’ probit for positive

trade yielded a coefficient of virtually zero with a P-value of close to unity. This suggests

ignoring missing values is not an issue for our data.

16

We perform 50 replications. Furthermore, we reject draws that spawn theoretically illegitimate values for xij

and gij.The first stage and homogeneous goods models use standard errors clustered by country-pair.


18 Bulletin

VI. Results

Here, we discuss the estimation results. In a homogeneous firm setting, we start with

benchmark specifications. We discuss the impact of MR on the estimates and then we

assess potential endogeneity. Thereafter, in a heterogeneous-firm setting, we show the

results of our two step procedure. The results in this section estimate trade flows using

developing countries, but we calculate MR terms by summing over all 116 countries and

perform comparative statics accordingly.17 Appendix B explores estimation based on all

countries. As discussed there, we were not satisfied with the extent to which an interaction

between logistics and a developing country dummy captured differences by income group.

More importantly, our results were not as reliable or robust when estimated on the full

sample.

Homogeneous firms

In Table 2, column 1 presents a standard augmented gravity model and the signs of the

coefficients are as one would expect. In particular, the international logistics coefficient

(on Li +Lj) is significant with a value of 0.606. To get a sense of magnitude, we calcu-

late the implied effect on exports of a one-standard deviation rise in the ILI. This modest

0.4 unit rise would for example put Rwanda’s logistics on a par with nearby Tanzania’s,

make Bulgaria’s like Romania’s or place Brazil just above Argentina. Such a rise would

raise exports e0.6*0.4−1=27%. Alternatively, we can use the coefficients on logistics and

distance to calculate this is equivalent to a reduction in distance of k

c*0.4=17%. In column

2, we add a full set of controls for MR for each bilateral variable – dij is replaced with(dij−MR dist

ij

)for example–and the logistics coefficient is 0.597. While the influence of

MR on estimation is small, we will see later that its comparative static impact will be very

important.

Column 3 presents the second stage of a 2SLS estimation, where we have the sum of

procedures in both exporter and importer as one instrument and their product as another.

Both variables have explanatory power in the first-stage and the overidentification test

is insignificant. Thus, conditional on one of these instruments being valid, we can legiti-

mately exclude both. This condition cannot be tested, but our regression of the residuals

on the exogenous variables yielded individually and jointly insignificant terms. The sec-

ond stage generates a coefficient of 0.697 on logistics, which is statistically insignificantly

higher than the column 2 coefficient. Column 4 presents the results where the first stage

has the number of procedures needed to start a business in the importer instrumenting

for logistics (Li +Lj). In the first stage, the instrument was significant and there is only

17

For comparative statics, a prior version of this paper multiplied by the 88 developing countries. We are gratefulto an anonymous referee for pointing out it is more appropriate to proxy for the whole world with the full sample ofcountries in our dataset.AvW (2003) have a two-country model as well as a multi-country model. In the former, theyuse only US and Canada information. In the latter, they estimate bilateral trade flows using only Canadian Provincesand US states but include data from other countries in their system of price equations. Our approach here is analogousto their multi-country model.



TABLE 2

Gravity regressions for homogenous firm model

1 2 3 4

Dependent variable: bilateral exports OLS OLS MR IV 1 IV 2

GDP 0.987*** 0.922*** 0.904*** 0.951***

[0.0161] [0.0157] [0.0295] [0.0274]

Logistics (ILI) 0.606*** 0.597*** 0.697*** 0.437**

[0.0566] [0.0556] [0.151] [0.138]

Distance −1.437*** −1.467*** −1.474*** −1.456***

[0.0437] [0.0496] [0.0501] [0.0500]

Border 1.491*** 1.071*** 1.074*** 1.066***

[0.166] [0.176] [0.176] [0.176]

Colony 0.653*** 0.501*** 0.508*** 0.491***

[0.116] [0.133] [0.134] [0.134]

Language 0.651*** 0.646*** 0.645*** 0.648***

[0.0769] [0.0956] [0.0955] [0.0956]

Samecountry 0.403* 0.398 0.387 0.416*

[0.227] [0.243] [0.243] [0.244]

Religion 0.122 0.826*** 0.829*** 0.820***

[0.107] [0.143] [0.143] [0.142]

Landlocked −0.752*** −0.648*** −0.656*** −0.635***

[0.0651] [0.0653] [0.0659] [0.0657]

Island −0.00127 −0.410*** −0.421*** −0.393***

[0.0702] [0.0694] [0.0697] [0.0700]

Constant −24.11*** −33.18*** −32.89*** −33.66***

[0.695] [0.602] [0.724] [0.704]

N 6939 6939 6939 6939

Instrument relevance† 0.134*** 0.171***

Instrument overidentification‡ 1.497 —

Endogeneity test§ 0.518 1.687

adjusted R2 0.593 0.589 0.589 0.588

Notes: Logistics is significant in all columns and robust to the inclusion of controls or the useof instrumental variables. Column 1 will serve as the benchmark. Column 2 includes MR terms.Columns 3 (1 instrument) and 4 (2 instruments) give insignificantly different coefficients eitherside of the OLS coefficient.Standard errors in brackets.Significance levels: *P <0.1, **P <0.05, ***P <0.01.†Shea (1997) adjusted partial R2; ‡Sargan (1958) Chi-square; §Wooldridge (1995) F-statistic.

one instrument so no overidentifying restriction to test.18 The coefficient on logistics is

0.437, which is insignificantly lower than the OLS coefficient.19 Given the insignificant

differences, we will use the column 2 estimate, which lies in between the 2SLS estimates,

18

We ran an alternative specification in which we include both importer and exporter procedures as separate instru-ments. This produces a coefficient of 0.49. Informatively, an auxilliary regression of the residuals found that theexporter logistics variable is significant while the importer’s was not.

