World Development Vol. 39, No. 12, pp.

20692079, 2011
2011 Elsevier Ltd. All rights reserved
0305-750X/$ - see front matter
www.elsevier.com/locate/worlddev

doi:10.1016/j.worlddev.2011.05.013

Trade Policy, Trade Costs, and Developing Country Trade

BERNARD HOEKMAN
World Bank and CEPR, Washington DC, USA

and
ALESSANDRO NICITA *
UNCTAD, Geneva, Switzerland
Summary. This paper reviews some indices of trade restrictiveness and trade facilitation and compares the trade impact of dierent
types of trade restrictions applied at the border with the eects of domestic policies that aect trade costs. Based on a gravity regression
framework, the analysis suggests that taris and non-tari measures continue to be a signicant source of trade restrictiveness for
low-income countries despite preferential access programs. The results also suggest that behind-the-border measures to improve logistics
performance and facilitate trade are likely to have a comparable, if not larger, eect in expanding developing country trade, especially
exports.
2011 Elsevier Ltd. All rights reserved.
Key words TARIFFS, nontari measures, trade facilitation, logistics, economic development, DOHA Round

1. INTRODUCTION

and trade facilitation performance levels prevailing in middleincome countries. We nd that the latter will have a substantially larger positive impact on trade volumes than the former.
Section 6 concludes.

The trade policy literature has for many years emphasized

the importance of taking into account the impact of a variety
of sources of trade costs in addition to import tarissee For
example, Deardor and Stern (1998) and Anderson and van
Wincoop (2004). Recent research on trade and development
has emphasized the magnitude of the trade costs associated
with administrative red tape and entry barriers, informed by
the emergence of new datasets, such as the OECDs Product
Market regulation database, the World Banks Doing Business indicators and Logistics Performance Index (LPI), as
well as rm-level surveys of the investment climate, and business environment that prevails in countries.
In this paper we assess the impacts of dierent sources of
trade costs on international trade. The objective is to compare
the eect of border barriers (import taris, adjusted for bilateral preferences, and non-tari measures) with other sources
of trade costs. Our interest is to explore the relative impacts
on trade volumes of dierent sources of policy-induced trade
costs.
The plan of the paper is as follows. Section 2 summarizes the
current pattern of tari protection and the aggregate of all
non-tari measures (NTMs) captured in the UNCTAD database on NTMs (WITS). Section 3 discusses how we take into
account the extensive system of preferential trade that has
been put in place by OECD countries as well as the numerous
preferential trade agreements concluded between subsets of
WTO members. Section 4 discusses some of the components
of the aggregate NTM measure, as well as indicators of trade
facilitation performance: the trading across borders component of the World Banks Doing Business report, and the
Logistics Performance Index. These indicators reect regulatory policies that directly aect trade costs but are not captured by the tari and NTM databases commonly used by
analysts. Section 5 presents the results of an empirical assessment of the relative trade impacts of these dierent sets of policies and the possible trade eects of convergence by
developing countries to the average levels of border protection

2. TRADE POLICIES
Trade policies can be broadly divided into taris (ad-valorem
and specic) and non-tari measures. Although taris are still
the most widely used policy instrument to restrict trade, their
relative importance has been declining. Trade liberalization,
whether unilateral, the result of agreements negotiated under
the auspices of the World Trade Organization, or the consequence of preferential trade agreements (PTAs), has greatly reduced the average level of applied taris. Conversely, the use of
NTMs has been increasing both in terms of the number of products covered and the number of countries utilizing them (World
Bank and IMF, 2008). The use of taris, specic duties and
NTMs in 2006 is illustrated in Figure 1. In general, the use of
non-tari measures increases with the level of economic development of countries. This is particularly true for NTMs, which
are increasingly used to regulate trade, especially in high income
countries. Similarly, specic duties, although aecting only a
relatively small share of total imports, are more prominent in
high income countries.
The type of NTMs included in the analysis of this paper is
limited by the availability of data. In particular, as a measure
of NTMs we use the ad-valorem equivalent of NTMs estimated in (Kee, Nicita, & Olarreaga, 2009). This ad-valorem
* We are grateful to Alan Deardor, Simon Evenett, Sheila Page, and Ben
Shepherd for comments on an earlier draft, and to three referees of this
journal for comments that greatly improved the paper. The authors accept
sole responsibility for any errors remaining. The views expressed are personal and should not be attributed to the World Bank or the United
Nations Conference on Trade and Development. Final revision accepted:
March 28, 2011.
2069

WORLD DEVELOPMENT

.2
.1

.08

simple average tariff

.04

specific duties

10

% lines subject to: NTM and specific duties

non-tariff measures

.06

simple average tariff

.1

.12

.3

2070

12

log gdp per capita

Figure 1. Use of Taris, Specic Duties, and NTMs (% of HS six digit lines), 2006.

is in higher-income economies and the fact that their imports

are highly skewed toward manufactures, which face relatively
low barriers.
Agricultural trade is much more restricted than manufactured products, both in terms of the TTRI and the OTRI,
especially in high income countries. This reects both higher
taris and greater use of NTMs in agricultural trade. A comparison of Figures 2 and 3 reveals that NTMs contribute substantially to the set of policies restricting global trade,
especially in agriculture. Dierences in the OTRI and TTRI
are also evident across geographic regions (Table 1). In general, East Asian, Central Asian, and East European countries
are less restrictive, while countries in South Asia and the Middle East and North Africa are more restrictive. This pattern is
similar for the TTRI and the OTRI, and for agriculture and
manufacturing.
As a result of unilateral reforms and bilateral and regional
agreements, global trade has been substantially liberalized in
recent years. Figure 4 presents scatter plots of the TTRI for
the years 2000 and 2006. While liberalization has been substantial in most countries, tari reduction has centered more
on manufacturing than agricultural products. Agricultural

