A partnership with academia

Building knowledge for trade and development

Vi Digital Library - Text Preview

Mapping the Tariff Waters

Working paper by Diakantoni, Antonia, Escaith, Hubert, 2009

Download original document (English)

Tariff water –the difference between bound and applied duties– provides relevant information on domestic trade policy and WTO trade negotiations. This paper examines the general and sectoral tariff structure of 120 economies, using exploratory data analysis.

Staff Working Paper ERSD-2009-13 Date: 1 December 2009

World Trade Organization
Economic Research and Statistics Division

Mapping the Tariff Waters

Antonia Diakantoni and Hubert Escaith


Manuscript date: 1 December 2009

Disclaimer: This is a working paper, and hence it represents research in progress. This
paper represents the opinions of the authors, and is the product of professional research. It is
not meant to represent the position or opinions of the WTO or its Members, nor the official
position of any staff members. Any errors are the fault of the author. Copies of working
papers can be requested from the divisional secretariat by writing to: Economic Research and
Statistics Division, World Trade Organization, Rue de Lausanne 154, CH 1211 Geneva 21,
Switzerland. Please request papers by number and title.

Mapping the Tariff Waters

Antonia Diakantoni and Hubert Escaith1


Tariff water –the difference between bound and applied duties– provides relevant information on
domestic trade policy and WTO trade negotiations. This paper examines the general and sectoral tariff
structure of 120 economies, using exploratory data analysis.

Keywords: Commercial Policy, Taxation, Tariff Duty, GATT-WTO, International Trade
Agreements, MFN, Bound Tariff, Tariff Water.

JEL: F13, H20

The authors acknowledge helpful comments and suggestions by Jean-Michel Pasteels and technical
support from Joaquin Montes and Stefan Froschl. Any remaining errors and analytical shortcomings
remain the sole responsibility of the authors.

1. This article represents the opinions of the authors, and it is not meant to represent the position or

opinions of the WTO or its Members, nor the official position of any staff members.



Governments negotiate at the WTO the bound tariffs levels, i.e. the maximum tariffs they are
authorised to charge on other WTO members. Countries differences in their bound levels as well as
in applied tariffs under the most favoured nation principle (MFN) are most likely related to both their
economic development and their national policies towards domestic protection and trade openness.
To the extend that trade negotiations are reflected in these policies, analysing the existing tariff
structures should expose any underlying negotiation strategies.

The present study examines the hypothesis that the particular structure of bound and applied tariffs of
a given country, reflects a series of trade policy decisions that are related to its economic development
and reveals national authorities' protection standpoint towards the various economic sectors. The
analysis identifies the countries having similar overall structures across sectors, then analyses the
specific patterns for each particular product sector. Finally, the degree of influence of countries'
socio-economic characteristics on their tariff structure is evaluated. In doing so, the analysis takes
into account that recently acceded members went through an accession process that was more
demanding than previous GATT practices.

The relative situation of applied and bound tariffs provides also information on the negotiation
margins of each economy, as open economies (low applied duties) with low bounds have less margin
at the negotiation table than open economies with high bounds. Economies with high applied duties
are in an intermediate situation as any reduction in the bound could cut the MFN applied and this may
be resisted by domestic pressure groups. In addition, some members are also allowed to keep some of
their products unbound while others have fixed bound duties at economic pointless levels i.e., much
above the prohibitive tariff (Foletti, Fugazza, Nicita and Olarreaga 2009).
Thus, a better understanding of the tariff structure can shed additional light on the strategies behind
trade negotiations and the building up of coalitions (Costantini, Crescenzi, De Filippis and Salvatici

Mapping the tariff waters has gained an additional interest after September 2008, when the global
crisis severely affected the international economy. 2 In such a crisis situation, it is feared that
governments will look for uncooperative exit solutions typical of the "Prisoner Dilemma", running
unilaterally their applied tariff within the bound commitments at the expense not only of their
partners, but eventually of themselves and the international governance.

After introducing the data and the statistical methodology, the study proceeds by exploring the tariff
data from a general to a sectoral perspective. The fourth part explores the relationship between the
tariffs structure and relevant socio-economic variables. Finally, the conclusion summarises the main



The analysis covers 120 economies with the European Union 27 countries counting as one and
Switzerland including Liechtenstein. While only WTO Members are considered, some had to be

2 In this paper, the term "tariff water" is used for the binding overhang, namely the difference between

bound and MFN applied tariffs. From a historical context, the expression was used, in a somewhat more
restrictive connection, in the Uruguay Round negotiations when countries offered tariff equivalents that were
obviously inflated (Goode 2003). It should not be confused neither with the notion of "policy space" as many
trade partners have concluded a series of regional or preferential trade agreements (PTA and RTA, respectively)
that legally restrict their margin of manoeuvre.


excluded from the analysis due to lack of information. The applied and bound duties are sourced
from the WTO Integrated Data Base (IDB) and the WTO Consolidated Tariff Schedules database
(CTS). The 2006 applied duties are considered; for 17 countries where this information is missing,
the latest available year is considered. Missing applied duties and ad valorem equivalents are sourced
from the UNCTAD's TRAINS database and ITC's MAcMap database, discarding the outliers (e.g. ad
valorem equivalents higher than 1000%).

In order to obtain cross-country comparability, national tariff line duties are first aggregated up to the
level of 6-digit HS sub-headings by calculating the simple average of included tariff lines. The water
between the bound and applied duty is calculated at the HS 6-digit level; for products where the
applied duty is not available (0.4% of all observations), the water has been left missing. Negative
values of the water (3.5% of all observations) due to reporting errors, binding violations or
inconsistencies in the calculation of ad valorem equivalents, are set to zero. Unbound products 3 (21.3
% of the observations) are considered sensitive for the economy; thus, the imputed water (at 6-digits)
is set equal to the 3rd quartile of the water observed for similar products (same HS 4-digit products, or
same HS 2-digit products when no 4-digit water is available). For products where the water at the HS
2-digit level is not available, the value has been left missing.

To provide economically meaningful information, applied duties and water have been aggregated by
sector by using simple averages, according to UNCTAD's 4 economic classification based on factor
intensity. This nomenclature regroups all HS products into 6 groups at the HS 4-digit level (

Table 1).

Table 1: Product sectors by factor intensity



Typical products

A Non-fuel primary commodities Agricultural products (fish included)

B Resource-intensive manufactures Raw materials (aluminium, paper, leather, silk, etc.)
C Low skill- and technology intensive manufactures Textiles and clothing

D Medium skill- and technology intensive manufactures Organic and inorganic chemicals, rubber, machinery, cars
E High skill- and technology intensive manufactures Pharmaceuticals and hi-tech products

F Mineral fuels Coal, petroleum, other energy.

Source: UNCTAD's economic classification based on factor intensity


In order to identify underlying patterns behind the large volume of data, a series of bottom-up
procedures inspired by Exploratory Data Analysis (EDA) are applied. EDA seeks "connections"
among complex multivariate data sets, without pretence of causality, that may help to formulate
statistical models to test. Because bound, applied and binding overhang are related with a strict
identity, the analysis focuses on applied tariff (the present situation) and the binding overhang (the
tariff water). 5 The category F (Fuels) has been excluded, because it is (i) usually unrepresentative of

3 and not available bound tariffs
4 The original classification follows the HS96 nomenclature; it was also transposed into the HS92 and

in HS02 in order to match all countries applied and bound duties.
5 Dropping one of the variables is not only an editorial choice, but a mathematical necessity. Because

the three components are linearly dependent, they cannot be analysed jointly through most statistical procedures
based on linear algebra.


national trade policy, (ii) particularly prone to measurement errors in notifications and interpretation,
and (iii) plagued by many missing observations.

A series of statistical techniques are implemented in the study. Underlying similarities and
dissimilarities between observations and variables are explored using clustering techniques in order to
(i) develop a typology of objects; (ii) understand the conceptual classificatory function behind the
typology; and (iii) formulate hypothesis about the data generation process. 6 Principal Component
Analysis (PCA) is used to identify the underlying multidimensional data structure. PCA displays the
data in a substantially reduced subspace of uncorrelated variables, that best preserves the statistical
information (variance) contained in the original data. 7 Discriminatory analysis is applied in the end,
with a view to verify whether the composition of the identified clusters during the analytical process
could be linked to any trade policy or economic performance variables.


Adopting the time honoured "general to specific" statistical analysis, this section initiates with a
global perspective, factoring-in all the tariff dimensions (sectors) into a common pattern,
representative –at least in the tariff space– of each country's trade policy. The second part is
successively focusing on each one of the tariff policy dimensions (sectoral tariffs) to scale more
precisely the tariff profile of each individual member.


The entire set of variables and observations is explored first, in order to provide some general
descriptive information of the dataset and to identify any general underlying patterns across both
products and countries.

1. Descriptive statistics on tariffs

Table 2 presents a set of descriptive statistics concerning the observed distribution of applied tariffs
and the binding overhang. Not surprisingly, the A category of agricultural products emerges almost
immediately as an outlier: it spreads across a large interval and has the highest variance in both
dimensions of applied and water.