19

We experimented with alternative variables and functional forms, including the importer’s documentation forimports, the importer’s documentation for exports, the importer’s costs of starting a business, the importer’s durationin days of starting a business and combinations thereof. These tended to produce significant coefficients in residualsregressions and/or significant overidentification tests as well as instances of insufficient explanatory power in thefirst stage.


20 Bulletin

for the homogeneous firm comparative static exercises. The results suggest endogeneity

may not be a concern, but we also need to consider potential bias due to firm heterogeneity.

Heterogeneous firms

Table 3 presents the results that account for firm- and/or country-selection. In column 1,

the probit model yields all the expected signs. Logistics affects the probability that a

country exports, which in our framework means logistics affect fixed costs. The Same-

country variable is also significant.

In column 2, we present the second stage of the Heckman two-step procedure, where

we have excluded the Samecountry variable.20 The logistics coefficient is a little higher

than in Table 2.Analogous to HMR, this is because failure to account for country selection

induces a negative correlation between logistics quality and the error term, because a low

logistics qualitymeans that unobserved trade frictionsmust be low on average for countries

to be observed as trading pairs. This in turn induces a downward bias in the homogeneous

logistics coefficient. However, the insignificant IMR and small difference in coefficients

suggest that country-selection is not an important issue in our application.

In column 3, we implement the full HMR procedure to account for both country- and

firm-selection. We omit the bilateral Samecountry variable but preserve its MR compo-

nent to be consistent with the theoretical specification (equation (40)).21 This specification

assigns low coefficients to logistics and distance but a high estimated d. Thiswould indicate

that the influence of logistics is only through the extensive margin. However, as in column

2, the IMR is still insignificant, so the next two specifications control for firm-selection

but not country-selection.

In column 4, we followManova (2008) by excluding no explanatory variables from the

second stage but dropping the IMR. This produces a logistics coefficient of 0.489 and an

estimate for d of 0.307. Excluding the Samecountry variable from column 5, the logistics

coefficient is 0.445 and d is 0.416.22 Column 5 is our preferred specification but, provided

we exclude the IMR, our results are not materially affected by the choice of exclusion

variable. Column 5 implies a one standard deviation improvement in the ILI is equivalent

to a 14% reduction in distance.23

Consistent with HMR, allowing for firm heterogeneity produces lower coefficients.

Comparing our two benchmarks for example, the homogeneous coefficient (column 2 of

20

We also ran a just-identified model with all first-stage variables included in the second stage. The results werevery similar. We also attained similar results using the number of regulations required to start a business in the firststage probit and excluding those from the second stage.

21

Excluding the MR component made no difference. Furthermore, as done by HMR, we attempted specificationswhich exclude religious similarity or the procedures or days needed to start a business. The specifications yieldeda large significantly negative Inverse Mills Ratio, a GDP coefficient well below unity and/or a positive distancecoefficient. These all suggest excluding these variables from the second stage is not appropriate for our dataset.We also attempted export documentation and/or import documentation in the importer and/or exporter, or combina-tions theoreof. While these are a priori highly plausible candidates for fixed but not variable trade costs, the resultssuggested they are not reliable identifying variables.

22

We do not discuss the significance of this variable because it must be non-zero for a well-defined gravity model,so it cannot be zero under the null hypothesis. Furthermore, the term ln(edxij −1) is only the ij-specific componentof x. The rest of it is subsumed in the constant.

23

Because all trade costs are equally affected by firm heterogeneity and multilateral resistance, we still calculatethis by taking the quotient of the logistics and distance coefficients.



TABLE 3

Gravity regressions for heterogeneous firm model

1 2 3 4 5

Dependent variable: bilateral exports Probit Country Country, firm Firm 1 Firm 2

GDP 0.244*** 0.941*** 0.679*** 0.868*** 0.840***

[0.00937] [0.0183] [0.135] [0.0477] [0.0462]

Logistics (ILI) 0.483*** 0.636*** 0.123 0.489*** 0.445***

[0.0342] [0.0704] [0.273] [0.105] [0.0948]

Distance −0.635*** −1.520*** −0.842** −1.314*** −1.270***

[0.0374] [0.0463] [0.365] [0.141] [0.128]

Border 0.175 1.162*** 0.881*** 1.020*** 1.059***

[0.175] [0.163] [0.215] [0.172] [0.184]

Colony 0.152** 0.563*** 0.365** 0.478*** 0.499***

[0.0689] [0.146] [0.160] [0.121] [0.143]

Language 0.418*** 0.680*** 0.247 0.550*** 0.527***

[0.0598] [0.0944] [0.254] [0.112] [0.115]

Samecountry 0.489** −4.123*** 0.281 −3.961**

[0.196] [1.367] [0.258] [1.566]

Religion 0.316*** 0.853*** 0.541*** 0.759*** 0.751***

[0.0865] [0.141] [0.204] [0.167] [0.146]

Landlocked −0.121*** −0.657*** −0.523*** −0.622*** −0.603***

[0.0296] [0.0582] [0.105] [0.0796] [0.0791]