.15

equivalent of NTMs captures the eect of quantitative restrictions, technical product regulations, anti-dumping and countervailing measures, and any monopolistic measure or
discretionary licensing. 1
The ad-valorem equivalent of NTMs can be combined with
data on taris into an overall trade restrictiveness index
(OTRI) to capture the eect of both types of measures (Kee
et al., 2009). To isolate the eect of taris from the overall index we also calculate the tari trade restrictiveness index
(TTRI). The dierence between the TTRI and OTRI is that
the OTRI includes the eect of both tari and NTMs, while
the TTRI captures only taris, both ad valorem and the ad
valorem equivalents of specic taris. 2 These indices are calculated on a bilateral basis, using the eectively applied tari
and taking into account all preferential regimes. Both the
OTRI and the TTRI are a measure of the uniform tari equivalent implied by observed trade policies aecting a countrys
importsthat is, they represent the ad-valorem tari that
would be needed to generate the observed level of trade. 3
The prevailing average TTRI and OTRI across countries is
plotted in Figures 2 and 3. Trade policies are generally more
restrictive in lower-income countries, reecting both lower tar-

TTRI

.1

Agriculture

Manufacturing

.05

All Goods

10

log gdp per capita

Figure 2. TTRI and GDP per capita, 2006.

12

2071

.3

.4

TRADE POLICY, TRADE COSTS, AND DEVELOPING COUNTRY TRADE

.2

OTRI

Agriculture

.1

All Goods

Manufacturing

10

12

log gdp per capita

Figure 3. OTRI and GDP per capita, 2006.

Table 1. OTRI and TTRI by developing country region, 2006

Region (developing countries only)

Total trade (%)

Agriculture (%)

Manufacturing (%)

East Asia

11.3
5.0

26.6
8.7

10.4%
4.8

Europe and Central Asia

10.1
4.5

25.9
10.3

9.0
4.0

Latin America

15.0
5.4

28.1
6.6

13.8
5.3

21.6
11.9%

32.3
12.1

19.4
11.8

South Asia

19.5
14.0

46.4
31.4

18.2
13.2

Sub-Saharan Africa

14.4
8.4

24.9
13.8

12.9
7.6

Middle East and N Africa

Note: OTRI gures in bold and TTRI gures in italics.

trade restrictiveness increased for some countries between

2000 and 2006. 4 The TTRI has declined both for all country
groups. Middle income economies have seen the largest decline, including in agriculture. By region, countries in East
Asia and Latin America (sub-Saharan Africa) have reduced
taris the most (least).
3. MARKET ACCESS AND THE EFFECT OF TRADE
PREFERENCES
The eect of trade policies on exporters access to markets
diers across trading partners and geographic regions. To
measure the average restrictiveness that exporters face in a
particular market the OTRI and TTRI discussed above are
computed in terms of exports. Those are reported in Table 2
where restrictiveness is provided by a market access perspective. We label these measures MA-TTRI and MA-OTRI.
Those can be thought as the equivalent uniform tari that applied to exports of a given country would result in the observed level of exports. These measure the restrictiveness of
policies confronting exporters from in each geographic region
and country group by taking into account preferential trade
regimes. Upper middle income countries generally enjoy better
market access in both developing and developed countries.

This is largely due to the composition of exports from these

countries, which are skewed toward manufacturing. Low income countries face more restrictive market access conditions
because their exports are more biased toward agriculture.
Across developing country regions, South Asia faces the
most restrictive market access, due to export composition
(agriculture, textiles and apparel) and because it has relatively
limited preferential access. Sub-Saharan countries have the
best market access, especially in high income countries, reecting again export composition (minerals, primary products,
plantation agriculture), as well as low or zero (preferential)
taris in many high income countries. By far the highest levels
of market access barriers apply to South-South trade ows.
Sub-Saharan African countries confront TTRIs and OTRIs
in low-income countries that are 3 to 4 times higher than those
that apply in middle- and high-income markets.
Comparing the MA-TTRI and the MA-OTRI suggests that
NTMs are generally more important in restricting trade than
taris: their measured ad-valorem equivalent is much higher
than existing taris. Standards, licensing requirements and
similar regulatory instruments typically aect all products
entering a market regardless of their origin, so that the impact
of NTMs is relatively similar across trading partners. Taris,
conversely, are generally negotiated on a bilateral basis, thus
giving some trading partners a substantial advantage in

WORLD DEVELOPMENT

.3
.2
0

.1

.2
.1
0

TTRI 2000

.3

Agriculture TTRI 2000

.4

.4

2072

.1

.2

.3

.4

.1

.3

.4

.4
.3
.2
.1
0

Manufacturing TTRI 2000

.2

Agriculture TTRI 2006

TTRI 2006

.1

.2

.3

.4

Manufacturing TTRI 2006

Figure 4. TTRI 2000 and 2006.

Table 2. Market access TTRI and OTRI, 2006

Importing countries

Exporting countries
High
income

Upper middle
income

Lower middle
income

Low
income

East
Asia

E. Europe
Cent. Asia

Latin
America

Mid. East
N. Africa

South Asia

Sub-Saharan
Africa

High income

6.3
2.4

5.7
1.2

7.9
2.5

9.1
2.4

8.3
2.6

5.1
1.1

7.0
1.5

4.3
0.8

10.4
3.1

4.4
0.7

QUAD

6.3
2.1

5.2
0.9

8.6
2.5

10.6
2.5

8.9
2.7

5.2
0.8

6.9
1.2

4.4
0.5

13.6
3.3

4.5
0.5

Upper middle

15.6
5.6

11.8
3.8

15.8
5.6

14.7
5.7

19.2
7.2

10.2
4.4

13.6
2.6

6.0
2.5

14.3
6.6

5.9
3.5

Lower middle

12.4
7.1

11.1
4.8

12.9
6.7

9.4
5.1

13.6
6.6

11.2
6.2

12.6
5.1

6.7
2.8

9.9
6.2

4.0
2.7

Low income

18.2
10.9

14.3
8.1

19.5
12.2

25.4
12.9

22.2
13.8

17.7
6.2

15.9
9.0

16.3
10.0

16.2
10.4

16.3
12.2

Note: MA-OTRI in bold; MA-TTRI in italics. QUAD countries are Canada, European Union, Japan, and USA.