A box plot of agricultural data 8 reveals the presence of outliers for applied tariff duties, with the
maximum lying far away from the normally expected range. Aside from agriculture, the group D, of
medium skill- and technology intensive manufactures (chemicals and cars), shows also an interesting
distribution. At the contrary of A, it is a very compact group, distributed much sharply and more
concentrated than a normal Gauss distribution. The distribution of applied tariffs in group C of low
skills- and technology intensive manufactures (textiles and clothing) is the closest to normality, in our
sample. It is relatively symmetric, albeit flatter than the normal distribution. Groups B and E come
in-between the previous cases.

6 Testing those hypothesis belongs to "confirmatory data analysis", not covered here.
7 PCA is similar to factor analysis, but it is more general (it works on the total variance of the

observation and does not imply any underlying statistical model).
8 Not published here; all the same, distribution patterns for individual variables can easily be inferred

from Table 2 and Figure 1.


Table 2: Descriptive statistics on applied MFN and binding overhang
Variables: a

Statistics: App_A App_B App_C App_D App_E Dif_A Dif_B Dif_C Dif_D Dif_E

No. of observations 120 120 120 120 120 120 120 120 120 120

No. of missing values 0 0 0 0 0 0 0 0 0 0

Minimum 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

Maximum 54.2 30.0 29.3 30.5 31.2 148.2 115.7 115.0 119.5 116.8

Freq. of minimum 2.0 4.0 3.0 3.0 5.0 2.0 3.0 3.0 3.0 3.0

Freq. of maximum 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

Range 54.2 30.0 29.3 30.5 31.2 148.2 115.7 115.0 119.5 116.8

1st Quartile 8.7 4.9 6.8 3.6 3.6 15.3 7.5 6.6 7.0 4.6

Median 12.9 8.3 11.6 6.5 8.1 26.5 22.6 17.3 21.2 17.4

3rd Quartile 17.1 12.6 16.3 8.5 11.1 62.0 40.5 36.9 41.9 36.4

Mean 14.3 9.1 12.4 6.9 7.8 38.8 28.5 25.9 26.9 23.5

Variance (n) 83.8 33.4 43.4 20.6 29.2 1077.1 724.4 666.0 617.9 528.3

Standard deviation (n) 9.2 5.8 6.6 4.5 5.4 32.8 26.9 25.8 24.9 23.0

Variation coefficient 0.6 0.6 0.5 0.7 0.7 0.8 0.9 1.0 0.9 1.0

Notes: a/ Variable codes used along the paper are constructed using a prefix plus a suffix. Prefix "App_" stands for applied
MFN, "Dif_" for difference between applied and bound. Suffix refer to product groups: A for Non-fuel primary
commodities; B for Resource-intensive manufactures; C for Low skill- and technology intensive manufactures; D for
Medium skill- and technology intensive manufactures; E for High skill and technology intensive manufactures (see Table 1).
All statistics are based on un-weighted average.
Source: Authors' calculation on the basis of WTO tariff data.

Applied tariffs are usually more regularly distributed than the binding overhang. The distribution of
water is characterised by its concentration towards the lower end of the spectrum, together with a
large variance of the observations. Here again, agriculture is an extreme case; all remaining groups
—the non agricultural products, or NAMA products— share apparently more or less the same

The scatter-plot matrix (Figure 1) allows to visualize the individual and the joint distribution of
applied tariff levels and tariff water, offering a fairly comprehensive view of the data. The diagonal
shows the distribution histogram of each individual variable and complements the statistics presented
in Table 2. Scatter-plots outside the diagonal illustrate, for each cell or "panel" formed by the
intersection of a row and column, the bivariate relationship between the variables appearing on the
horizontal and vertical axis. To facilitate the visualization of the bivariate relationship, a confidence
interval ellipse is centred on the sample means of the respective variables. It indicates the sample
covariance between the two variables (a measure of correlation) and its orientation. Highly correlated
variables will exhibit an elongated ellipse along the first or second diagonal of the graph (positive or
negative correlation).


Figure 1: Scatter-plot matrix of applied tariffs and difference

Source: Authors' calculation on the basis of WTO tariff data.

The histograms on the diagonal of the scatter-plot matrix show a clear distinction between the
distribution of applied tariffs and water. The tariff waters share a common profile with a concentration
of observations on low values and with a regular decrease of frequency when moving to higher values
for the water. This patter is steeper for higher technology contents (categories D and E) and more
diffuse and irregular for agriculture. Applied MFN show a wider difference in histograms, from log-
normal (groups A, B and D) or asymmetric (E) to a more normal distribution around the average (C).

There is usually a strong bilateral interrelation between tariff waters, as shown by the elongated
feature of the covariance ellipse. The relationship is strong for NAMA but, once again, water in
agriculture is less closely related to water in other categories of goods (i.e., some countries may have
high water for agriculture and low water for NAMA, or conversely). The relationship among applied
tariffs is also present, albeit weaker, between B and the other NAMA categories C, D and E, and
between D and E. The covariance between {C ; D}, and {C ; E} exists, but is weaker than for the
other NAMA categories. As far as agriculture is concerned, no clear bivariate relationship appears
from the graphs, as evidenced by the rounder ellipses. Similarly, there is no direct relationship
between applied and differential tariffs.


The degree of similarity/dissimilarity among the product groups has been also be more formally
measured by testing the following hypothesis: do the 5 product sectors belong to the same statistical
population as far as the applied duties and water are concerned? For the applied duties the hypothesis
was rejected, confirming that each group receives a specific tariff treatment and validating that the
five sectors can be considered as separate entities. Even when the comparison is limited to NAMA
groups (B, C, D and E), the hypothesis is rejected.

The situation for the water is somehow similar as the hypothesis has been also rejected, but the risk of
wrongly rejecting the hypothesis is about 0.2, while for the applied tariffs was almost nil. Moreover,
when the sample is reduced to the three NAMA groups B, C, E, the hypothesis of same origin is
accepted: water is sufficiently similar across these three sub-group to consider them as homogeneous
for this criteria.

Thus, from an analytical perspective, applied tariffs and binding overhang seem to follow a different
logic. Agriculture follows a clearly specific internal logic (sui generis) in terms of applied duties and
water. Moreover, the MFN applied levels in the NAMA sector are clearly distinct but the water looks
rather homogeneous across NAMA products.

2. Geographical mapping of tariffs

By categorizing tariffs (0 to 5%; between 5 and 15%; higher than 15%) and by colouring high values
with dark colours, an illustration of the average applied tariffs in agriculture and in NAMA for the
120 economies is provided in the world maps 1 and 2 respectively. 9 The word maps 3 and 4 provide
an overview of the average tariff water for agricultural and NAMA products by using the same scale
and colours.

Countries seem to apply in average higher duties in agriculture than in NAMA, as shown in maps 1
and 2. For instance, in the Americas Mexico and Colombia look much darker in the agricultural
sector, like Egypt in Africa and Norway, Switzerland and Iceland in Europe. Similarly Japan in Asia,
Turkey and Saudi Arabia in Middle East apply higher duties in agriculture and appear darker in map
1. Nevertheless, a few countries applying lower duties in agriculture can be identified like Rwanda
and Burundi in Africa, as they appear lighter in the agriculture map.

Globally, in agriculture the average applied duty is 14% and the water 39%; in NAMA the
corresponding values are 9% and 26% and as expected, are lower than in agriculture. It has to be
reminded that averages include AVEs which can fluctuate significantly from one year to another and
the average NAMA water includes a high number of estimates for the unbound products. 10

9 NAMA products values are calculated by summing up sectors B, C, D and E.
10 The NAMA groups, B, C, D and E include equivalent numbers of unbound products.


Map 1: Applied duties for agricultural products

Note: Economies with applied duties in average from 0 to 5% appear in white, between 5 and 15% in grey and higher than
15% in black. Crossed diagonals indicate that data are not available for the country.
Source: Authors' calculation on the basis of WTO tariff data.

Map 2: Applied duties for NAMA products

Note: Economies with applied duties in average from 0 to 5% appear in white, between 5 and 15% in grey and higher than
15% in black. Crossed diagonals indicate that data are not available for the country.
Source: Authors' calculation on the basis of WTO tariff data.

The visual exploration of the water in agriculture, see map 3, indicates that Latin America economies
are characterised by deep water.


Map 3: Tariff water for agricultural products

Note: Economies with water in average from 0 to 5% appear in white, between 5 and 15% in grey and higher than 15% in
black. Crossed diagonals indicate that data are not available for the country.
Source: Authors' calculation on the basis of WTO tariff data.

Map 4: Tariff water for NAMA products

Note: Economies with water in average from 0 to 5% appear in white, between 5 and 15% in grey and higher than 15% in
black. Crossed diagonals indicate that data are not available for the country.
Source: Authors' calculation on the basis of WTO tariff data.

In most African and in some Asian countries, the margins in agriculture look also very high.
Developed countries show rather low margins regardless their geographical location.