Island 0.0685* −0.408*** −0.522*** −0.427*** −0.477***

[0.0404] [0.0667] [0.0688] [0.0651] [0.0737]

Constant −13.32*** −34.38*** −19.05*** −29.45*** −27.94***

[0.385] [0.982] [7.344] [2.425] [2.303]

Inverse Mills Ratio 0.273 −0.725

[0.229] [0.585]

Delta 1.053* 0.307 0.416*

[0.558] [0.241] [0.225]

N 9350 6939 6939 6939 6939

adjusted R2 0.589 0.589 0.589 0.589

Notes: Column 1 is the probit estimate. Column 2 controls for country selection (only) using the Heckman 2-stepmodel. Column 3 follows the HMR procedure with the exclusion of the bilateral component of the same countryvariable (theMR component is included). Column 4 followsManova (2008) by not excluding any trade cost variablesbut excludes the Inverse Mills Ratio. Column 5 excludes the bilateral component of the same country variable andthe Inverse Mills Ratio and is the basis for the country-level simulations.Standard errors in brackets (bootstrapped in columns 2–5).Significance levels: *P <0.1, **P <0.05, ***P <0.01.

Table 2) is 0.597 while the firm-level coefficient (column 5 of Table 3) is 0.445. However,

this does not mean that the country-level effect is smaller.

Table 4 in fact demonstrates much higher bilateral country-level effects, which are

calculated for each country-pair by allowing for the extensive margin (cf. equation (38)).

For reference, the first column presents values from the homogeneous firms model. This

is followed by the firm-level coefficient from our preferred heterogeneous firms model.

The subsequent values reveal substantial variation in the extensive margin and hence

the country-level effect. Moreover, the minimum value of 0.648 is greater than that im-

plied by the homogeneous model, so the homogeneous firm model has in this applica-

tion underestimated the country-level effect of logistics for all countries. The source of


22 Bulletin

TABLE 4

Summary statistics for bilateral export response to improved logistics

Country level

Homogeneous Firm-level Mean p25 p50 p75 SD min max

0.597 0.445 0.882 0.770 0.855 0.971 0.143 0.648 1.522

Notes: The homogeneous statistic refers to the ILI coefficient in column 2 of Table 2. Otherstatistics are derived from column 5 of Table 3.

variation across country-pairs and hence across income-groups is driven by variations in the

extensive margin. Differentiation of equation (38) with respect to xij would show that the

extensive margin effect is higher for countries with a lower value of xij, which implies

the extensive margin effect is higher for those who tend to have a lower proportion of

firms exporting. This in turn implies, for example, that more distant countries with lower

logistics quality would have a bigger increase in bilateral exports, ceteris paribus.

The average country-level effect of 0.882 on bilateral trade is almost 50% higher than

that implied by the homogeneous firms model. It also suggests that the country-level effect

is on average half due to the intensive margin and half due to the extensive margin. In

other words, approximately half the country-level effect is due to new firms entering the

export market.24

VII. Simulations: total country-level exports

To understand the effects of logistics on a country’s total exports in the heterogeneous

goods model, we must aggregate over the bilateral elasticities calculated in Table 4. It is

also about time we recognized the importance of MR more explicitly by factoring in the

actual size of the country. The distribution of world GDP is highly skewed. China accounts

for about 5% of world output. This is about the same as the next three biggest developing

countries combined (Brazil, India andMexico). This 10% share exceeds that of the other 84

developing countries in our sample! Mean GDP is about seven times as big as the median

and, in our full sample, the world mean for non-African countries is almost 40 times as

big as that for African countries.

Because of MR, this extreme skewness means the effect on a particular country or

subset of countries can be very different to the average. We do not allow for country-level

entry in the simulations because HMR attribute very little of the rise in international trade

to the formation of new bilateral relationships and the IMR was insignificant.

Homogeneous firms

To calculate the effect for a particular country j, equations (32) and (34) imply that we

multiply the logistics coefficient by n to get k and then multiply that by sj

(1− sj

). Using

the estimate k

n=0.597 and n=116 gives k≈69. Table 5 has the (semi-)elasticities for all

24

See Eaton, et al. (2007); Manova and Zhang (2009); and Bernard, et al. (2009) for empirical analyses of marginsfor particular countries using microeconomic data.