market access. With the increase in reciprocal and nonreciprocal preferential trade agreements, almost all trade ows today
are aected by some sort of tari preference. This is particularly true for high income countries, where market access is affected by an increasing number of such agreements.
The proliferation of preferential trade arrangements makes
it important to properly measure preferential margins in
assessing the relative market access conditions confronting
exporters. This is done in the TTRI and OTRI in a direct
way, as the calculations take into account the bilateral market
access conditions that apply. But what matters for a given
country, however, is the relative preference (the relative market access conditions), not just the absolute level of prevailing
barriers at the border.
Commonly used measures of preference margins compare
the preferential tari to the most-favored-nation (MFN) rate.
This will overestimate the relative preference enjoyed by coun-

tries as in most instances other countries will also have preferential access. In practice it is possible that preferential rates
granted to a particular country, although lower than MFN,
still penalize it relative to other countries that benet from
even lower or zero taris. To calculate the relative preferential
margin the focus needs to be on the average advantage in
tari percentage points that a given basket of goods enjoys
when exported from country A as compared to when it originates in other countries.
To clarify with an example, in what follows we calculate the
relative preferential margin that Mexico enjoys in the US by
using as the counterfactual the average tari for Mexicos export bundle if this were to originate in other countries. The relative preferential margin is the dierence between the bilateral
trade-weighted preferential tari imposed by the US on
Mexico and that counterfactual. There are two sets of weights
when calculating this margin: rst, the counterfactual, which is

TRADE POLICY, TRADE COSTS, AND DEVELOPING COUNTRY TRADE

a weighted average of taris imposed on all other (potential)

exporters to the US; and second, the preferential margin,
which is an average constructed across many tari lines.
To measure the counterfactual, we rst calculate the tradeweighted average tari at the tari line level that an importer
(the US) imposes on all other countries except the country for
which the preferential margin is calculated (Mexico). This is
done by using (US) bilateral imports as weights, so as to take
into account the supply capacity of (US) trading partners. We
then aggregate across tari lines using (Mexican) exports (to
the US) to take care of dierences in product composition
across partners. 5
A simpler alternative measure would be to compare the (US)
import weighted average tari imposed on a country (Mexico)
with that imposed on all other countries. 6 This approach uses
total imports (by the US) at the HS six digit level as weights. A
problem with this method is that it disregards product composition: if Mexicos export bundle to the US is not representative of the composition of US imports (e.g., Mexican exports
to the US are mainly agricultural, while US imports are
mainly manufactured goods) using exclusively US imports as
weights in the calculation of the counterfactual would likely
lead to biased results. 7
A further complication arises in the aggregation across tari
lines. A proper aggregation would take into account that imports of some goods are more responsive to changes in prices
than others. In theory, imports that are less sensitive to prices
(inelastic) should be given less weight as taris change as
they would have little eect on overall volumes of trade. 8
To correct for this, HS six digit product lines are aggregated
using the import demand elasticities. 9
The relative preferential margin (RPM) for exports from
country j is then calculated as:
P k k !
mimpm;hs tm;hs
PP
impjk;hs k;hs P mimpk
RPM j

m;hs

hs

PP
k

PP

impjk;hs k;hs

hs

impjk;hs k;hs tkm;hs

 PP
; mj
impjk;hs k;hs
k

hs

hs

where, imp are imports, e is the import demand elasticity, t is

the tari, k indexes importers, hs are HS 6 digit categories, and
v are exporters competing with country j in exporting to country k. The second term in the equation is simply the MA-TTRI
while the rst term is the MA-TTRI but calculated on the basis of taris applied to competitors of country j. This index can
be calculated bilaterally (i.e., Mexicos relative preferences in
the US market) or at the country level (i.e., Mexicos overall
level of relative preferences for its exports relative to all its export markets). In the latter case all bilateral trade ows of a
particular country are considered.
This measure of preference margin can be positive or negative, depending on the advantage or disadvantage of the country with respect to other exporters. Table 3 reports relative
preferential margins averaged by region. All regions have positive relative preferential margins with themselves. This indicates the importance of regional trade agreements. The most
eective regional agreements in terms of preferences are in
Latin America, where countries enjoy a relative preferential
margin of about 3%. Latin America both enjoys and provides
a substantial preferential margin to the USA and Canada,
reecting trade agreements with the US and within the region
(MERCOSUR, etc.). This is mirrored by the negative prefer-

2073

ence that countries outside Latin America face when they export to that region. Relative preference margins, whether
positive or negative, are much smaller for other regions. Countries in Sub-Saharan Africa, for example, enjoy a relative preferential margin of only about 0.5% in the EU, as they compete
both among themselves and other countries to which the EU
provides preferences (Eastern Europe, North Africa and Latin
America). Relative preferential margins are mostly negative
for East Asian states.
Country-specic estimates of the relative preference margin
for a number of Sub-Saharan countries are reported in Table
4. Bilateral preferences are substantial in only a few cases, and
in a few instances they are actually negative, putting the countries concerned in a situation similar to that applying to East
Asian economies. Only Madagascar has relatively large preferential margins in more than two markets, while most countries
have meaningful preferential margins in only one or two markets, if any.
4. OTHER TRADE COSTS
The foregoing discussion illustrates that NTMs as well as
taris are a signicant source of barriers to trade. A question
to which we return below concerns the relative importance of
various barriers, especially from a developing country perspective. In particular, given the still high taris in agricultural
products, remaining taris may still have large eects relative
to NTMs, especially in the case of South-South trade ows.
On the other hand, the average impact of NTMs in regulating
imports into higher-income countries clearly suggests that action to reduce their trade-impeding eects could have high
payos.
Besides taris and NTMs, we are also interested in the impact of other trade related costs. Internal trade and transactions costs may be of equal if not greater importance in
reducing volumes of trade. Many of these trade costs reect
the domestic economic environment: the legal and regulatory
framework, the eciency of infrastructure services and related regulation, customs clearance procedures, administrative red tape, etc. Data on domestic trade costs are often
lacking. However, the World Bank has recently initiated
the collection of data for a large number of developing countries on the performance of logistics services and on the internal costs associated with shipping goods from the factory
gate to the port, and from ports to retail outlets. The rst
is captured by the Logistics Performance Index (World Bank
2007); the second is covered by the Doing Business database
(World Bank, 2008). All of these indicators capture dimensions of prevailing domestic regulatory regimes that aect
trade.
The Doing Business cost of trading indicator measures the
fees associated with complying with the procedures to export
or import a 20-foot container, measured in US dollars
(Djankov, Freund, & Pham, 2010). These include costs for
documents, administrative fees for customs clearance and
technical control, terminal handling charges and inland transport. The cost measure does not include taris or trade taxes
and only ocial costs are recorded. The indicator is part of
the Doing Business trading across borders index. The methodology, survey instruments and data are available at http://
www.doingbusiness.org.
The Logistics Performance Index (LPI) provides a snapshot
of the supply chain performance of countries. Based on a
worldwide survey of global freight forwarders and express
carriers, the LPI measures the logistics friendliness of the