For several countries, the tariff water in NAMA (map 4) is lower than in agriculture; it is not the case
for Australia, New Zealand and some African countries like Central African Republic, Guinea Bissau,
Sierra Leone, Angola, etc. Apart from Latin American countries, there is no evidence that the water
in developing countries is related to the geographical region. The visual exploration shows also the


diversity of situations regarding tariff schedules across countries and between agriculture and NAMA

3. Tariffs by stages of processing

Tariff policy is defined not only by the absolute level of market protection provided by applied and
bound tariff, by also by the "effective protection" or "effective rate of assistance" provided by tariff
escalation. Tariff escalation results in progressively higher import duties on semi-processed products
than on raw materials, upwards to finished products. As mentioned by the Dictionary of Trade Policy
Terms (Goode, 2003), this practice protects domestic processing industries and discourages the
development of processing activity in the countries where the raw materials originate.

The extend of tariff escalation is explored by classifying products according to their stages of
processing, from raw (stage 1) to partially processed (stage 2) and processed (stage 3) and computing
the corresponding indicators of applied MFN, bound tariffs and water. The results presented in , show
a complex pattern of escalation within product groups A, B, C and D in terms of applied duties, and
within product groups A, C and D in terms of bound duties. A de-escalation pattern is noticeable
across product groups for both applied and bound tariffs as duties decrease with the complexity and
intensity of factoring (or, conversely, increase with the intensity of low skilled labour). 11 This pattern
is more consistent with the protection of domestic labour than the promotion of industrialization
through effective assistance. It should also be noted than those patterns observed for the sample of
120 members hide large variances across countries, and the coefficient of variation is always
comprised between 0.7 and 1.

As shown in the first panel of Figure 2 within each of the NAMA product groups, applied tariffs tend
to be higher for the processed products (stages 2 and 3) belonging to group C (low skill and low
technology manufactures). These labour intensive products, such as textile and clothing, receive a
more significant protection than other manufactures in terms of tariff escalation. The tariff escalation
pattern for bound averages is slightly different than for applied duties. Within each product group,
countries tend to protect more the processed products (process3) but unprocessed goods tend to be
more protected than semi-processed. Group B –raw materials- makes the exception, as the bound
protection is higher for un-processed products (stage 1).

It should be also noted that the higher protection observed in panel 1 (applied tariffs) for labour
intensive manufactures results more from an evolution of practices than an a priori policy, as the
escalation graph for the bound tariffs (second panel) does not show significant differences between
groups B, C and D.

11 Comparing product groups, the relatively more technical intensive groups D and E have lower

applied and bound tariffs than the other product groups; similarly, agriculture is always more protected than
manufactured products. It should be noted that, when performing "within group" analysis, tariff escalation
within the products "E" (high skill- and technology intensive manufactures) is not meaningful under the present
classification in three stages, as this category includes practically no unprocessed articles.


Figure 2: Tariffs By Stage of Processing and Product Group, 2006




Process1 Process2 Process3



Bound Tariffs









Difference Bound-Applied

Applied MFN

Notes: Three stages of processing, from low (Process 1) to high (Process 3). Product groups: A: Non-fuel primary
commodities; B: Resource-intensive manufactures; C: Low skill and technology intensive manufactures; D: Medium skill
and technology intensive manufactures; E: High skill and technology intensive manufactures.
Source: Authors' calculation on the basis of WTO tariff data.


4. Looking for tariff policy patterns

After this first examination of the data, the present section is using EDA techniques to identify
potential underlying structures in the data. The research is conducted in two directions: the variables
(tariff duties) and the observations (countries). Structuring the variable space means looking for
common tariff patterns across countries, while studying the observation space provides information on
economies, based on their use of applied tariffs and binding overhang.

(a) Cluster analysis of tariff variables

The Agglomerative Hierarchical Clustering (AHC) of tariff variables is straightforward, as showed in
Figure 3, and confirms the initial results of previous sections; the dendrogram should be read
bottom-up by considering that the most similar objects are paired first. The tariff water presented at
the left panel of the graph, clusters more rapidly than average tariffs, indicating more homogeneity
than the applied MFN tariffs.

Figure 3: Hierarchical clustering of tariff variables














Notes: a/ Product groups: A, Non-fuel primary commodities; B, Resource-intensive manufactures; C, Low skill- and
technology intensive manufactures; D, Medium skill- and technology intensive manufactures; E, High skill- and technology
intensive manufactures
Source: Authors' calculation on the basis of WTO tariff data.

For both applied and water, agriculture stands out by its dissimilarity to other product groups. Its
specificity measured by the dissimilarity index on the graph, is larger in the "applied tariff" dimension
than in the binding overhang. Variables related to groups B, C and D are close in the water
dimension, but less in the applied tariffs. Group C of low skill manufactures (textiles and clothing)
differs from the other NAMA products, in the applied tariffs dimension.


(b) Cluster analysis of the economies

The analysis of the 120 "observations", included in our sample by visual examining a series of AHC
experiments, indicated that countries could be "appropriately" grouped into 5 clusters (Figure 4). 12

Figure 4: Hierarchical clustering of observations













Source: Authors' calculation on the basis of WTO tariff data.

The composition of the 5 clusters is defined by using the K-means clustering.13 The classification as
showed in Table 3, produces two large clusters, 1 and 2, including respectively 41 and 34 countries,
and three medium sized clusters of about 15 economies. Clusters 1 and 4 are the most "compact",
with the lowest within-class variance. Cluster 5 is the "loosest" one, with large within-class variance
and high average distance to centroid. In other words, the typical tariff profile of classes 1 and 4 is
shared by most of its members, while the association of class 5 participants with a "typical pattern" is

All clusters demonstrate relatively higher applied tariffs for group A (agriculture) and C (low
skill/capital manufactures), although there is more dispersion concerning the treatment of agriculture.
Product groups D and E (medium and high skill manufactures) tend to be generally less protected.

12 Appropriate means that this number provided for reasonably balanced sub-samples, while

maintaining significant distances between each cluster. As often in EDA, it was the result of a subjective
judgement based on the interpretation of graphs. The following sections will show that this decision can also be
based on quantitative information criteria.

13 The procedure divides the dataset by minimising the within-group sum of squared errors in terms of
the Euclidean distance from the group centroid. – i.e. the mean of variables included in the cluster. Because K-
means is sensitive to initial partitioning, 10 simulations were run using different random initial partitions, and
the best case was selected on the basis of minimum within-class variance, i.e., the partition giving the most
compact clustering.


Table 3: Classification of economies in five clusters.


















1 41

Albania, Armenia, Australia, Botswana, Côte d'Ivoire, Cambia, Canada, China, Congo, Croatia, Cuba,
Ecuador, EC, FYROM Macedonia, Gabon, Georgia, Guinea, Hong-Kong, Japan, Jordan, Korea Rep.,
Kyrgyz, Macao, Madagascar, Mauritania, Moldova, Mongolia, Namibia, Nepal, New Zealand, Oman,
Qatar, Saudi Arabia, Sing, South Africa, Swaziland, Switzerland, Taipei Chinese, UAE, USA, Viet
Nam. 325

2 34

Argentina, Bahrain, Venezuela, Bolivia, Brazil, Brunei, C. African Rep., Chile, Costa Rica, Djibouti,
Dominican Rep., Egypt, El Salvador, Fiji, Guatemala, Guinea Bissau, Haiti, Honduras, Indonesia,
Maldives, Mexico, Morocco, Nicaragua, Panama, Papua NG, Paraguay, Peru, Philippines, Senegal,
Sierra Leone, Sri Lanka, Thailand, Turkey, Urugay. 564

3 14
Benin Burkina Faso, Burundi, Colombia, Iceland, India, Israel, Malaysia, Mali, Myanmar, Niger,
Norway, Tunisia, Zimbabwe. 1079

4 14
Angola, Antigua & Barbuda, Belize, Dominica, Ghana, Grenada, Guyana, Jamaica, Malawi, Pakistan,
St. Lucia, St. Vincent, Trinidad & Tobago, Uganda. 456

5 17
Bangladesh, Barbados, Cameroon, Chad, Dem. Rep. Congo, Kenya, Kuwait, Lesotho, Mauritius,
Mozambique, Nigeria, Rwanda, St, Kitts & Nevis, Solomon Isl., Tanzania, Togo, Zambia. 2434

Note: Classification based on applied tariffs and binding overhang in the five product groups A to E (see Table 1). The
maximum distance to centroid indicates possible outliers.
Source: Authors' calculation on the basis of WTO tariff data.

When contrasted with other groups, cluster 1 is characterized by comparatively low values for applied
tariffs and water for all product groups (Table 4). At the other extreme, cluster 5 is made of countries
having both high applied tariffs in average (except in agriculture), and high binding overhangs. Class
4 is somewhat similar to class 5 albeit much more homogeneous, as seen above, but with lower values
in both tariffs and waters.

Table 4: Tariff profile of the five clusters

Statistics:a Average Std. Deviation Coefficient of variation

Variables: Applied Water Applied Water Applied Water

1 7.7 5.7 2.7 2.4 0.4 0.4

2 10.5 24.4 3.1 1.2 0.3 0.0

3 12.6 25.4 5.4 19.8 0.4 0.8

4 10.8 51.2 3.3 11.5 0.3 0.2

5 12.5 76.9 2.9 10.1 0.2 0.1

Notes: a/ computed on the values of each class centroid for the five categories of products (A to E).
Source: Authors' calculation on the basis of WTO tariff data.