TABLE5

Co

un

try-

leve

lel

ast

icit

ies

aft

era

cco

un

tin

gfo

rM

ult

ila

tera

lR

esis

tan

ce

Exp

ort

erS

ha

reH

om

og

eneo

us

Het

ero

gen

eou

sE

xpo

rter

Sh

are

Ho

mo

gen

eou

sH

eter

og

eneo

us

China

5.34%

3.50

3.93

Jordan

0.03%

0.02

0.03

Brazil

1.90%

1.29

1.53

Cote

d’Ivoire

0.03%

0.02

0.03

India

1.82%

1.24

1.45

Ethiopia

0.03%

0.020

0.026

Mex

ico

1.81%

1.23

1.35

Bolivia

0.03%

0.019

0.023

Russia

0.99%

0.68

0.81

Ugan

da

0.02%

0.015

0.021

Argen

tina

0.89%

0.61

0.72

Jamaica

0.02%

0.017

0.021

Turkey

0.70%

0.48

0.56

Parag

uay

0.02%

0.017

0.021

Indonesia

0.59%

0.41

0.47

Ghan

a0.02%

0.013

0.017

Poland

0.56%

0.39

0.44

Honduras

0.02%

0.014

0.016

South

Africa

0.45%

0.31

0.37

Sen

egal

0.02%

0.011

0.015

Thailand

0.45%

0.31

0.35

Nep

al0.02%

0.012

0.015

Iran

0.38%

0.26

0.31

Mozambique

0.02%

0.011

0.015

Egypt

0.35%

0.24

0.29

Mau

ritius

0.02%

0.011

0.014

Ven

ezuela

0.37%

0.26

0.29

Gab

on

0.02%

0.011

0.014

Malay

sia

0.32%

0.22

0.25

Zim

bab

we

0.02%

0.011

0.014

Colombia

0.28%

0.19

0.22

Cam

bodia

0.01%

0.010

0.012

Pak

istan

0.26%

0.18

0.22

Alban

ia0.01%

0.009

0.011

Chile

0.27%

0.18

0.22

Mad

agascar

0.01%

0.009

0.011

Philippines

0.26%

0.18

0.21

Zam

bia

0.01%

0.008

0.011

Algeria

0.19%

0.13

0.16

Nicarag

ua

0.01%

0.009

0.011

Peru

0.18%

0.13

0.15

Guinea

0.01%

0.007

0.010

Ban

gladesh

0.17%

0.12

0.14

Burkina

0.01%

0.007

0.010

Nigeria

0.16%

0.11

0.13

Mali

0.01%

0.006

0.010

Hungary

0.16%

0.11

0.13

Haiti

0.01%

0.007

0.009

Roman

ia0.14%

0.10

0.11

Pap

uaNew

Guinea

0.01%

0.007

0.009

Vietnam

0.13%

0.09

0.10

Ben

in0.01%

0.005

0.008

Morocco

0.12%

0.08

0.10

Chad

0.01%

0.005

0.007

Tunisia

0.07%

0.05

0.06

Rwan

da

0.01%

0.005

0.007

Syrian

0.06%

0.04

0.05

Niger

0.01%

0.004

0.006

Uruguay

0.06%

0.04

0.05

Lao

0.01%

0.005

0.006

(co

nti

nu

edo

verl

eaf)


24 Bulletin

TABLE5

con

tin

ued

Exp

ort

erS

ha

reH

om

og

eneo

us

Het

ero

gen

eou

sE

xpo

rter

Sh

are

Ho

mo

gen

eou

sH

eter

og

eneo

us

Dominican

0.06%

0.04

0.05

Malaw

i0.01%

0.004

0.006

Leb

anon

0.06%

0.04

0.05

Togo

0.00%

0.003

0.004

Guatem

ala

0.06%

0.04

0.05

Mau

ritania

0.00%

0.003

0.004

SriLan

ka

0.06%

0.04

0.05

Sierra

0.00%

0.002

0.004

Ecu

ador

0.06%

0.04

0.05

Mongolia

0.00%

0.002

0.003

CostaRica

0.05%

0.04

0.04

Burundi

0.00%

0.002

0.002

Sudan

0.05%

0.03

0.04

Guyan

a0.00%

0.001

0.002

Bulgaria

0.05%

0.03

0.04

Djibouti

0.00%

0.001

0.002

Ken

ya

0.04%

0.03

0.04

Gam

bia

0.00%

0.001

0.002

Angola

0.04%

0.03

0.04

Liberia

0.00%

0.001

0.001

Tan

zania

0.04%

0.02

0.03

Solomon

0.00%

0.001

0.001

Pan

ama

0.04%

0.03

0.03

Guinea-B

issau

0.00%

0.000

0.001

ElSalvad

or

0.04%

0.03

0.03

Comoros

0.00%

0.000

0.001

Yem

en0.04%

0.02

0.03

Mea

n0.24%

0.16

0.19

Cam

eroon

0.03%

0.02

0.03

Med

ian

0.03%

0.02

0.03

Note

s:Elasticitiesforea

chco

untryassu

methat

only

that

countryim

proves

itsinternational

logistics

qualitybyoneunitwhile

othershold

logistics

constan

t.



88 developing countries. The trade impact clearly depends onwhether you are the Comores

or China.We average over all exporters to get themean elasticity of 0.16. This value, which

we call the ‘typical’ elasticity, is about a quarter of that for a country of average-size 1

n.

Argentina’s share of GDP is only slightly higher than the average (1/116) and its compar-

ative static effect is accordingly only slightly higher than that given by the homogeneous

coefficient. China’s elasticity is more than 20 times bigger than the typical elasticity. Sim-

ilarly, Brazil, India and Mexico all have elasticities more than double those implied if we

ignore MR. On the other hand, 82 out of 88 countries are below average, so ignoring MR

typically overestimates the elasticity and sometimes does so by a large amount. While this

calls for estimation of elasticities at a country level, we are not necessarily saying that k

is inappropriate: by implicitly giving a greater weighting to bigger countries, it may be a

good summary measure.

Heterogeneous firms

For each exporter, the elasticity varies by importer when firm heterogeneity is included.

Therefore, for each exporter, we sum the bilateral response over all its importers, weighting

by the level of exports:

∂mj

∂Lj

=1∑

l

Mlj

∑

l

[Mlj

k

n

(1+

dedxlj

edxlj −1

)]. (41)

The average-size total country response is 0.766, which is higher than the homogeneous

firms benchmark that did not account for MR in estimation (0.606) and implies a one

standard deviation improvement in the ILI would raise exports by 36%. To incorporate

MR fully in comparative statics, we follow analogous procedures to before and compute

nj = sj

(1− sj

) ∂mj

∂Lj

.

nj is a key object of interest and forms the basis for our main substantive result. We expect

bigger exporters to have smaller elasticities in general because they have a lower exten-

sive margin effect, but this effect is empirically dominated by the MR effect. Overall, the

biggest countries have the biggest elasticities, which would be true for all country-specific

determinants of trade costs, not just logistics.