2074

WORLD DEVELOPMENT
Table 3. Relative preference margins, 2006 (percentage points)

Importers
East Asia
East Europe Central Asia
Latin America
Middle East and North Africa
South Asia
Sub-Saharan Africa
High Income Countries
Australia and New Zealand
Canada
European Union
Japan
USA

East
Asia

East Europe
Central Asia

Latin
America

Middle East and

North Africa

South Asia

Sub-Saharan
Africa

High income
countries

0.22
0.01
2.54
0.29
0.21
0.10
0.46
0.18
1.00
0.05
0.34
0.67

0.06
0.45
1.88
0.24
0.08
0.03
0.42
0.61
0.85
1.07
0.02
0.03

0.09
0.37
2.98
0.25
0.04
0.06
0.71
0.28
1.75
0.98
0.07
1.01

0.02
0.39
0.51
0.91
0.26
0.02
0.19
0.08
0.01
0.64
0.00
0.08

0.03
0.20
2.13
0.22
2.03
0.12
0.46
0.23
1.79
0.70
0.70
0.22

0.01
0.04
1.22
0.10
0.15
0.30
0.13
0.11
0.02
0.51
0.08
0.11

0.03
0.15
1.69
0.03
0.05
0.06
0.08
0.10
1.01
0.50
0.13
0.03

Table 4. Relative preference margins for selected african countries in high income markets, 2006 (percentage points)
Exporter
Angola
Benin
Burkina Faso
Cent.African. Rep
Chad
Cote DIvoire
Cameroon
Congo
Ethiopia
Ghana
Kenya
Madagascar
Mali
Mozambique
Mauritania
Malawi
Niger
Nigeria
Rwanda
Sudan
Senegal
Togo
Tanzania
Uganda
Zambia
Zimbabwe

Australia and New Zealand

Canada

European Union

Japan

USA

Other High Income Countries

4.70
0.00
4.23
1.93
4.61
0.04
0.02
0.00
0.11
0.15
0.04
1.43
1.89
0.54
0.37
0.04
2.27
0.29
4.20
1.33
1.41
0.00
0.15
0.03
0.37
0.10

0.00
2.72
1.02
0.65
0.78
0.04
0.16
0.01
0.34
0.06
0.48
7.53
1.63
0.17
4.37
0.35
0.52
0.03
0.14
0.00
0.60
0.06
0.00
0.16
0.82
0.04

0.04
0.02
0.60
0.06
0.08
0.36
0.40
0.07
0.50
0.92
1.25
3.89
0.43
4.46
0.37
0.02
0.04
0.06
0.02
0.06
2.29
0.32
1.13
1.45
0.13
0.60

0.03
0.23
0.02
0.07
0.00
0.04
0.01
0.08
0.06
0.02
0.08
0.83
2.84
0.53
6.95
0.05
0.43
0.00
0.20
0.00
2.99
0.28
0.05
0.22
0.82
0.01

0.08
0.18
0.40
0.03
0.17
0.02
1.16
0.27
0.01
0.38
0.92
0.97
0.24
0.18
0.00
2.96
2.01
0.13
0.05
0.00
0.22
1.42
0.32
-0.01
0.56
0.02

0.02
0.02
0.27
0.03
0.01
0.24
0.27
0.01
0.76
0.59
0.68
2.92
0.29
2.76
0.22
0.01
0.02
0.05
0.01
0.03
0.98
0.12
2.33
1.22
-0.32
0.25

countries surveyed. Feedback from the survey is supplemented with data on the performance of key components
of the logistics chain. For the analysis in this paper we use
the overall LPI score. This score summarizes six sub-indicators (eciency of clearance process, quality of trade and
transport infrastructure, ease of arranging competitively
priced shipments, competence and quality of logistic services,
ability to track and trade consignments, and timeliness of
shipments within the expected delivery time. The LPI indicators range from 1 (worst) to 5 (best). In practice, the LPI
score varies from 1.2 (Afghanistan) to 4.2 (Singapore). The
underlying methodology and data on the LPI are available
at http://www.worldbank.org/lpi.
Table 5 reports the average of these indices by income country groups. Developing countries generally have weaker trade
facilitation performance than higher-income economies.

5. AN EMPIRICAL ASSESSMENT
The literature on trade costs has recently devoted some
attention to the role that domestic costs play in aecting international trade. The impact of dierent sources of trade costs is
frequently assessed through the inclusion of specic variables
(generally measures of infrastructure availability such as
roads, railways, phone lines, etc.) in gravity type models. In
general, the literature supports the hypothesis that domestic
trade costs and the economic business environment are significant determinants of the volume of trade between countries
(Limao & Venables, 2001; Wilson, Mann, & Otsuki, 2003;
Anderson & Marcouiller, 2002; Francois and Manchin,
2007). What follows builds on the existing gravity model literature to investigate the importance of trade and related regulatory policies on trade ows.