Compared with 4 and 5, clusters 2 and 3 shows similar high applied rates in all product groups, (and
somewhat higher for agriculture) but lower water (except for agriculture). The difference between
clusters 2 and 3 themselves is more subtle, and relates mostly to the dispersion of the tariff water
rather than the applied level. Cluster 2 is characterized by more homogenous pattern of applied tariffs
and water across the five product groups, while the third cluster shows higher variation, especially in
tariff water.

The last step of the EDA on the general tariff profiles applies a principal component analysis to the
variables and observations. Most of the variance (82%) is explained by the first two components
while the third explains only 8% of the variance. The first factor (53% of total variance) organises the


observations by the level of binding overhang (see Table 5) and the second by the applied tariffs for
non agricultural goods. Finally, the third dimension is determined by the level of applied tariffs in
agriculture. Figure 5 provides a projection of the 120 observations on a two dimensional graph,
defined by the first two components as axis, and with the third component indicated by the size of the

Table 5: PCA factor loading

D1 D2 D3

A 0.0 0.3 1.0
B 0.2 0.9 0.2
C 0.1 0.9 0.2
D 0.1 1.0 0.1
E 0.2 0.9 0.1

A 0.9 0.1 0.1
B 1.0 0.1 0.0
C 1.0 0.1 0.0
D 1.0 0.2 0.0

E 0.9 0.2 0.0

Note: Factor loading after applying varimax rotation.
Source: Authors' calculation on the basis of WTO tariff data.

Figure 5: Principal Component Analysis of clusters





































43 4












1 5





1 1






1 1





1 11




















Mauritius, Lesotho,






Notes: After varimax rotation. The numbers refer to the clusters, the size of the bubble indicates the score on D3.
Source: Authors' calculation on the basis of WTO' tariff data.

The ordering from left to right of the clusters according to their score in terms of binding overhang
appears clearly on the graph. At the contrary, the second dimension of NAMA applied tariffs, does
not discriminate between clusters: for each cluster, there are observations scattered across a large
range of applied tariffs. The exception is cluster 4, which shows a clear concentration around the
average value. It appears also that clusters 2 and 3 are overlapping and could be merged without


loosing too much information, at least according to the two principal components. Observation
scoring relatively high on D3 (high applied in agriculture) tend to agglomerate on low value of D1.

From the projection, Djibouti appears as a clear outlier in the sample; at the contrary, Tanzania looks
more as a extreme value for group 5, rather than a real outlier.14 Mauritius, Lesotho and Kuwait are
exocentric with respect to group 5, which visibly has the highest within-group variance (Table 3).
Finally, Norway stands out for its low tariffs in non-agricultural products (second axis) and its high
level of agricultural protection (third factor, shown by the size of the point in the graph).


Moving from general to specific, in this section the applied tariffs and water are explored by product
sector. The research looks in every product sector for similarities and dissimilarities between the
economies and investigates how the observations can be best regrouped (or split).

At first, the most evocative results obtained with clustering techniques are presented. Then, the
negotiation margins are illustrated by following a simpler and more intuitive approach based on the
relative position of each economy in relation to applied duties and water. The results for only two
products are presented in this paper:

• group A -agricultural products-, for its specific role in partitioning the sample, and
• group C -textiles and clothing-, a symbolic NAMA product.

1. K-means cluster analysis for groups A and C

The k-means cluster analysis method is used again to partition the observations. The optimal number
of clusters is defined on the basis of the Calinski - Harabasz pseudo-F index 15. For group A and C
the optimal number of clusters obtained is two and seven (see also footnote 13). 16

In figure 6 are projected the 120 economies' average values of applied duties (horizontal axis) and
binding overhang (vertical axis) in the agricultural sector. The cluster analysis regroups low overhang
economies in cluster 1 and high overhang economies in cluster 2. The applied tariffs factor does not
discriminate between clusters. It's worth mentioning that developed economies, with the exception of
Iceland and Norway, are all included in cluster 1 which counts 80 economies.

Extreme values lay at the top-left and bottom-right of the plot opposing Lesotho, Bangladesh, Nigeria
and Mauritius to Tunisia, Egypt, Norway and Morocco. Djibouti and Tanzania depicted as extreme
values/ outliers in the previous section are now laying near to the centres of clusters 1 and 2 and have
rather a 'normal' tariff profile in this sector. Consequently, their agricultural tariff policy had to
significant influence in their position in the global analysis.

14 Reprocessing the data without Djibouti did not change substantially the results.
15 The larger the index, the better the partition provided by the corresponding number of clusters. In

the previous chapter, a more intuitive visual examination based on an initial hierarchical clustering was used to
define this optimal number. The present approach is more objective, and easier to apply when many sub-groups
have to be treated successively.

16 For the remaining product groups, the obtained optimal number of clusters was 5 for group B, 4 for
group D , 8 for both E and F.


Figure 6: K-means cluster analysis for agricultural products, product group A

Source: Authors' calculation on the basis of WTO's tariff data.

The scatter diagram for the textiles and clothing in figure 7, gives a different picture of the 120
economies. Many data points have moved to the centre of the diagram compared to agriculture. The
k-means cluster analysis regroups the data points into 7 clusters.

The clusters are well balanced and include equivalent number of countries. The smaller cluster is
cluster 7, which includes only 7 economies and has the highest dispersion. Again, the binding
overhang is the factor that basically discriminate the economies and forms the clusters.

Tanzania is an extreme value in this group of products, and has the highest overhang. It is included in
cluster number 7 together with Kuwait and Mauritius. Djibouti, revealed as outlier in the global
analysis shown in the previous section, lays at the bottom-right of the scatter together with Zimbabwe,
Morocco and Viet Nam. Lesotho, almost an outlier for agriculture, has a 'normal' profile for textiles
and clothing, and belongs to cluster 3. Norway has also a completely different behaviour in this
product group, and is now part of cluster 1.


Figure 7: K-means cluster analysis for textiles and clothing, product group C

Source: Authors' calculation on the basis of WTO tariff data.

It is now clear that economies behave differently in their tariff policy according to the product sectors.
In agriculture, economies with similar tariff policies form basically two groups; whereas in textiles
and clothing sector, the situation is more complex: a larger variety of tariff policies is observed,
leading to establish seven "policy" clusters.

However, cluster analysis is a technique best used for dealing with multiple criteria; it is probably too
sophisticated when dealing with only two dimensions, i.e. applied and water for a single product
group. It is also sensitive to the presence of outliers. The following section offers a more intuitive
and robust approach.

2. Illustrating the negotiation margins: groupings based on percentiles

In this section a different way to regroup countries is proposed: straightforward categories based on
applied duties and tariff water percentiles are used to partition the data points. The 33rd and 66th
percentiles 17 are used to regroup economies according to three categories (low / middle/ high) of
applied duties and three categories (low / middle/ high) of binding overhang. These 3x3 categories
split the scatter diagram of the 120 economies in 9 distinct groups (table 6) for every product group.

17 Defined as the values below which stand 33% and 66% of all observations, respectively.


Table 6. Visual presentation of the percentile approach

Group 1 includes very open economies applying
relatively low duties (below the 33rd percentile of
applied duties of the sample) and having very shallow
water (below the 33rd percentile of the binding
overhang of the sample). Conversely, group 9
gathers very protectionist economies with very high
applied duties (above the 66th percentile of applied
duties of the sample) and very high margins (above
the 66th of the binding overhang of the sample).
Compared to cluster analysis, where groupings are
based on abstract topological considerations, the
percentile approach provides a more intuitive
interpretation of the groupings and allows a cross-
products / cross-countries comparison.

Table 7 shows the 9 percentile classes for agriculture. Classes 1 and 2 include the majority of
developed countries and confirm, for example, the openness of the USA, Australia and New Zealand
markets in the agricultural sector. EC and Canada are included in the second group as they apply in
average, higher duties in agriculture than Class1 countries. It's interesting to observe that among
developed countries the most protective in agriculture are Japan and Switzerland (class 3).
Table 7: The percentiles approach in product group A (Agriculture) - the composition of classes

Members of the class Number


Albania, Armenia, Australia, Croatia, Cuba, Haiti, Hong-Kong China, Kyrgyz Rep., Macao China, Moldova,
Mongolia, New Zealand, Oman, Qatar, Saudi Arabia, United Arab Emirates, United States



Côte d'Ivoire, Canada, China, Ecuador, European Communities, FYROM Macedonia, Georgia, Madagascar,
Panama, Peru, Senegal, Chinese Taipei


Cambodia, Central African Republic, Congo, Japan, Jordan, Korea Rep., Morocco, Switzerland, Thailand,
Turkey, Viet Nam 11


Argentina, Bahrain, Botswana, Brazil, Brunei Darussalam, Chile, Guatemala, Honduras, Indonesia, Namibia,
Nicaragua, Paraguay, Philippine, Singapore, South Africa, Swaziland, Uruguay



Angola, Benin, Venezuela, Bolivia, Costa Rica, Dominican Rep., El Salvador, Guinea, Guinea Bissau, Mali,
Mauritania, Nepal, Sierra Leone


CLASS 6 Djibouti, Egypt, Fiji, Gabon, Israel, Maldives, Mexico, Papua New Guinea, Sri Lanka, Tunisia 10

CLASS 7 Burundi, Kuwait, Lesotho, Malaysia, Mauritius, Myanmar 6


Antigua and Barbuda, Burkina Faso, Colombia, Dem. Rep. of Congo, Malawi, Mozambique, Niger, Nigeria,
Pakistan, Rwanda, St. Kitts and Nevis, St. Lucia, St. Vincent and Grenadines, Togo, Trinidad and Tobago



Bangladesh, Barbados, Belize, Cameroon, Chad, Dominica, Ghana, Grenada, Guyana, Iceland, India, Jamaica,
Kenya, Norway, Solomon Isl., Tanzania, Uganda, Zambia, Zimbabwe

Source: Authors' calculation on the basis of WTO tariff data.