Alongside the homogeneous model elasticities, Table 5 presents the heterogeneous

model elasticity for each country with respect to an improvement in its own logistics. The

average value of nj across 88 exporters is 0.18, which implies a one standard deviation rise

in a country’s ILI would raise exports by 7.67%. Recall that such an improvement would

place Brazil on a par with Argentina next door. Brazil’s size means its calculated response

of 84% is more than triple the benchmark of 27%.More dramatically, Rwanda’s small size

means its trade response of 0.028% would be barely one per cent of the benchmark.

By way of summary, we make two final comparisons using estimates we have already

presented. Our final estimate of the response to a one standard deviation improvement in

logistics quality for a country of average size is 36%. This is somewhat higher than the

27% response implied by our benchmark model in column 1 of Table 2. However, this

masks the huge variation in response by country and, due to the skewed distribution of


26 Bulletin

country size, is not typical. Averaging over each country-level response, our measure of

the typical response is only 7.67%, which is barely one quarter of the effect implied by the

benchmark. This difference is an order of magnitude greater than typical estimation bias.

VIII. Concluding discussion

We have seen consistent evidence that an exporter’s logistics increase exports. Our pre-

ferred heterogeneous specification indicates that the elasticity of total exports with respect

to a change in logistics for a country of average size is 0.766. However, most countries

are much smaller than the average, so most countries have much smaller effects. After

calculating elasticities for all countries, the typical (mean) elasticity is only 0.18, which

implies a one standard deviation improvement in logistics would raise exports by 7.67%.

By contrast, the benchmark linearmodel would have produced estimates of 27%,which

is a large exaggeration for the typical country and a 100-fold exaggeration for Rwanda. The

chief reason for these exaggerations is that standardmethods ignore the general equilibrium

effects operating through MR, which are stronger for smaller countries and hence dampen

their trade responses by more. In contrast, a handful of large countries have elasticities that

are much higher than that given by the benchmark.

One of our key results is that economically small countries have small export elas-

ticities. When evaluating the case for logistics upgrades, this result should be considered

alongside the costs of improving logistics, which are also likely to increase with country

size. Upgrading international logistics for the whole of Brazil requires far more resources

than upgrading in Rwanda.25 This is an important factor in the context of broader evalua-

tions of net welfare gains and evaluations of aid-for-trade on a country-by-country basis.

While we have emphasized the importance of MR for individual country responses,

a summary view is arguably better captured by the elasticity for the average-size coun-

try, where bigger countries are implicitly given a bigger weighting. Allowing for firm

heterogeneity, a one standard deviation improvement gives a trade response of 36% for

an average-size country, which is higher than given by the benchmark. Nonetheless, the

cross-country variation in trade responses is overwhelmingly driven by differences in MR.

Our study has been of a unilateral improvement by a country, but a global analysis

requires an investigation of multilateral improvements. Wilson et al. (2005) simulate the

effect of bringing all countries with below-average measures of trade facilitation half way

up to the global average. Such a simulation is beyond the scope of this paper, but one that

takes proper account of firm heterogeneity and especially MR would be an informative

enterprise. Finally, we noted that statements about one transport cost expressed in other

trade cost equivalents are immune to the issues we have discussed. For any country, our

results indicate that logistics improvement is equivalent to a 14% reduction in distance.

Final Manuscript Received: May 2012

25

Furthermore, the functional form we have chosen does not readily allow for decreasing returns, nor does it allow

for variations in the gravity parameter k. This is standard in the literature. However, this parameter may vary acrosscountries because the products they export may differ (Djankov et al., 2010) or because the elasticity of substitution

is not constant. Even so, it would take very large variations in k to extinguish the relationship between size and thegeneral equilibrium response. More generally, a proper evaluation of welfare gains and losses would be needed.



References

Anderson, J. (1979). ‘A theoretical foundation for the gravity equation’, American Economic Review, Vol. 69,

pp. 106–116.

Anderson, J. and vanWincoop, E. (2003). ‘Gravity with gravitas: a solution to the border puzzle’, American

Economic Review, Vol. 93, pp. 170–192.

Anderson, J. and vanWincoop, E. (2004). ‘Trade costs’, Journal of Economic Literature, Vol. 42, pp. 691–751.

Arvis, J., Mustra, M., Panzer, J., Ojala, L. and Naula, T. (2007). ‘Connecting to Compete: Trade logistics in

the global economy’, The World Bank.

Baier, S. and Bergstrand, J. (2009). ‘Bonus Vetus OLS: a simple method for approximating international

trade-cost effects using the gravity equation’, Journal of International Economics, Vol. 77, pp. 77–85.

Baier, S. and Bergstrand, J. (2010). ‘Approximating general equilibrium impacts of trade liberalizations using

the gravity equation:Applications to nafta and the European economic area’, in Brakman S. and van Bergeijk

P. (eds), The Gravity Model in International Trade: Advances and Applications, Cambridge: Cambridge

University Press.

Baranga, T. (2009). ‘Unreported Trade Flows and Gravity Equation Estimation’, mimeo.