TRADE POLICY, TRADE COSTS, AND DEVELOPING COUNTRY TRADE

2075

Table 5. Measures of trade costs (averages by country group)

Policy/indicator
Trade policy tari (imposed) (%)
Trade policy NTB (imposed) (%)
Trade policy tari (faced) (%)
Trade policy NTB (faced) (%)
Trade policy RPM (%)
Doing business costs of exports (USD)
Doing business costs of import (USD)
Logistic performance index (higher better)

High income

Middle income

Low income

1.98
4.45
2.77
4.56
0.04
774.40
813.60
3.90

6.86
4.45
2.79
4.72
0.44
867.20
1024.20
3.00

10.22
6.71
6.02
5.07
0.23
949.30
1212.00
2.80

To capture the eect of traditional trade policies as opposed

to other trade-related costs we use a traditional cross-section
gravity model that includes time invariant trade impediments
(distance, adjacency, common language, access to the sea)
augmented with a set of trade policy variables. Border barriers
are captured by the TTRI, the NTM component of the OTRI
(dened as the dierence between the OTRI and the TTRI)
and the relative preferential margin (RPM). Deeper economic
integration is controlled for by adding a dummy for the existence of a bilateral trade agreement. Domestic trade costs are
captured by the two trade facilitation indicators discussed
above: the LPI and the Doing Business cost of trading. 10
The dataset covers 105 countries. All data are for 2006.
Trade ow data are from WITS (i.e., United Nations
COMTRADE), GDP data are from the World Banks World
Development Indicators, the standard gravity variables are
obtained from the Trade Production and Protection database
(Nicita & Olarreaga, 2007), and trade policy data are obtained
from the OTRI database (Kee et al. 2008; Kee et al. 2009). Finally, data on the LPI and domestic trade costs were sourced
from the LPI and the Doing Business websites.
The approach used to estimate the gravity model follows the
existing literature. However, we are subject to the limitation of
a cross-sectional dataset. Thus the presence of unobserved relative trade impediments that a country has with all its trading
partners is controlled for by using multilateral resistance terms
rather than the country-time xed eect commonly used in panel settings. Thus, multilateral resistance (Anderson & Van
Wincoop, 2003) is proxied by adding multilateral resistance
variables as in Baier & Bergstrand, 2009 and Baier, Bergstrand
and Mariutto, 2010. 11 This methodology produces consistent
estimates and, contrary to using country xed eects, allows
the estimation of the impact of country specic factors such
as the LPI score and the cost of trading as measured in Doing
Business.
A recurring problem with gravity estimation is the presence
of zero trade observations. As the gravity model is estimated
in a log-normal specication, it will discard observations
where there is no trade. Recent procedures to take into
account zero trade ows are the Poisson estimation (pseudoPoisson maximum likelihoodPPML), or a two-stage estimation procedure. Although the incidence of zeros in our sample
is limited (trade ows occur across about 90% of countries),
our preferred model is PPML. This model, besides being
robust to truncation, also controls for heteroskedasticity by
using robust standard errors. 12
Table 6 reports the trade related PPML estimated coecients for a series of specications with and without the
domestic trade costs terms. We start with the simpler specication consisting in a traditional cross section gravity model,
augmented with two trade policy variables: the TTRI discussed above and the OTRI purged by the TTRI component
so as to isolate the marginal eect of NTMs. This specication

controls for multilateral resistance with the a-theoretical

terms. 13 Specication 2 is identical to 1 but multilateral resistance is controlled for by adding multilateral resistance terms
as in (Baier & Bergstrand, 2009). 14 Specication 3 is identical
to 2 but adds the RPM term, while specications 4 and 5 add
the Doing business and the LPI terms, respectively. Finally,
specication 6 includes all trade policy and trade cost variables.
Results are generally typical of those of gravity equation
models. Distance is an important determinant of bilateral
trade, as is a common border and common language. Landlocked countries tend to trade less and larger and more populous countries tend to trade more. With regard to trade policy
variables the coecients are of the expected sign and are signicant across all specications. In particular, countries that
are partners in a reciprocal trade agreement are found to have
about 20% more trade. Both taris and NTMs are statistically
signicant determinants of bilateral trade. On average, a
reduction in taris (as measured by the TTRI) of 10% would
increase trade volumes a little more than 2%, while a similar
reduction in NTMs would add another 1.7% (as in specication 2). 15
Relative preferences matter a lot. Countries with an average
1% point advantage over competitors are estimated to trade
about 2.3% more with their counterpart. This suggests that,
even controlling for deeper integration (the bilateral agreement), tari preferences are signicant not only in and of
themselves (in terms of aecting the TTRI), but also in regard
to the relative advantage that is provided versus other countries (the RPM). Moreover, the inclusion of the relative preference margin results in somewhat lower coecient estimates
for the eect of a bilateral trade agreement. This suggests that
bilateral trade agreements have impacts that go beyond that
provided by nonreciprocal tari preferences.
Besides trade policy, the impact of trade costs on trade ows
is also estimated to be sizeable. Internal trade costs, as captured by the Doing Business indicator, are found to be similar
in magnitude for imports and exports. The elasticity of imports to the domestic cost of importing is about 0.50, and that
of exports to the domestic cost of exporting is about 0.48. That
is, a 10% reduction in the cost associated with importing
(exporting) would increase imports (exports) by about 5%
(4.8%). Signicant and qualitatively similar results are also
found in regard to the LPI score. However, the larger coecient on the LPI for exporters indicates that the quality of
logistics matters relatively more for them. One possible explanation could be that competition among exporters is generally
more intense, so that better logistics services can provide more
of an edge. The last specication (6) includes both the Doing
Business indicators and the LPI score. The results suggest
some overlap in the costs captured by these two indices as
the coecient estimates for specication 6 are lower than in
specications 4 and 5. Nonetheless, the fact that all coecients

2076

WORLD DEVELOPMENT
Table 6. Gravity model results.

