Comparison of Revealed Comparative

Advantage Indices with Application to Trade

Tendencies of East Asian Countries

Elias SANIDAS, Yousun SHIN

Department of Economics, Seoul National University

1
 Comparison of Revealed Comparative Advantage Indices
with Application to Trade Tendencies of East Asian Countries

Abstract

One of the most powerful propositions of classical trade theory is that the
pattern of international trade is determined by comparative advantage. That is,
a country with the comparative advantage in a given commodity exports, and the
other with the comparative disadvantage imports. Thus, the question has been
where then the comparative advantage originates from, and there have been
numerous attempts to identify the economic conditions that determine
comparative advantage.

Ballance et al. (1987) provided a simple theoretical framework that allows

us to clearly look at the relationship between the theoretical notion of
comparative advantage and the measures of comparative advantage we
practically obtain.

According to the above diagram, economic conditions that vary

across countries determine the international pattern of comparative advantage
, which lies under the pattern of international trade, production and
consumption . The relationship between , and can be
understood as what the international trade theories have been trying to identify:
what kind of economic conditions determine comparative advantage that makes
the trade takes place, and how the trade is going to affect the economy. The
classical and neo-classical trade models (Ricardo, 1817/1951, Ohlin, 1933) claims
that a country with an economic condition in which it has an ability to produce a
given commodity at a relatively lower costs, i.e. comparative advantage, exports
the commodity, while the other with comparative disadvantage imports. The new
trade theory, which explains the occurrence of intra-industry trade based on
imperfect competition and economies of scale, does not directly use the term
„comparative advantage‟; however, in the above framework, having economies of
scale can be also interpreted as having comparative advantage in a broader
sense that it reflects a lower opportunity cost and cases international trade.

Despite the powerful influence and usefulness of these trade theories, it

has been always difficult to apply the theoretical concept of comparative
advantage in empirical analyses, especially when trying to measure the
comparative advantage in analyzing trade performance, since the notion of
comparative advantage usually takes into account autarkic variables, such as
autarkic relative prices and autarkic production costs, which are not observable.
Thus, as the second-best methodology, indices of revealed comparative
advantage , which are our interest here, are constructed based on

2
and possibly other post-trade variables in order to identify the underlying
pattern of comparative advantage . That is, due to the practical limitations,
RCA indices are made to function as a tool to trace back the pattern of CA by
using the results assumably governed by the pattern of CA.

One of the first attempts to measure comparative advantage was

Balassa‟s (1965) RCA index (BI) using the variables generated from the post-
trade equilibria, which is so far the most widely used index in analyses of
comparative advantage. Although used by many researchers, BI has been under
critique for its alleged incomparability and inconsistency, and therefore several
other attempts to measure comparative advantage have been taken place to
overcome the shortcomings of BI.

Those newly suggested indices can be classified in three classes: trade-

cum-production indices containing both of trade and production variables, e.g.
Lafay index (LI) (Lafay, 1992); exports-only indices containing only exports
variables, e.g. symmetric RCA index (SI) (Dalum et al., 1998), weighted RCA
index (WI) (Proudman and Redding, 2000), and additive RCA index (AI) (Hoen
and Oosterhaven, 2006); and indices using hypothetical situation such as
comparative-advantage-neutral point, e.g. normalized RCA (NI) (Yu et al., 2009).

There are several ways of using the RCA indices in analyzing trade
performance. The most common ways are a) to simply examine whether a given
country has a comparative advantage in a given sector by comparing the
calculated value and the comparative advantage neutral point 1 , b) make a
comparison across sectors within a given country or across countries with respect
to a given sectors by using rankings in order of the calculated index values, and c)
to examine how much of comparative advantage or disadvantage a given country
gained during the period of interest by directly comparing the calculated index
values.

Furthermore, the indices also can be used in econometric analyses, such

as in Galtonian regression in order to see the structural changes of trade
performance. Galtonian regression was initially introduced by Cantwell (1989) to
measure technological comparative advantage and subsequently used by several
other scholars to measure trade comparative advantage. This simple OLS
method allows us to compare two cross-sections at two different points of time,
and tells us how much change in the structure of trade specialization in a given
country is made during the period of interest.