Behar, A. and Nelson, B. D. (2012). ‘Trade Flows, Multilateral Resistance and Firm Heterogeneity’, IMF

Working Paper, forthcoming.

Behar, A. and Venables, A. (2011). ‘Transport costs and international trade’, in de PalmaA. Lindsey R. Quinet

E. and Vickerman R. (eds), Handbook of Transport Economics, Cheltenham: Edward Elgar.

Bergstrand, J. (1985). ‘The gravity equation in international trade: some microeconomic foundations and

empirical evidence’, The Review of Economics and Statistics, Vol. 67, pp. 474–481.

Bergstrand, J., Egger, P. and Larch, M. (2007). ‘Gravity Redux: Structural Estimation of Gravity Equations

with Asymmetric Bilateral Trade Costs’, mimeo.

Bernard, A., Jensen, J., Redding, S. and Schott, P. (2007). ‘Firms in international trade’, The Journal of

Economic Perspectives, Vol. 21, 105–130.

Chaney, T. (2008). ‘Distorted gravity: the intensive and extensive margins of international trade’, American


Clark, X., Dollar, D. and Micco, A. (2004). ‘Port efficiency, maritime transport costs, and bilateral trade’,

Journal of Development Economics, Vol. 75, pp. 417–450.

Djankov, S., Freund, C. and Pham, C. (2010). ‘Trading on time’, The Review of Economics and Statistics,

Vol. 92, pp. 166–173.

Eaton, J., Eslava, M., Kugler, M. and Tybout, J. (1997). ‘Export dynamics in Colombia: firm-level evidence’,

NBER Working Paper, 13531.

Eaton, J., Kortum, S., and Kramarz, F. (2011). ‘The anatomy of international trade: evidence from French

firms’, Econometrica, Vol. 79, pp. 1453–1498.

Feenstra, R. (2004). Advanced International Trade: Theory and Evidence, Woodstock: Princeton University

Press.

GrowthCommission. (2008).The Growth Report: Strategies for Sustained Growth and Inclusive Development,

Commission on Growth and Development Final Report, Washington, DC.

Helpman, E., Melitz, M. and Rubinstein, Y. (2008). ‘Estimating trade flows: trading partners and trading

volumes’, Quarterly Journal of Economics, Vol. 123, pp. 441–487.

Helpman, E., Melitz, M. and Yeaple, S. (2004). ‘Export versus FDI with heterogeneous firms’, American


Huchet-Bourdon, M., Lipchitz, A. and Rousson, A. (2009). ‘Aid for trade in developing countries: complex

linkages for real effectivenes’, African Development Review., Vol. 21, pp. 243–290.

Hummels, D. (2001). Time as a Trade Barrier, GTAPWorking Papers.

Limao, N. and Venables,A. (2001). ‘Infrastructure, geographical disadvantage, transport costs, and trade’, The

World Bank Economic Review, Vol. 15, pp. 451–479.

Luttmer, E. (2007). ‘Selection, frowth, and the size distribution of firms’, Quarterly Journal of Economics,

Vol. 122, pp. 1103–1114.

Manova, K. (2008). ‘Credit Constraints, Heterogeneous Firms, and International Trade’, NBER Working Paper

14531.


28 Bulletin

McCallum, J. (1995). ‘National borders matter: Canada-US regional trade patterns’, American Economic

Review, Vol. 85, pp. 615–623.

Melitz, J. (2008). ‘Language and foreign trade’, European Economic Review, Vol. 52, pp. 667–699.

Melitz, M. (2003). ‘The impact of trade on intra-industry reallocations and aggregate industry productivity’,

Econometrica, Vol. 71, pp. 1695–1725.

Nordas, H. and Piermartini, R. (2004). Infrastructure and Trade, World Trade Organisation, Staff Working

Paper ERSD-2004-04.

Portugal-Perez, A. and Wilson, J. S. (2010). Export Performance and Trade Facilitation Reform : Hard and

Soft Infrastructure, Policy Research Working Paper Series, The World Bank .

Sargan, J. (1958b). ‘The estimation of economic relationships using instrumental variables’, Econometrica,

Vol. 26, pp. 393–415.

Shea, J. (1997). ‘Instrument relevance in multivariate linear models: a simple measure’, The Review of Eco-

nomics and Statistics, Vol. 79, pp. 348–352.

Tinbergen, J. (1962). Shaping the World Economy: Suggestions for an International Economic Policy, The

Twentieth Century Fund, New York.

Wilson, J., Mann, C. and Otsuki, T. (2005). ‘Assessing the benefits of trade facilitation: a global perspective’,

The World Economy, Vol. 28, pp. 841–871.

Wooldridge, J. (1995). ‘Score diagnostics for linear models estimated byTwo Stage Least Squares’, inMaddala

G., Phillips P. and Srinivasan T. (eds.), Advances in Econometrics and Quantitative Economics, Oxford:

Blackwell.