It is also interesting to notice that Madagascar and Senegal are part of class 2, with very low water.
The majority of developing countries are expected to be included in upper groups, where margins are
higher. What is most probably unexpected is to find in the class 9 of very protective economies with
very high margins Iceland, India and Norway. In the textiles and clothing sector, the majority of
developed countries are, as expected, in class 1, except Australia and New Zealand which are in class
4. Class 1 is the most numerous and also includes many developing countries. 18

18 Albeit the uni-dimensional clustering is done using percentiles and is by definition balanced, the
intersection of two uni-dimensional clusters needs not to be balanced (indeed, the intersection can be empty).


Table 8: The percentiles approach in product group C (Textiles and Clothing)- the composition of classes
Members of the class number


Armenia, Canada, Croatia, European Communities, Georgia, Hong-Kong China, Iceland, Japan, Korea Rep.,
Kyrgyz Rep., Macao China, Moldova, Norway, Oman, Saudi Arabia, Singapore, Switzerland, Chinese Taipei,
United States 19

CLASS2 Albania, Cambodia, China, Cuba, FYROM Macedonia, Guinea, Jordan, Malaysia, Mauritania 9


Benin, Botswana, Burkina Faso, Burundi, Côte d'Ivoire, Congo, Gabon, Mali, Namibia, South Africa, Swaziland,
Viet Nam



Australia, Brunei Darussalam, Chile, Haiti, Israel, Mongolia, Myanmar, New Zealand, Panama, Papua New
Guinea, Qatar, Turkey, United Arab Emirates


Dominican Rep., Ecuador, Honduras, Indonesia, Madagascar, Paraguay, Peru, Philippine, Sri Lanka, Thailand,
Uruguay 11


Argentina, Venezuela, Brazil, Central African Republic, Colombia, Djibouti, Egypt, India, Maldives, Mexico,
Morocco, Nepal, Niger, Pakistan, Senegal, Tunisia

CLASS7 Angola, Bahrain, Costa Rica, Jamaica, Kuwait, Mauritius, Nicaragua, Trinidad and Tobago 8

Antigua and Barbuda, Barbados, Belize, Bolivia, Dem. Rep. of Congo, Dominica, El Salvador, Fiji, Ghana,
Grenada, Guatemala, Guyana, St. Kitts and Nevis, St. Lucia, St. Vincent and Grenadines, Solomon Isl.. 16

Bangladesh, Cameroon, Chad, Guinea Bissau, Kenya, Lesotho, Malawi, Mozambique, Nigeria, Rwanda, Sierra
Leone, Tanzania, Togo, Uganda, Zambia, Zimbabwe 16

Source: Authors' calculation on the basis of WTO tariff data.

It's worth mention that some Least Developed Countries (LDC) appear in classes 2 and 3 - like Benin,
Burkina Faso, Burundi, Cambodia, Guinea, Mali and Mauritania-, an unexpected result considering
that these classes have very shallow water. This tends to indicate that tariff policies are not always
directly related to the development factor, an hypothesis that will be further tested in the next section
of the paper.

3. Negotiation margins: A cross-country and cross-product comparison

The percentiles approach divides in every sector the data points into 9 groups and consequently allows
a cross-product / cross-countries comparison. Table 9 shows the results for selected countries (all
developed economies and selected CIS, developing countries and LDCs). For each country and group
of product the applied duties (high, middle and low) are illustrated with colours (black, grey and
white) and the water (large, middle and low) with waves (≈,∼, ).
Developed countries are expected to naturally have low values in the water and therefore to be
included in classes 1, 2 and 3. Conversely, LDCs are expected to have high values in the water and to
a certain extent be included in classes 7, 8 and 9.

Surprisingly some of the developed countries have high applied duties (coloured black or grey) for
some products –agriculture, textiles and clothing, pharmaceuticals- and have even high water (two
waves). CIS countries show a 'developed economy' profile and appear as open as developed countries
without benefiting from water in their tariffs.

In the other extreme, LDCs at the bottom-left of table 9 show a variety of profiles but globally apply
high duties (with some exceptions) and do not always benefit from high margins in the binding

Developing countries, in the right part of

Table 9, have very different types of profiles, from 'developed country' profiles —Albania, Croatia,
Hong-Kong China, Macao China, Singapore etc.— to 'LDC country' profiles —Antigua and
Barbuda, Bolivia, Dominica, Nigeria, St. Kitts and Nevis, St. Lucia, St. Vincent, Barbados, Belize,
Grenada, Guyana, Kenya, Pakistan, Cameroon, Ghana and Zimbabwe.


Table 9: Negotiation margins for selected countries by product group

Note: The level of applied duties, high, middle and low, are illustrated with black, grey and white and the level of water,
large, middle and low, with waves (≈,∼, ).
Source: Authors' calculation on the basis of WTO tariff data.



This inventory of selected economies so called "negotiation margins" presented in the previous
section tends to indicate that the variety of tariff policies followed by the countries are rather product-
specific, and only loosely related to the economic development dimension. The present sections
explore more thoroughly the relationship between trade policy and economic factors. Two
complementary analyses are undertaken: the relationship between socio-economic variables and
tariffs is explored initially at the product group level and then at the global level. Lastly, the socio-
economic variables are used to determine the tariff policy and therefore, endorse any obtained
analytical conclusions.


A series of structural socio-economic variables and macroeconomic indicators are associated to the
initial tariff indicators including the size of the economy in terms of population and GDP, the average
per capita income, and the source of value added (agriculture, industry, services). Trade indicators
cover openness, the relative share of trade in goods and services, and the main categories of imports
and exports of merchandises (food, fuels and manufactures). Macroeconomic variables include
inflation, current account balance, government expenditure and taxes. Other qualitative variables
influencing the trade negotiations, such as the date of WTO membership, the level of development
and the geographical location are also added. In total, 75 socio-economic variables are included.
Most of the quantitative variables are grouped in a triplet including the 1995 and the 2006 levels, the
variation from 1995 to 2000, and from 2000 to 2006. The socio-economic data are sourced from the
World Bank 's World Development Indicators.

1. Exploring the socio-economic variables

Many of the socio-economic variables exhibit strong co-linearity, either because the indicators are
partially redundant or because they are linked through a functional relationship.. Principal component
analysis is applied to the initial dataset to reduce multi co-linearity and isolate, out of the 74 variables,
the relevant variables to be retained. In order to avoid endogeneity issues, variables in level are
measured in 1995, at the conclusion of the UR that fixed the bound tariffs, and values for 2006 are
reflected through two sets of rates of growth (from 1995 to 2000, and from 2000 to 2006). A second
analysis is performed with the 2006 level, in order to check the robustness of results (results did not
differ substantially, except for the ordering of the components).

The first result is disappointing as no clear pattern can be established in the extended dataset. The
main component, supposed to capture most of the information, explains less than 10% of the total
variance. Any conclusion or interpretation of the results would have a very weak explanatory power:
the four major principal components accounted for only 30% of the variance, and one would have to
include up to 22 dimensions to explain at least 80% of the total variance.
With these caveats in mind, the socio-economic data set is structured by the following factors: (1)
overall economic development dimension, (2) small oil and service oriented economies; (3) trade
openness and macroeconomic variables.

• The first component F1 explains 10% of total variance. It is associated with rural countries on
its positive segment, with a high share of agriculture in GDP, and low manufacturing and
services contribution. This rural orientation remains extensive, with a low use of fertilizers:
those are predominantly poor countries, as measured by the per capita income, and have a
high population growth and a high share of food imports and fuel imports; They are
characterized by a high weight of inelastic imports, corresponding to a low share of
manufactures in the imports. The governmental consumption has also a low incidence in
GDP. As for their external sector, the countries with high scores for this dimension were
highly indebted countries in 1995, but experienced a rapid decrease of their external
indebtedness between 2000 and 2006. Because their large current account deficit has not


improved markedly during the period, the reduction in external debt may be due to the impact
of official assistance, such as the initiatives adopted in the framework of the Millennium
Development Goals. Their exports are mainly agricultural products and low in manufactures.
This dimension is closely associated with high levels of applied MFN (except in agriculture)
and deep water (Table 10). Sub-Saharan Africa, as a region, is closely associated with
positive values of F1.