Variable

Poisson Maximum Likelihood Estimation(Dependednt variable, volume of trade)

Expected Sign

GDP Importer
(log)
GDP Exporter
(log)
Population Importer
(log)
Population Export
(log)
Multilateral Resistance
(Distance)
Multilateral Resistance
(Board)
Land Importer
(dummy)
Land Exporter
(dummy)
Distance (log)

+
+
+
+
+
+
+
+





Common Board
(dummy)
Common language
(dummy)
Bilateral trade agreement
(dummy)
Trade policy tari (log)

Trade policy NTM(log)

Trade policy RPM

DB import costs(log)

DB import costs (log)

LPI importer(index)

LPI exporter(index)

Pseudo R-squared
Observations

+
+
+

(1)
***

0.837
(0.041)
0.739***
(0.030)
0.056
(0.040)
0.146***
(0.059)
1.246***
(0.163)
1.413***
(0.131)
0.019
(0.131)
0.078
(0.100)
0.784****
(0.054)
0.428***
(0.143)
0.038**
(0.140)
0.199**
(0.092)
0.209***
(0.041)
0.185***
(0.053)

0.872
10920

(2)
***

0.722
(0.045)
0.620***
(0.024)
0.089**
(0.042)
0.191***
(0.048)
0.007***
(0.001)
0.005***
(0.015)
0.030
(0.139)
0.109
(0.104)
0.786***
(0.054)
0.216
(0.159)
0.228
(0.171)
0.227**
(0.088)
0.211***
(0.041)
0.172***
(0.054)

0.874
10920

(3)
***

0.724
(0.045)
0.622***
(0.023)
0.084**
(0.042)
0.187***
(0.047)
0.007***
(0.001)
0.004***
(0.015)
0.037
(0.138)
0.106
(0.104)
0.784***
(0.054)
0.184
(0.159)
0.210
(0.166)
0.199**
(0.087)
0.204***
(0.041)
0.173***
(0.054)
0.023**
(0.011)

0.879
10920

(4)
***

0.701
(0.041)
0.617***
(0.022)
0.077**
(0.036)
0.168**
(0.033)
0.006**
(0.001)
0.001***
(0.014)
0.012
(0.131)
0.061
(0.103)
0.748***
(0.050)
0.331**
(0.156)
0.228
(0.165)
0.222***
(0.081)
0.176***
(0.044)
0.134***
(0.047)
0.023**
(0.010)
0.496***
(0.094)
0.485***
(0.108)

0.888
10920

(5)
***

0.552
(0.041)
0.318***
(0.058)
0.209***
(0.049)
0.354***
(0.061)
0.008***
(0.001)
0.044**
(0.018)
0.080
(0.132)
0.168
(0.104)
0.701***
(0.052)
0.374**
(0.152)
0.270*
(0.162)
0.203***
(0.085)
0.221***
(0.042)
0.167***
(0.050)
0.030***
(0.008)

0.671***
(0.142)
0.905***
(0.168)
0.885
10920

(6)
0.606***
(0.069)
0.430***
(0.053)
0.149***
(0.051)
0.312***
(0.051)
0.007***
(0.001)
0.029*
(0.015)
0.032
(0.129)
0.126
(0.105)
0.705
(0.050)
0.401
(0.150)
0.261
(0.161)
0.213***
(0.083)
0.198**
(0.044)
0.146***
(0.046)
0.027***
(0.008)
0.324***
(0.094)
0.222
(0.096)
0.408***
(0.149)
0.701***
(0.150)
0.897
10920

Note: Robust standard errors are in parentheses. Multilateral resistance terms in the rst specication are a-theoretical: partner GDP-weighted averages of
each importer and exporter distance from all of its trading partners.
*
Signicance level of 10%.
**
Signicance level of 5%.
***
Signicance level of 1%.

maintain their signicance suggests that each is capturing distinct dimensions of the trade facilitation environment.
Although PPML is often used in gravity model estimation,
other models may t the data better. As a robustness check,
we also estimate the model in standard OLS (by adding 1 to
the trade values and taking the log), zero inated Poisson
(ZIP), negative binomial regression (NBREG), and zero inated negative binomial regression (ZINBR) as in Burger,
van Oort, and Linders (2009). 16 Table 7 reports the relevant
coecients for specication 6 of table 6 estimated with these
alternative models. 17 In general, the coecients are similar
across the various models. This is partly due to the limited presence of zero observations in our dataset (around 10%). However, while the results of the PPLM and the ZIP are quite
similar, some dierences arise with respect to OLS, NBREG
and ZINB. The alternative models indicate that taris play a

relatively more important role than NTMs. The NBREG and

ZINB models, but not OLS, also suggest lower impacts of
domestic trade costs (Doing Business and LPI). All in all, however, the models suggest qualitatively similar results: domestic
trade costs are a signicant determinant of trade ows. 18
From a policy perspective a relevant question is how much
dierent sources of trade costs matter for low income countries. To get a sense of this we can assess the magnitude of
the trade-impeding eect of border trade policies and internal
trade costs. Table 8 reports some simple calculations of the impacts of trade policy on trade ows. The calculations give the
increase in trade for low income countries if they were to converge to a set of policies that would generate the observed
average levels of the various policy indicators in middleincome countries (as reported in Table 5). The calculations
are based on the results of specication 6 in Table 6. 19

TRADE POLICY, TRADE COSTS, AND DEVELOPING COUNTRY TRADE

2077

Table 7. Robustness checks: alternative models

Trade policy tari (log)

Trade policy NTB (log)
Trade policy RPM (index)
DB Import Costs (log)
DB Export Costs (log)
LPI importer (index)
LPI exporter (index)

0LS

PPML

NBREG

ZIP

ZINB

0.315***
(0.025)
0.030
(0.025)
0.016*
(0.010)
0.098*
(0.057)
0.394***
(0.057)
0.357***
(0.083)
1.182***
(0.087)

0.198**
(0.044)
0.146***
(0.046)
0.027***
(0.008)
0.324***
(0.094)
0.222**
(0.096)
0.408***
(0.149)
0.701***
(0.150)

0.312***
(0.035)
0.053*
(0.028)
0.023**
(0.010)
0.240***
(0.078)
0.201***
(0.070)
0.279**
(0.109)
0.135
(0.108)

0.197***
(0.044)
0.146***
(0.046)
0.027***
(0.008)
0.326***
(0.094)
0.224**
(0.096)
0.403***
(0.149)
0.695***
(0.150)

0.299***
(0.032)
0.054**
(0.026)
0.025**
(0.010)
0.245***
(0.070)
0.168***
(0.065)
0.300***
(0.100)
0.150
(0.101)

Standard errors are in parentheses.