Appendix A

TABLEA.1

Logistics Performance Index (LPI) and International Logistics Index (ILI) by country

Country ILI LPI Country ILI LPI Country ILI LPI

Rwanda 1.90 1.77 Dominican Rep. 2.52 2.38 Argentina 3.10 2.98

Djibouti 1.99 1.94 Gambia 2.52 2.52 India 3.14 3.07

Sierra 2.04 1.95 Malawi 2.52 2.42 Saudi Arabia 3.14 3.02

Gabon 2.10 2.10 Egypt 2.52 2.37 Indonesia 3.17 3.01

Albania 2.11 2.08 Morocco 2.53 2.38 Poland 3.19 3.04

Chad 2.11 1.98 Iran 2.55 2.51 Turkey 3.26 3.15

Niger 2.14 1.97 P. New Guinea 2.55 2.38 Hungary 3.26 3.15

Solomon Islands 2.16 2.08 Ethiopia 2.56 2.33 Czech Rep. 3.28 3.13

Guyana 2.18 2.05 Angola 2.59 2.48 Israel 3.34 3.21

Algeria 2.20 2.06 Bangladesh 2.60 2.47 Chile 3.40 3.25

Tanzania 2.20 2.08 Uganda 2.61 2.49 China 3.42 3.32

Mauritius 2.25 2.13 Cambodia 2.61 2.50 Thailand 3.43 3.31

Syrian 2.26 2.09 Uruguay 2.62 2.51 Greece 3.52 3.36

Burundi 2.26 2.29 Benin 2.63 2.45 Portugal 3.57 3.38

Mongolia 2.26 2.08 Honduras 2.63 2.50 South Korea 3.61 3.52

Togo 2.28 2.25 Colombia 2.63 2.50 Spain 3.61 3.52

Nepal 2.29 2.14 Paraguay 2.64 2.57 Malaysia 3.63 3.48

Ghana 2.31 2.16 Costa Rica 2.68 2.55 South Africa 3.66 3.53

Nicaragua 2.31 2.21 Guatemala 2.70 2.53 Italy 3.68 3.58

Haiti 2.31 2.21 Ecuador 2.71 2.60 France 3.85 3.76

Lao PDR 2.36 2.25 Cameroon 2.72 2.49 New Zealand 3.87 3.75

(continued overleaf )



TABLEA.1

continued

Country ILI LPI Country ILI LPI Country ILI LPI

Burkina Faso 2.38 2.24 Venezuela 2.72 2.62 United States 3.89 3.84

Mozambique 2.38 2.29 Pakistan 2.73 2.62 Finland 3.91 3.82

Madagascar 2.39 2.24 Kenya 2.74 2.52 Norway 3.92 3.81

Jamaica 2.40 2.25 Brazil 2.78 2.75 Australia 3.94 3.79

Senegal 2.40 2.37 Mauritania 2.78 2.63 Belgium 3.96 3.89

Guinea-Bissau 2.41 2.28 El Salvador 2.83 2.66 Denmark 3.98 3.86

Yemn 2.41 2.29 Sudan 2.84 2.71 Canada 4.04 3.92

Cote d’Ivoire 2.44 2.36 Philippines 2.87 2.69 Ireland 4.06 3.91

Russian Fed. 2.44 2.37 Guinea 2.88 2.71 United Kingdom 4.08 3.99

Bolivia 2.46 2.31 Peru 2.91 2.77 Japan 4.09 4.02

Mali 2.46 2.29 Tunisia 2.92 2.76 Switzerland 4.10 4.02

Zimbabwe 2.46 2.29 Mexico 3.01 2.87 Hong Kong 4.10 4.00

Lebanon 2.48 2.37 Jordan 3.01 2.89 Austria 4.16 4.06

Zambia 2.50 2.37 Jordan 3.02 2.89 Germany 4.16 4.10

Comoros 2.51 2.48 Bulgaria 3.04 2.87 Sweden 4.18 4.08

Nigeria 2.51 2.40 Romania 3.05 2.91 Netherlands 4.25 4.18

Liberia 2.51 2.31 Kuwait 3.08 2.99 Singapore 4.28 4.19

Sri Lanka 2.51 2.40 Vietnam 3.09 2.89 Mean 2.92 2.80

Notes: LPI is World Bank Index. ILI is our own summary measure of the components affecting internationallogistics.Source: World Bank and author’s calculations.

TABLEA.2

Regression results for full sample of countries

Dependent variable: bilateral exports 1 2 3 4

OLS MR IV Contry,firm Firm

GDP 0.866*** 0.245 0.689*** 0.933***

[0.0123] [1.100] [0.118] [0.0362]

Logistics (ILI) 0.840*** −1.119 0.282 1.203***

[0.0864] [5.462] [0.466] [0.183]

Logistics developing 0.158 8.369 0.347 −0.0744

[0.102] [16.77] [0.252] [0.144]

Distance −1.363*** −1.418*** −0.979*** −1.588***

[0.0372] [0.112] [0.294] [0.101]

Border 0.795*** 0.840*** 0.718*** 1.027***

[0.149] [0.257] [0.202] [0.164]

Colony 0.300*** 0.375* 0.23 0.414***

[0.116] [0.216] [0.151] [0.127]

Language 0.713*** 0.798*** 0.449** 0.835***

[0.0726] [0.195] [0.194] [0.0723]

Samecountry 0.633*** 0.521 −3.901*** −3.683***

[0.226] [0.380] [1.215] [1.329]

Religion 0.634*** 0.528** 0.409** 0.740***

[0.111] [0.261] [0.180] [0.0894]

Landlocked −0.550*** −0.26 −0.477*** −0.564***

[0.0498] [0.582] [0.0601] [0.0534]

(continued overleaf )


30 Bulletin

TABLEA.2

continued

Dependent variable: bilateral exports 1 2 3 4

OLS MR IV Contry,firm Firm

Island −0.400*** 0.0931 −0.483*** −0.410***

[0.0484] [1.067] [0.0599] [0.0533]