• Countries with a high score for the second component F2 (capturing 8% of total variance) are

associated with high inflation levels over the entire period, albeit decreasing between 2000
and 2006. They also tend to have high military spending in 1995, decreasing between 1995
and 2000. The share of industries predominantly extractive, representing in 1995 a large
proportion of GDP, has further increased between 1995 and 2000. Correspondingly, the share
of services in the economy has decreased markedly since 1995, and particularly between 1995
and 2000. Albeit a low share of agriculture in GDP in 1995, it has increased during the
reporting period. This latter characteristic is not found in other components. Their external
sector dominated by exports of fuels, has strengthened their current account position between
1995 and 2000. Despite registering no particular current account deficit in the base year, these
countries had accumulated an important stock of external debts. Indebtedness decreased
quickly between 1995-2000 and also, but less so, after 2000. Regional specificities are rather
diffuse: Middle-East and African countries tend to be somewhat associated with the positive
segment of F2, while CIS countries are closer to the negative values.

• The contribution of the two other axes in explaining the data structure is even lower (about

6% of total variance for each component) but reveals some additional discriminating factors.
F3 shares some of the characteristics of F2 in the macroeconomic side (like high inflation) but
is defined by larger countries (in population and total GDP), slightly poorer than the average
in 1995 but which registered a high increase in the income per capita over the entire 1995-
2006 period. The countries ranking high on this component tend to have a low participation of
services in GDP, and no particular specialization in agriculture or in manufactures, at the
difference of F1 (agriculture) and F2 (extractive industries) countries. External sector
variables indicate that high scores on F3 are associated with relatively closed economies, with
low trade coefficients in goods and, particularly, in services. They also register low values of
foreign direct investment, which further decreased after 2000. F4 countries tend to suffer from
moderate inflation, slowing down after 2000. At the difference of F3, those are small
countries in terms of population and GDP. Despite some increase in the share of rural
population, these countries show a sharp reduction in both demographic growth and share of
agriculture in GDP, especially after 2000. Open to trade, they received high FDI flows. CIS
countries tend to be associated with high scores on F4. No specific regional pattern has been
identified for F3.

2. Associating tariff policy and socio-economic variables

(a) EDA on tariffs and socio-economic data sets.

The exploratory procedure combines the results of the tariff policy exploration (applied tariffs and
water, clusters and scores) to the principal component analysis applied to the socio-economic
variables. The additional variables are simply projected on the resulting principal components, but do
not interact with their computation. Thus these components are strictly determined by the socio-
economic variables, as described in the previous section.


(i) Tariff by products and socio economic variables

At first, the applied tariffs and water for each product group are included as additional variables in the
socio-economic Principal Component Analysis. As shown in Table 10, the only clear relation
between tariff indicators (excluding agriculture) and socio-economic variables is the level of
development, measured by the first component F1. But this remains a loose relationship as F1
explains only 10 per cent of the total variance. In addition, the correlation of NAMA tariff variables
with the development variable F1 remains low (between 0.3 and 0.6). In addition, this closer
relationship is limited to the applied level, and no strong systematic link exists between water and
development. Finally, agriculture behave differently from the other product groups.

Table 10: Tariff variables and Socio-Economic Principal Components

(percentages and correlation coefficients)
F1 F2 F3 F4

Variability (%) 9.9 8.4 6.3 5.6

Cumulative (%) 9.9 18.4 24.6 30.2

Factor loading:

App_A -0.02 -0.02 -0.08 -0.08

App_B 0.5 0.1 -0.1 -0.2

App_C 0.5 0 0 -0.3

App_D 0.4 0 -0.1 -0.2

App_E 0.6 0 -0.2 -0.2

Diff_A 0.3 0.1 -0.2 0

Diff_B 0.4 0.2 -0.2 -0.1

Diff_C 0.3 0.2 -0.2 -0.1

Diff_D 0.4 0.2 -0.2 -0.1

Diff_E 0.3 0.2 -0.2 -0.1

Notes: F1 to F4 are the four principal components extracted from socio-economic variables; applied tariffs and differences
appear as supplementary variables. Factor loadings are similar to correlation coefficients in a normalized PCA.
Source: Authors' calculation on the basis of WTO tariff data.

The second socio-economic factor, F2 explaining 8% of total socio-economic variance, differentiates
oil exporters from the other economies. Its impact on tariff structure is almost nil, as it is also the case
for the remaining two principal components F3 and F4.

(ii) Overall tariff pattern and socio-economic variables

The previous section explored the relationship between socio-economic structure and tariffs for each
individual product group. But a country's trade policy is defined by considering the entirety of
products. This holistic approach led to 5 tariff policy clusters, as seen in previous sections of this
paper. The relationship between these 5 clusters and the principal components representing the socio-
economic variables is even more diffuse than for the product groups, with a correlation between 0
and 0.3 in absolute value (Table 11). Therefore, no clear and strong relationship can be identified
between the overall tariff policy and broad socio-economic variables.


Table 11: Tariff policy clusters and Socio-Economic Principal Components

(percentages and correlation coefficients)

PCA on Tariffs: Tariff Policy Clusters:
Factors: D1 D2 D3 1 2 3 4 5

F1 0.3 0.5 -0.2 0.0 -0.3 0.2 0.0 0.1

F2 0.2 0.0 0.0 -0.1 -0.1 0.1 0.0 0.1

F3 -0.2 0.0 -0.1 0.0 0.2 -0.1 0.0 -0.1

F4 0.0 -0.2 0.0 -0.1 0.1 -0.1 -0.2 0.3

Selected variables:

Fertilizers 2006 0.1 -0.4 0.0 0.0 -0.1 0.2 -0.1 0.0

GNI per cap. 1995 -0.1 -0.5 0.1 0.2 -0.2 0.1 -0.1 0.0

Trade services/GDP1995 0.3 0.0 0.1 -0.2 -0.1 -0.1 0.3 0.1

Export Food 1995 0.3 0.1 -0.1 -0.3 0.1 -0.1 0.4 0.0

Agricultural GDP 1995 0.1 0.4 -0.1 -0.1 -0.1 0.1 0.0 0.1

Ext. Debt 1995 0.2 0.3 -0.1 -0.1 0.0 -0.1 0.1 0.1

Manufacture GDP 1995 -0.3 -0.3 0.1 0.1 0.2 0.0 -0.3 -0.2
Notes: F1 to F4 are the four principal components extracted from socio-economic variables; applied tariffs and differences
appear as supplementary variables. Factor loadings are similar to correlation coefficients in a normalized PCA.
Source: Annex 1.

The strongest relationship observed between clusters and individual socio-economic variables never
exceeds 0.5 in absolute value (Annex 1). In addition, the robustness of the results was checked by
substituting the 1995 levels by the final 2006 levels, endorsing the weakness of the conclusions: we
are more in a situation of nuances than of contrasts. With these caveats in mind, the few significant
relations between tariff policy clusters and socio-economic variables that can be mentioned refer to
Cluster 1. This subset consists of economies with low applied duties and water in all products, and
regroups recently acceded WTO members and developed countries characterized by a urban
population, mature demography, low inflation, a high per capita GNI and a low weight of agriculture
in both production and trade. Even though no clear-cut regional identification can be associated, these
countries are mainly located in Europe, including the CIS.

Characterizing other clusters is much more tentative. Cluster 2 is somehow symmetric to cluster 1
from the socio-economic variables perspective; the correlation between the two clusters is -0.5. It
contains relatively poor countries, with low incidence of public services, located in Latin America or
the Caribbean. Cluster 3 economies distinguish themselves from the previous cluster countries for
their higher governmental consumption, an increase of agriculture share in GDP and more intensive
use of fertilizers. This is the only cluster associated with high applied tariffs in agriculture. Cluster 4
economies having high tariffs and deep water in both agriculture and NAMA, appear to be relatively
poor rural developing countries, mainly food exporters suffering from inflation and external debt.
These economies have nevertheless benefited from FDI flows. Cluster 5 economies are also poor
countries, predominantly rural and very similar to cluster 4 economies with high water in both
agricultural and NAMA sectors. They are still far from achieving demographic maturity, with high
population growth and have benefited from a reduction in their external debt.

Once again, these correlations are weak, and do not define clear-cut socio-economic pattern. It is clear
that the national socio-economic dimensions captured by the principal components do not improve
significantly our understanding of national tariff policies. But, as often in statistics, the null
hypothesis is not deprived of significance and despite the apparent weak outcome, this is an important
result. There is no over-determination on the overall tariff policy by broad economic considerations:
the decision-making process governing the definition of national tariff policy is the result of more
complex interactions.


(b) Discriminating tariff patterns by using economic variables

The final step of the analysis consists in investigating whether socio-economic variables, despite the
low individual correlation observed in the previous sections, can still reliably determine the five tariff
policy profiles. As a first step, a discriminant analysis investigates for each country, if its initial
clustering (based on purely tariff information) could have been inferred on the basis of socio-
economic data. The confusion matrix, obtained by comparing the prior classification (initial clusters
built on purely tariff data) to the results deduced from the socio-economic variables, provides some
interesting information, especially on countries where the classification based on tariff structure
appears unusual considering the economic situation.