However, the normality of error terms assumed in the OLS regression

hinders using the RCA indices in regression analysis due to the existence of

1 For example, the comparative advantage neutral point of BI is unity. When a given country
shows BI=1.5 in a given sector, the country is considered to have a comparative advantage in the
sector.
3
outliers, which results in violating the normality. Thus we suggest trying some
transformation skills such as log transformation and Box-Cox transformation in
order to make the distributions of indices normal. We also suggest using the
robust regression and the quantile regression that yield more effective results
with the existence of outliers and resolve the normality issue, and interpreting
the relevant results in a different way from Cantwell‟s (1989) together with the
Spearman rank correlation coefficients.

The aim of this paper is also to systemically compare all major attempts
of measuring comparative advantage thorough RCA indices, examine the pros
and cons of these indices, and the relationship between them, and thus
eventually in order to find out how to adequately use them. To do that, we first
theoretically examine the six RCA indices with regard to the ways of using them.
Then we apply this discussion in real cases by taking an example of East Asian
countries, namely, China, Japan and South Korea: we calculate the six indices
for the three countries, using ITC (International Trade Centre) trade data from
1995 to 2008 based on Harmonized System (HS) 2-digit level of aggregation,
which consists of 98 sub-headings (or sectors). Besides, more South East Asian
counties are added in cross-country analysis in order to make more appropriate
comparison.

Thus, we find, first of all, that using different RCA indices yields very
different results when used in non-econometric comparative analysis: in
analyzing trade performance, one needs to be careful interpreting the results by
using different indices. Secondly, we find that there is not a perfect RCA index:
each index has advantage and disadvantages depending on the ways of using it,
although the NI seems to have more favorable features as an RCA index than
the others. For example, the SI is not comparable across sectors or countries,
while the NI is. Thirdly and lastly, we also find that, when using the RCA
indices with the robust regression and the quantile regression and making an
interpretation as suggested in this study, we can witness that the difference
across the RCA indices is much less.

Key Words: Comparative Advantage, Revealed Comparative Advantage Index, Trade

Specialization, Trade Performance Measure, Galtonian Regression

4
1. INTRODUCTION
One of the most powerful propositions of classical trade theory is that the
pattern of international trade is determined by comparative advantage. That is,
a country with the comparative advantage in a given commodity exports, and the
other with the comparative disadvantage imports. Thus, the question has been
where then the comparative advantage originates from, and there have been
numerous attempts to identify the economic conditions that determine
comparative advantage.

Despite the powerful influence and usefulness of these trade theories, it

has been always difficult to apply the theoretical concept of comparative
advantage in empirical analyses, especially when trying to measure the
comparative advantage in analyzing trade performance, since the notion of
comparative advantage usually takes into account autarkic variables, such as
autarkic relative prices and autarkic production costs, which are not observable.
Thus, as a second-best methodology, indices of revealed comparative advantage
(RCA), which are our interest here, are constructed based on post-trade variables
in order to identify the underlying pattern of comparative advantage. That is,
due to the practical limitations, RCA indices are made to function as a tool to
trace back the pattern of comparative advantage by using the results supposedly
governed by the pattern of comparative advantage.

The major innovation regarding this issue was done by Balassa‟s (1965),
which is so far the most widely used index in analyzing trade performance. There
are three aims using the BI indices in analyzing trade performance. The most
traditional and common ways are a) to simply examine whether a given country
has a comparative advantage in a given sector by comparing the calculated value
and the comparative advantage neutral point (dichotomous measure)2, b) make a
comparison across sectors within a given country or across countries with respect
to a given sectors by using rankings in order of the calculated index values
(ordinal measure), and c) to examine how much of comparative advantage or
disadvantage a given country gained during the period of interest by directly
comparing the calculated index values (cardinal measure). However, the BI has
been criticized for not achieving comparability especially in terms of the second
and third aims (e.g. Bowen, 1983, Cai and Leung, 2008, De Benedictis and
Tamberi, 2001, De Benedictis and Tamberi, 2004, Hillman, 1980, Yu et al., 2009).