*
Signicance level of 10%.
**
Signicance level of 5%.
***
Signicance level of 1%.

Table 8. Eects of policy convergence by low income countries to middle

income average (percent increase)
Policy/Indicator
Trade Policy Tari
Trade Policy NTB
Trade Policy RPM
Doing Business
Logistic Performance Index

Increase in
Imports (%)

Increase in
Exports (%)

6.5
4.9
0.6
5.0
8.5

10.6
1.0
0.6
1.9
15.1

These results suggest that traditional trade policies still have

a substantial trade-restricting eect for low income countries.
A reduction of their TTRI to the level observed in middleincome countries would increase their imports by about
6.5%. Improving market access through reductions in taris
in export markets also remains an important policy objective.
If low income countries were to improve their market access to
that which confronts middle income countries this would increase their exports by more than 10%. NTMs are also a
strong determinant of trade ows, but our simulation suggests
that their impact is generally less than that of taris. In the
case of exports in particular the limited impact is due to the
fact that the level of NTMs faced by low- and middle-income
countries is similar. Thus, this result should not be interpreted
as NTMs not being an important factor aecting trade ows.
The eects of relative preferences, for low income countries are
on average very small. This is driven by the small dierence in
the average relative preference margin for middle- and low-income countries (0.2%). Relative preferences are important on
a country by country basis, not across the averages that apply
to each of broad grouping by income.
Domestic costs, measured by the Doing business and the
Logistic Performance Index are quantitatively important. Of
the two, the LPI has the largest eect on trade. The domestic
trade costs captured by the Doing Business variable have less
of an impact on trade. However, given that the Doing Business
and the LPI indicators capture, at least in part, similar domestic trade costs, their relative importance is dicult to assess. A
more proper interpretation is to use their sum as the total effect of domestic costs. This suggests that trade costs represent
an important bottleneck for low income countries, both in regard to imports (13.5%) and to exports (17%).

As a caveat, it should be borne in mind that these results

originate from a regression based on cross sectional data. This
generally does not allow unambiguous conclusions regarding
the direction of causation, as issues of endogeneity may not
be completely controlled for. However, the results clearly indicate that there is strong correlation between trade costs, both
domestic and at the border, and trade ows. In general terms,
these results indicate that administrative and regulatory policies are at least as important as traditional border policies in
impeding trade. This supports the recent focus of many developing countries on taking action to facilitate trade by reducing
domestic trade costs.
6. CONCLUDING REMARKS
The focus of policy debates and international cooperation is
more and more on non-tari measures, regulatory policies,
and on eorts to facilitate trade. In this paper we investigate
the impact on trade of a subset of such policies, and compare
these with the eects of traditional trade policies. Our ndings
indicate that while traditional trade policies continue to be
important in developing countries as well as for some sectors
in high-income countries (agriculture in particular), non-tari
measures and domestic trade costs are also of great importance. In particular, the analysis in this paper suggests that
the impact of reducing the costs associated with policies that
reduce the behind-the-border domestic cost of trade could
have a greater payo than further reductions in border barriers such as taris and NTMs, or seeking additional trade preferences. Our results imply that focusing attention on the
policies that aect logistics performance and the internal cost
of trading is likely to generate large trade gains, especially in
terms of exports. Given that the LPI and the Doing Business
cost of trading capture the eect of a multiplicity of specic
policies, from a practical perspective this suggests that a
priority for both research and policy advice is to further
unpack these ndings.
The analysis also makes clear that there are still large trade
gains to be had from traditional trade liberalization. While
bilateral and regional trade agreements have proliferated, progress at the multilateral level has unfortunately been slow.
Bringing the Doha Round to a successful conclusion and thus
capturing some of these potential trade gains is, therefore,

2078

WORLD DEVELOPMENT

important. Making progress to reduce internal trade costs

does not require or hinge on multilateral (or bilateral) negotiations the costs that are incurred by traders in developing
countries can and should be reduced through unilateral

actions. The analysis in this paper strongly supports the

argument made by Ikenson (2008) that there is great scope
to enhance economic growth opportunities while Doha
sleeps.

NOTES
1. Although important, other types of NTMs such as rules of origin and
government procurement could not be included in the analysis of this
paper as comprehensive and suitable data are lacking.

trading. Although this is an important question that is the subject of recent

research (Lawless, 2010; Bernard et al., 2009), it is beyond the scope of this
paper.

2. The inclusion of NTMs in the OTRI is done through estimation of advalorem tari equivalents. Both the TTRI and the OTRI provide a
measure of the uniform tari equivalent of observed policies that is needed
to generate the observed level of trade for a country. See Kee et al. (2009)
for details.

11. EachP of these

(var) is constructed as
P bilateral
P P variables
m
MRvarij k varik m varm
k
m vark .
j

3. These indices are superior to more commonly used indicators such as

average taris or NTM frequency and coverage ratios as they take into
account the elasticity of import demand with respect to prices. In
calculating the indices more weight is given to products for which demand
is more responsive to changes in prices (so that smaller movements in
prices produce larger shifts in imports).

13. That is, by adding ad hoc remoteness variables constructed as the

GDP-weighted averages of each countrys distance from all of its trading
partners (Wei, 1996; Baier & Bergstrand, 2004.). We also estimated the
model by alternatively using importer and exporter xed eects. This
resulted in qualitatively similar coecients.

4. As NTM data have not been updated recently, the change in the OTRI
is not reported.

14. Following the existing literature, we add two terms for multilateral
resistance in our preferred specication: distance and border. Results
obtained by adding two additional terms (language and trade agreements)
were similar.