Table 12: Confusion matrix for the five clusters (discriminant analysis based on socio-economic
1. Full sample

from \\ to 1 2 3 4 5 Total % correct
1 32 1 1 0 0 34 94.1%
2 0 30 0 0 1 31 96.8%
3 1 1 9 0 1 12 75.0%
4 0 1 1 11 0 13 84.6%
5 1 0 0 0 13 14 92.9%

Total 34 33 11 11 15 104 91.3%
2. Cross validation

from \\ to 1 2 3 4 5 Total

1 25 3 4 0 2 34 73.5%
2 0 26 2 2 1 31 83.9%
3 2 3 7 0 0 12 58.3%
4 0 3 2 8 0 13 61.5%
5 2 2 2 1 7 14 50.0%

Total 29 37 17 11 10 104 70.2%
Source: Authors' calculation.

Overall, most countries are found to be properly classified in the 5 clusters, except for cluster 3 where
25% of the countries seem misplaced according to their socio-economic characteristics (see part 1 of
table 8).

To cross validate the results and verify the robustness of the initial classification, each country has
been re-classified, after removing it from the sample (see part 2 of Table 12). Thanks to this
"jacknifing" process, the discriminant function is computed independently of the particular country
being classified. 19 The difference between the full sample analysis results and the cross validation
results is inversely dependent on the size of the sample and the degree of the relationship: small size
clusters or weak relationships result in large differences.

Surprisingly, cluster 1 —the largest one and the most closely associated to the principal dimensions of
the socio-economic data— reports 74% of correct assignments compared to 94% in the full sample
analysis. The difference of 20 percentage points indicates the wide variance of socio-economic

19 The general idea behind jacknifing in statistics lies in systematically recalculating the estimates
leaving out one observation at a time from the sample set; the method is normally used to estimate the sampling
distribution of a statistic. It was used here with the purpose of deriving more robust estimates of the discriminant


characteristics that exist in this cluster, including the recently acceded members to the WTO. Cluster 2
appear to be the most robust with 84% of correct assignments.

The remaining smaller size clusters (less than 15 observations), are more vulnerable to the validation
procedure. Cluster 5 is also the loosest one, as only half of its members are correctly classified
according to the validation procedure. Cluster 5 looks more like a residual grouping of very different
data points sharing a few common features (in particular, a low development level). An interesting
feature of cluster 3 is that all reassigned countries went to clusters 1 and 2; cluster 3 mainly includes
developed countries with relatively protectionist tariff policies.

When looking at individual economies, in the group of developed countries Norway and Iceland are
reassigned from Cluster 3 to 1, Japan and Korea from 1 to 2. On the side of developing countries,
Kuwait and unexpectedly Bangladesh move from Cluster 5 to 1 and Guyana from 4 to 2. Moves from
3 to 2 relates to relatively advanced developing countries such as Colombia or Tunisia. On the other
hand, Gabon and Madagascar are reclassified from 1 to 3 and Botswana, South Africa from 1 to 5 (the
re-assignation of SSA to the 5th cluster is nevertheless loose, as the discriminant function identified a
25% probability of pertaining to group 1). China and Guinea move from 1 to 3, and Nicaragua from 2
to 4.

Additional explorations incorporated the regional dimension. Geographical variables improve the
direct classification, but deteriorate the cross-validation results and show that regional groupings are
poor predictors of tariff patterns. This result (not shown here) was already perceptible earlier in the
study, when looking at the large differences across patterns in maps 1 to 4.


In the present study 120 economies are analysed in the basis of their applied duties and binding
overhang; the main purpose is to illustrate the margins that economies, under WTO rules, are allowed
to use unilaterally at the MFN level, and to identify and understand potential interactions determining
tariff policies. The analysis, which proceeds from general to specific, applies exploratory data
analysis to identify underlying patterns behind the data, without any pretence at identifying the
rational behind the observed structure. Five broad clusters of countries are determined on the basis of
their tariff policy, and the relationship between this classification and socio-economic structures is

A first element of conclusion is that tariff policies seem to protect more the labour intensive sectors,
with some variations in the case of agriculture, rather than pursuing traditional protectionist industrial
policies through "effective protection". A second element of conclusion is that in tariff policies, the
distinction between developing and developed countries is broadly relevant, but not overly
determinant. For example, while the extreme clusters 4 and 5 are constituted of developing countries,
there is a mixture of developed and developing economies in all remaining clusters. A third
conclusion is that agricultural tariff policy becomes a strong discriminating factor across countries as
soon as other variables, such as the level of development level, are taken into consideration.

When the product sectors are considered independently, economies seem to follow different tariff
strategies to protect their products. Globally and for the majority of countries, the agricultural sector
is more protected than the NAMA in terms of applied duties and water, while textiles and clothing is
the most protected NAMA sector. For a given country, the tariff structure varies across sectors. Even
if developed economies tend to share more common features than developing ones, no clear
relationship can be established between the level of development and the overall tariff structure.

From the negotiation margins perspective, developing countries that have joined the WTO recently
faced stricter negotiations with their partners resulting in relatively low applied duties and water; as a
result, these countries, like the developed economies, have a reduced negotiation margins in almost all


On the other side of the spectrum, high negotiation margins are not a privilege for LDCs; some
developing and developed countries benefit from deep water in some sectors. Unexpectedly, some
LDCs show very small negotiation margins as they apply low duties and have only shallow water in
some of the product sectors. On the other hand, some developed countries make use of high applied
duties in addition to deep water in agriculture. Globally, applied tariffs and water respond to a
different logic.

By exploring the relationship between the tariff policy and the socio-economic dimensions, the only
clear relation which emerged for individual product groups (excluding agriculture) is the level of
development. Specifically, in the NAMA sector, the applied tariffs for raw materials and
pharmaceutical/hi-tech products are highly correlated to the level of development. However, the
relationship remains limited to the level of the applied tariffs; no strong systematic link has been
identified between the water and the socio-economic dimension. Agriculture behaves also differently
from other product groups and its level of protection seems not related to the level of development.

The relationship between the overall tariff policy (considering all product groups together) and the
socio-economic variables is even more diffuse, and no strong relationship emerges between tariff
policy clusters and the socio-economic context. Consequently, we can conclude that trade policy is
not over-determined by economic considerations: the decision-making process defining a precise
trade policy is the result of more complex interactions.

At last, discriminant analysis is used to validate the robustness of the clustering results and
investigates to what extend the belonging of countries to a tariff policy cluster could be predicted
based only on their socio-economic profile. The first and most numerous cluster of low protection
economies with shallow water, is particularly affected as many countries are reclassified in other
clusters. This indicates the wide variance of socio-economic conditions within the cluster, but also
the bias created by the countries of recent accession to the WTO, with a tariff structure resulting from
strict bilateral negotiations rather than socio-economic factors. Conversely, a few industrialised
countries show a wider than expected dispersion of their tariff in relation to their development status.

It is often believed that national trade policies are largely determined by socio-economic concerns,
and that structural factors and factor endowments, which orient the productive specialisation of
economies should determine the position of each country in the negotiation, as well as its propensity
in entering into strategic coalitions. By showing that actual tariff structures are only loosely
determined by such structural factors, our results do not support these premises.

* *



Bchir M. H., Jean S. and Laborde D. (2005), " Binding overhang and tariff cutting formulas". CEPII

Costantini V., Crescenzi R., De Filippis F. and Salvatici L. (2007), "Bargaining Coalitions in the
WTO Agricultural Negotiations". The World Economy pp. 863891.

Decreux Y. and Fontagné L., (2009), " Economic impact of potential outcome of the DDA". CEPII

Foletti L., Fugazza M., Nicita A. and Olarreaga M. (2009) "Smoke in the Water". World Bank,
Policy Research Working Paper 5050.

Fontagné L., Guérin J-L. and Jean S. (2005), " Market Access Liberalisation in the Doha Round:
Scenarios and Assessment". Blackwell Publishing Ltd 2005, p. 1073-1094.

François F. and Martin W. (2004),"Commercial policy variability, bindings, and market access".
European Economic Review 48 (2004) 665-679.

François F., Martin W. and Manole V. (2005),"Choosing Formulas for Market Access Negotiations:
Efficiency and Market Access Considerations" World Bank, Policy Research Working Paper
Series number 3474.

Francois J., Van Meijl H., Van Tongeren F (2003), " Trade Liberalization and Developing Countries
under the Doha Round". Tinbergen Institute TI 2003-060/2.

Gibson P., Wainio J., Whitley D., And Bohan M. (2001), "Profiles of tariffs in global agricultural
markets". Agricultural Economic Report No. (aer796) 44 pp, January 2001.

Goode W. (2003), Dictionary of Trade Policy Terms, Fourth Edition.
Jones K. (2007), "Regionalism and the problem of representation in the WTO". UNU-CRIS Working

Papers W-2007/12.
World Bank - World Development Indicators (on-line data base, data extracted in July 2009).
WTO - Integrated Data Base, Consolidated Tariff Schedules database (on-line data bases, extracted in

February 2008)
WTO - UNCTAD - ITC (2006) World Tariff Profiles.