Therefore, several other attempts to measure comparative advantage

have been taken place to overcome the shortcomings of the BI. The newly
suggested RCA indices can be classified into three classes: trade-cum-production
indices containing both of trade and production variables, e.g. Lafay index (Lafay,

2 For example, the comparative advantage neutral point of BI is unity. When a given country
shows BI=1.5 in a given sector, the country is considered to have a comparative advantage in the
sector.
5
1992); exports-only indices containing only exports variables, e.g. symmetric
RCA index (Dalum et al., 1998), weighted RCA index (Proudman and Redding,
2000), and additive RCA index (Hoen and Oosterhaven, 2006); and indices using
hypothetical situation such as comparative-advantage-neutral point, e.g.
normalized RCA (Yu et al., 2009).

More specifically, these indices can be used in econometric analyses, such

as in Galtonian regression in order to see the structural changes of trade
specialization. Galtonian regression was initially introduced by Cantwell (1989)
to measure the structural changes of technological comparative advantage over
time and subsequently used by several other scholars in terms of trade
comparative advantage. This simple OLS method allows us to compare two
cross-sections at two different points of time, and tells us how much change in
the structure of trade specialization in a given country is made during the period
of interest, e.g. how stable or not the trade specialization of the country is.
However, since the simple OLS model assumes the normal distribution of error
terms, which the RCA indices in most cases do not follow, it has been
problematic to directly use the RCA indices in the regression analysis, although
there have been some solutions suggested (Dalum et al., 1998, Laursen, 1998).

The main aim of this paper is to inform empirical researchers about the
issues related to measuring comparative advantage and to suggest a practical
strategy in analyzing trade performance by using the RCA indices. Precisely, the
hypotheses are: a) the problems inherent in the RCA indices can be resolved; b)
the RCA indices are useful; c) it is possible to formulate a strategy for
researchers.

To check these hypotheses, we first conduct an in-depth exploration of

selective RCA indices including the BI, and how to properly use them in
econometric analysis as well as in non-econometric comparative analysis. As
Vollrath (1991) points out, there can be as many RCA indices as there are
combinations and transformations of post-trade variables, and there actually
have been numerous attempts to measure comparative advantage. Nonetheless,
we limit our interest into the Balassa‟s RCA index and its five alternatives
mentioned above that tried to adjust the problems of the BI3; the discussion
about these selective indices will still be able to cover in general most of the
issues related to RCA indices. We find that each index has pros and cons and
none of them can be „the perfect one,‟ thus it is important to well acknowledge
their characteristics and properly use them.

Then, we apply this exploration into real cases by taking an example of

East Asian economies, namely, China, Japan and South Korea. Although, we add
more South East Asian countries, namely, Hong Kong, Indonesia, Malaysia,

3 It is beyond the scope of this study to examine all of the possible indices. Further research can
be based on the present study.
6
Philippines, Singapore and Thailand, for cross-country analysis in order to
conduct more appropriate comparison. We calculate the six RCA indices for the
three countries (and for the other South East Asian countries when needed)
using ITC (International Trade Centre) trade data from 1995 to 2008 based on
Harmonized System (HS) 2-digit level of aggregation, which consists of 98 sub-
headings (sectors). We empirically compare the six RCA indices in terms of
dichotomous, ordinal and cardinal measurement in order to examine how
different these indices are in an empirical sense.

Also, we try to investigate whether the six RCA indices can be used for
regression analyses and how they should be used: we find that in most of the
cases the six RCA indices violate the normality assumption of the error terms in
OLS regression, which implies that using the RCA indices possibly can lead to an
invalid inference. Thus, we apply different ways of using the RCA indices in
regression analysis: we suggest trying the log transformation and the Box-Cox
transformation in order to obtain normality; and the robust regression, the
quantile regression methods, and the Spearman rank correlation, which do not
assume normality and so on as it will be explained further below.