High income 0.44 26.63 1.062 −0.376

[0.369] [54.68] [0.871] [0.513]

Constant −32.56*** −36.88*** −21.42*** −34.88***

[0.492] [9.049] [6.393] [1.800]

Inverse Mills Ratio −1.079**

[0.479]

Delta 0.584 −0.382**

[0.474] [0.162]

N 10057 10057 10057 10057

Instrument relevance † −0.001

Endogeneity test‡ 5.989***

adj. R2 0.679 0.324 0.68 0.68

Notes: Column 1 is estimated by OLS. Column 2 uses two instruments. Column 3 follows the HMR procedurewith the exclusion of the bilateral component of the same country variable (the MR component is included). Column4 excludes the bilateral component of the same country variable and the Inverse Mills Ratio.Standard errors in brackets (bootstrapped in columns 3–4).Significance levels: *P <0.1, **P <0.05, ***P <0.01.†Shea (1997) adjusted partial R2 for both variables; ‡ Wooldridge (1995) F-statistic.

Appendix B: Dropping high-income exporters

We explore the implications here of performing estimates on the full sample and including

a dummy for developing countries. Theoretically, if we want to capture a different effect

of a trade friction on a developing versus developed country, our equation becomes

mij =w+ yi + yj− cdij +k

2(Li +Lj)+

k′

2

(DiLi +DjLj

)+wij + ln

(PiPj

)r−1, (42)

where Dk =1 if country k is developing and k′ is the additional logistics coefficient. When

si =1/n, the MR effect of the dummies is captured by

MRDij=

∑

l

sl

∑

h /= l

sh(DlLl +DhLh)−

∑

h /= i

sh(DiLi +DhLh)−

∑

l /= j

sl(DlLl +DjLj)

=−2

nDL−

(DiLi +DjLj

)(n−2

n

)

where DL= 1

n

∑l

DlLl. Then

mij =w+ yi + yj− c(dij−MR dist

ij

)+

k

n(Li +Lj)+

k′

n

(DiLi +DjLj

)+wij +jMR

fij. (43)

A priori, we have econometric concerns regarding high-income exporters. First, as part of

the HMR procedure, very high values of qij can be predicted for many country pairs such



that they are practically indistinguishable from unity and from one another. HMR truncate

the values of qij above 0.9999999 such that there is a mass of estimates of gij and zij at a

particular value. Baranga (2009) notes this can have a big impact on estimates of d and spec-

ulates this may have been done to ensure d comfortably exceeds zero. Preliminary analysis

revealed that all our predicted values above 0.9999999 were generated by high-income

exporters and that high-income countries generally generated high predicted probabilities.

Rather than being stuck with indistinguishable values or truncating arbitrarily, we remove

this well defined group of countries. Second, including logistics and its interaction with the

income level means we have two potentially endogenous variables for which instruments

must be found. This places a greater demand on the instrumentation procedure.

Column 1 of Table A2 presents the OLS results for a homogeneous firms model with

MR controls. It now includes a dummy equal to one if the exporter is classified as a high

income country and an interaction between logistics and income level which applies only

to developing countries as in equation (43). This produces a higher logistics coefficient

than the analogous estimate in Table 2 together with insignificant estimates for the income

dummy and interaction. The insignificance may be due to high multicollinearity – the

variance inflation factors for the income dummy and the interaction are 118 and 72 respec-

tively – but could imply that income level makes no difference. However, the coefficient

is different to that for developing countries estimated alone. This may be due to differ-

ent coefficient estimates for the other trade costs variables. In other words, fully isolating

separate developing country effects may require interactions with income for many trade

cost variables and not just logistics. Unfortunately, IV estimates are unreliable. Using

the sum of start-up procedures in both exporter and importer as one instrument and their

product as another – as done before – yields very poor identification of the coefficients of

interest. They are all individually insignificant, but the coefficients imply logistics quality

reduces exports in rich countries and has a huge positive impact on developing country

exports.

In the heterogeneous firms model, we present two estimates where the bilateral com-

ponent of the Samecountry variable is excluded. Column 4 controls for country and firm

selection. The estimates implied for developing countries are individually insignificant

but jointly significant at the 5% level. Compared to the specification controlling for coun-

try and firm selection for developing countries, this assigns a relatively greater role to the

intensive margin and a smaller role to the extensive margin.As was the case for developing

countries only, specifications controlling for country and firm selection were not robust to

the choice of excluded variable. Unlike the developing country estimates, there is a large

and negatively significant IMR term. Therefore, it is not entirely surprising that attempting

to control only for firm-selection yields non-sensical results. Most importantly, it produces

a negative delta term, which is theoretically inconsistent and precludes the calculation of

country-level comparative statics.

While we have provided estimates based on the exclusion of the Samecountry var-

iable for consistency and brevity, we stress that these results are representative of a

range of alternative specifications, such as alternative exclusion restrictions and drop-

ping the high income dummy. We tried alternative instrument combinations and func-

tional forms for the IV estimates, which consistently yielded poorly identified results

as well as negative delta coefficients or IMR terms. We additionally estimated spec-


32 Bulletin

ifications on the full sample without the income dummy or interactions. The results

were more plausible for the IV estimates but equally untrustworthy for the heteroge-

neous firms models. Furthermore, the OLS coefficient on logistics (0.876) was higher

than that for the analogous developing country estimate. The results are available on re-

quest.