PCA on Tariffs: PCA on Socio-economic data: Tariff Policy Clusters: Regions: Recent accession

Variables or components D1 D2 D3 F1 F2 F3 F4 1 2 3 4 5 ASPA CIS EUR LAC MENA NAM SSA RAM-0 RAM-1

Inflation (2000) 0.1 -0.1 0.0 0.2 0.5 0.4 0.4 -0.1 0.0 -0.1 0.2 0.0 -0.1 0.0 -0.1 0.0 0.0 0.0 0.2 -0.2 0.2

Inflation (2006) 0.2 0.0 0.0 0.3 0.5 0.5 0.3 -0.2 0.1 -0.1 0.3 0.0 -0.1 0.0 -0.1 0.0 -0.1 -0.1 0.2 0.0 0.0

dInflation(2000-2006) -0.1 0.1 0.0 -0.2 -0.5 -0.4 -0.4 0.1 0.0 0.0 -0.2 0.1 0.1 0.0 0.0 0.1 0.0 0.0 -0.1 0.2 -0.2

FDI (1995) 0.2 -0.2 -0.1 0.0 0.2 -0.2 0.4 -0.1 -0.1 -0.1 0.3 0.1 0.0 0.0 -0.1 0.2 -0.1 0.0 -0.1 0.0 0.0

dFDI (2000-2006) 0.0 0.0 0.2 0.0 -0.3 -0.5 0.1 -0.1 0.0 0.1 0.1 -0.1 0.0 0.1 0.0 0.1 0.2 -0.1 -0.2 -0.2 0.2

Military 1995 0.0 -0.1 -0.1 -0.3 0.4 0.0 0.2 0.1 -0.1 0.0 0.0 0.0 0.0 0.0 0.0 -0.2 0.6 0.0 -0.1 -0.2 0.2

dMilitary(1995-2000) -0.1 0.1 0.1 0.3 -0.4 -0.1 -0.4 -0.1 0.2 0.1 -0.1 -0.1 0.0 0.0 -0.2 0.1 -0.2 0.0 0.1 0.2 -0.2

Population 1995 -0.1 0.1 0.1 -0.2 -0.1 0.4 -0.2 0.1 -0.1 0.1 -0.1 0.0 0.3 -0.1 -0.1 -0.1 -0.1 0.1 -0.1 0.0 0.0

dPopulation (1995-2000) 0.3 0.4 -0.2 0.4 0.4 -0.2 -0.4 -0.3 0.0 0.1 -0.1 0.3 -0.1 -0.4 -0.4 -0.2 0.1 -0.1 0.5 0.1 -0.1

dPopulation (2000-2006) 0.3 0.4 -0.2 0.6 0.3 -0.2 -0.4 -0.3 0.1 0.2 0.0 0.2 -0.1 -0.4 -0.3 -0.2 0.1 -0.1 0.6 0.1 -0.1

Rural 1995 0.3 0.4 0.0 0.7 -0.2 0.1 0.0 -0.2 -0.1 0.0 0.2 0.2 0.1 0.0 -0.2 -0.2 -0.3 -0.2 0.5 -0.1 0.1

Trade/GDP 1995 0.1 -0.2 -0.1 -0.2 0.2 -0.4 0.5 0.0 -0.1 -0.1 0.2 0.0 0.1 0.0 -0.1 0.0 0.0 -0.1 -0.1 -0.1 0.1

Current Balance 1995 -0.2 -0.2 0.1 -0.5 0.0 0.1 -0.3 0.1 0.1 0.0 -0.1 -0.1 0.2 -0.1 0.2 -0.1 0.2 0.1 -0.2 0.1 -0.1

dCurrent Bal (1995-2000) 0.0 0.0 0.0 0.0 0.7 0.0 0.1 0.1 -0.1 0.0 -0.1 0.1 0.0 0.0 0.0 -0.2 0.2 -0.1 0.1 -0.1 0.1

Fertilizers 2006 0.1 -0.4 0.0 -0.5 0.2 -0.3 0.1 0.0 -0.1 0.2 -0.1 0.0 0.1 -0.1 0.1 -0.1 0.3 -0.1 -0.3 0.1 -0.1

GNI per cap. 1995 -0.1 -0.5 0.1 -0.8 0.2 -0.2 -0.1 0.2 -0.2 0.1 -0.1 0.0 0.0 -0.1 0.4 -0.1 0.2 0.3 -0.4 0.2 -0.2

dGNI/H (2000-2006) -0.1 -0.1 0.0 -0.1 0.0 0.6 0.5 0.2 -0.1 -0.1 0.0 0.0 0.2 0.4 0.0 -0.1 0.0 -0.1 -0.2 -0.3 0.3

d2GNI/H (1995-2006) 0.1 0.0 -0.1 0.1 0.1 0.5 0.1 0.0 0.0 -0.1 0.0 0.1 0.1 0.2 -0.2 -0.1 0.1 -0.2 0.0 -0.1 0.1

Gov. Consumption 1995 0.0 -0.3 0.0 -0.4 0.0 -0.2 0.3 0.1 -0.3 0.2 0.1 0.0 -0.2 0.1 0.2 -0.1 0.3 0.0 -0.1 -0.1 0.1

Food imports 1995 0.1 0.3 -0.1 0.6 -0.1 -0.4 0.2 -0.1 -0.1 0.0 0.2 0.1 -0.3 0.1 -0.1 0.0 0.1 -0.2 0.3 -0.2 0.2

Manufacture imports 1995 0.1 -0.2 0.0 -0.5 0.2 0.0 -0.3 -0.1 0.0 0.0 0.0 0.1 0.1 -0.4 0.1 0.2 -0.1 0.2 -0.2 0.2 -0.2

Trade goods/GDP1995 0.1 -0.2 -0.1 -0.2 0.3 -0.3 0.5 0.1 -0.1 -0.1 0.1 0.0 0.1 0.0 -0.1 -0.1 0.1 -0.1 0.0 0.0 0.0

Trade services/GDP1995 0.3 0.0 0.1 0.0 0.1 -0.6 0.5 -0.2 -0.1 -0.1 0.3 0.1 0.0 -0.1 -0.1 0.2 0.0 -0.1 -0.1 -0.1 0.1

Export Food 1995 0.3 0.1 -0.1 0.4 -0.3 -0.3 0.1 -0.3 0.1 -0.1 0.4 0.0 -0.2 0.0 -0.1 0.4 -0.2 -0.1 0.1 0.1 -0.1

Export Fuels 1995 0.0 0.0 0.0 -0.1 0.8 0.0 0.0 0.1 -0.1 -0.1 0.0 0.0 -0.1 0.0 0.0 -0.1 0.3 0.0 0.0 -0.2 0.2

Export manuf. 1995 -0.2 -0.2 0.2 -0.6 -0.2 0.3 0.1 0.2 -0.1 0.0 -0.1 -0.1 0.4 0.0 0.2 -0.2 0.0 0.2 -0.4 0.0 0.0

Agricultural GDP 1995 0.1 0.4 -0.1 0.8 -0.4 0.1 0.1 -0.1 -0.1 0.1 0.0 0.1 0.0 0.3 -0.1 -0.3 -0.2 -0.2 0.4 -0.2 0.2

dAg-GDP (1995-2000) 0.1 0.1 0.1 -0.1 0.2 -0.2 -0.4 -0.2 0.2 0.0 0.0 0.0 0.0 -0.4 -0.1 0.0 0.1 0.1 0.2 0.3 -0.3

Ext. Debt 1995 0.2 0.3 -0.1 0.6 0.4 -0.1 0.1 -0.1 0.0 -0.1 0.1 0.1 -0.2 -0.1 -0.2 0.0 0.0 -0.1 0.4 0.0 0.0

dExt. Debt (1995-2000) -0.2 -0.1 -0.1 -0.2 -0.6 0.1 -0.1 0.1 0.0 0.1 -0.1 -0.1 0.1 0.3 0.1 0.0 0.0 0.0 -0.2 0.0 0.0

dExt. Debt (2000-2006) -0.2 -0.3 0.1 -0.6 -0.2 -0.1 0.1 0.0 0.0 0.0 0.0 -0.2 0.2 -0.1 0.2 0.3 0.0 0.1 -0.5 0.0 0.0

Industrial GDP 1995 -0.1 -0.2 0.0 -0.4 0.6 0.3 0.1 0.1 0.0 -0.1 0.0 -0.1 0.1 -0.1 0.0 -0.1 0.2 0.0 -0.1 0.0 0.0

dIndustrial GDP (1995-2000) -0.1 0.1 -0.1 0.2 0.5 -0.2 0.0 0.2 -0.1 0.0 0.0 0.0 0.0 0.1 -0.1 -0.1 0.0 0.0 0.1 -0.2 0.2

Manuf. GDP 1995 -0.3 -0.3 0.1 -0.5 -0.2 0.3 0.0 0.1 0.2 0.0 -0.3 -0.2 0.3 0.1 0.1 0.0 0.0 0.1 -0.4 0.1 -0.1

Service GDP 1995 -0.1 -0.2 0.1 -0.6 -0.1 -0.4 -0.2 0.0 0.1 0.0 0.0 -0.1 -0.1 -0.3 0.1 0.4 0.1 0.2 -0.4 0.2 -0.2

dServiceGDP (1995-2000) 0.0 -0.2 -0.1 0.0 -0.6 0.3 0.3 0.0 0.0 0.0 0.0 0.0 0.0 0.3 0.1 0.0 0.0 0.0 -0.2 -0.1 0.1

Note: Correlation coefficients, except for the "D" and "F" PCA components: Factor loading.
Sources: Authors' calculation on the basis of WTO tariff data and World Bank's World Development Indicators.