5. As trade ows are generally reported at the six digit level of the
Harmonized System (HS), this indicator is constructed at the HS 6 digit
level instead of the tari line level.
6. This is the approach taken by Low, Piermartini, and Richtering
(2008).
7. This simpler methodology would be consistent with a framework
where export composition is a function of the structure of the tari of the
importing country (everything else equal, exports would concentrate in
product lines with lower taris), however, as the intent is to assess the
preferential margin applied on the existing structure of trade, the measure
controlling for product composition is to be preferred. Also, export
composition is more likely to be determined by other factors aecting
comparative advantage, such as level of development, land quality,
climate, geography and labor force skills.
8. Complicating the framework even more, one should take into account
substitution possibilities across similar products originating in dierent
countries. For simplicity, we abstract from this and assume that
substitution elasticities are equal to one.
9. See Kee, Nicita, and Olarreaga (2008) for the methodology used to
estimate import demand elasticities.
10. This approach allows us to assess the impact of border and domestic
trade impediments on international trade. However, it does not allow
disentangling how these impediments aect the xed and variable costs of

12. See Silva and Tenreyro (2006) for the rationale behind Poisson MLE.

15. The eect of NTMs is captured at the margin, that is given the eect
of the existing tari structure.

16. For ZIP and ZINBR the presence of a zero trade ow is determined
by the standard gravity variables.
17. Results from other specications are qualitatively similar.
18. To identify the most appropriate model for our analysis we base the
choice on visual inspection of how the various functional forms t the
data. Given the limited presence of zeros in the dataset, the choice is
restricted between the PPML and the NBREG. A plot of the meanvariance relationship of the PPML and NBREG did not indicate a clear
winner. The Poisson model has a better t for most of the data, while
negative binomial appears to t better the few extreme high values such as
intra-NAFTA trade and China-USA trade. In the absence of a clearly
superior model and to be consistent with the mainstream literature on this
subject, we choose to report the main results of Table 6 based on the
PPML model.
19. For the logged variables the change in trade is simply the estimated
elasticities multiplied by the change in the policy variable. For the index
variables (RPM and LPI) the change in trade is calculated as the
exponential of the dierence in the index multiplied by the coecient,
minus 1.

REFERENCES
Anderson, J., & Marcouiller, D. (2002). Insecurity and the Pattern of
Trade: An Empirical Investigation. Review of Economics and Statistics,
84(2), 342352.
Anderson, J., & Van Wincoop, E. (2003). Gravity with gravitas. A solution
to the border puzzle. American Economic Review, 93, 170192.

Anderson, J., & Van Wincoop, E. (2004). Trade Costs. Journal of

Economic Literature, 42(3), 691751.
Baier, S., & Bergstrand, J. (2004). Economic determinants of
free trade agreements. Journal of International Economics, 64,
2963.

TRADE POLICY, TRADE COSTS, AND DEVELOPING COUNTRY TRADE

Baier, S., & Bergstrand, J. (2009). Bonus vetus OLS: A simple method for
approximating international trade-cost eects using the gravity equation. Journal of International Economics, 77(1), 7785.
Baier, S., Bergstrand J & R. Mariutto. (2010). The Growth of Bilateralism.
CAGE Online Working Paper Series 12, Competitive Advantage in the
Global Economy (CAGE).
Bernard, A. B., Bradford, J. J., Redding, S. J., & Schott, P. K. (2009).
The Margins of US Trade. American Economic Review, 99(2), 487
493.
Burger, M. J., van Oort, F. G., & Linders, G. M. (2009). On the
specication of the gravity model of trade: zeros, excess zeros and zeroinated estimation. Spatial Economic Analysis, 4, 167187.
Deardor, A. V. & Stern R. M. (1998). Measurement of Nontari
Barriers:Studies in International Economics. Ann Arbor MI: University
of Michigan Press.
Djankov, S., Freund, C., & Pham, C. (2010). Trading on Time. Review of
Economics and Statistics, 92(1), 166174.
Francois, J. & Manchin M. (2007). Institutions, Infrastructure, and Trade.
Policy Research Working Paper No. 4152. The World Bank.
Ikenson, D. (2008). While Doha Sleeps: Securing Economic Growth
through Trade Facilitation. Cato Institute Trade Policy Analysis No.
37.
Kee, H. L., Nicita, A., & Olarreaga, M. (2009). Estimating trade
restrictiveness indices. The Economic Journal, 119(534), 172199.

2079

Kee, H. L., Nicita, A., & Olarreaga, M. (2008). Import demand elasticities
and trade distortions. Review of Economics and Statistics, 90(4), 666682.
Lawless, M. (2010). Deconstructing gravity: trade costs and extensive and
intensive margins. Canadian Journal of Economics, 43(4), 11491172.
Limao, N., & Venables, A. J. (2001). Infrastructure, Geographical
Disadvantage, Transport Costs, and Trade. World Bank Economic
Review, 15(3), 451479.
Low, P., Piermartini, R., & Richtering, J. (2008). Multilateral Solutions to
the Erosion of Non-Reciprocal Preferences in NAMA. In B. Hoekman, W. Martin, & C. Primo Braga (Eds.), Trade Preference Erosion.
Palgrave-McMillan.
Nicita, A., & Olarreaga, M. (2007). Trade, Production, and Protection
Database 19762004. World Bank Economic Review, 21(1), 165171.
Silva, S., & Tenreyro, S. (2006). The Log of Gravity. The Review of
Economics and Statistics, 88(4), 641658.
Wilson, J. S., Mann, C., & Otsuki, T. (2003). Trade Facilitation and
Economic Development: A New Approach to Measuring the Impact.
World Bank Economic Review, 17(3), 367389.
World Bank and IMF. (2008). Global Monitoring Report, 2008. Washington DC: World Bank.
Bank, World. (2007). Connecting to Compete. Trade Logistics in the local
Economy. Washington DC: World Bank.
Bank, World. (2008). Doing Business Report 2008. Washington DC: World
Bank.

Available online at www.sciencedirect.com