Thus, we find, first of all, that using different RCA indices yields different
results: in analyzing trade performance, one needs to be very careful interpreting
the results by using different indices, although the NI has the most favorable
theoretical features that are needed in comparative analyses. Secondly, when
using the RCA indices for regression analyses, since in most of the cases the
assumptions for OLS regression are not met, one would rather find the robust
regression or the quantile regression a more effective way of using them. In
addition, we also suggest a different way of interpreting the results, instead of
the existing way that simply compares the regression coefficients.

This study provides only an exploratory analysis on the RCA indices and
how to use them: despite the wide use and popularity of the BI in analyzing
trade specialization performance, there has not been a lot of work done on
actually how to use RCA indices so far. As mentioned before, there is not a
perfect RCA index: each index has advantages and disadvantages depending on
circumstances. That is, despite the problems of using indices, we can still
manage to find the underlying potential and the ways to properly interpret them
according to the strategy suggested here.

Section 2 examines the relationship between the concept of comparative

advantage from trade theories and the RCA indices. Section 3 explores the
Balassa‟s and the five alternative RCA indices in a mathematical and statistical
sense, and suggests ideas of how to use them. Section 4 empirically shows the
process of applying the ideas suggested in Section 3 with an example of East
Asian economies. Section 5 summarizes and concludes.

7
2. THEORETICAL UNDERPINNING

2. 1. Comparative Advantage

The classical and neo-classical trade models (Ricardo, 1817/1951, Ohlin,

1933) claim that a country with comparative advantage, i.e. an economic
condition in which it has an ability to produce a given commodity at a relatively
lower costs, exports the commodity, while the other with comparative
disadvantage imports. Ricardian model finds that international trade takes place
from differences in labor productivity between countries while the Hecksher-
Ohlin model explains it based on the differences in factor endowments and
intensities.

The New Trade Theory, which explains the occurrence of intra-industry

trade taking into account the assumptions of imperfect competition and
increasing returns of scale, does not directly use the term „comparative
advantage‟: Krugman (1979) clearly stated that he developed “a simple, general
equilibrium model of non-comparative advantage trade” (p.469). The term
„comparative advantage‟ here, however, is used in a traditional sense: that is, it
should be interpreted as what triggers inter-industry trade, contrasted to
„increasing returns of scale‟ that causes intra-industry trade.

In contrast to this traditional notion comparative advantage has also been

used to contain a broader meaning than simply what causes inter-industry trade:
we may call it „a broader notion of comparative advantage.‟ The modern notion of
comparative advantage can be easily found in the literature: for example,
„internal returns to scale as a source of comparative advantage‟ (Tybout, 1993),
or „product differentiation as a source of comparative advantage‟ (Hummels and
Levinsohn, 1993). Palley (2008, p.201)‟s following argument seems to be helpful
to understand this more clearly:

“There is a new emphasis on the fact that comparative advantage in the modern
world is created and not endowed. In the 18th century world, trade was driven by
the search for exotic spices and raw materials. In that epoch, climate and natural
resource endowments significantly determined the pattern of comparative
advantage [in a traditional sense]4, and little could be done to alter this pattern. In
today‟s economy, comparative advantage [in a broader sense] is driven by
technology, and technology can be importantly influenced by human action and
policy.”

Putting aside the discussion between the traditional and broader notion of
comparative advantage, Davis (1997) claims that having the crucial link between

4 The terms in the brackets are added by the author for emphasis.
8
endowments and trade volumes, it is possible to explain the large volume of
North-North trade with the conventional model based on the concept of
comparative advantage. That is, the traditional notion of comparative advantage
can still be the main theoretical explanation for the pattern of international
trade, although the emergence of the New Trade Theory put an equal importance
on increasing returns of scale as much as on comparative advantage in a
traditional sense. (De Benedictis and Tamberi, 2001)

In summary, the notion of comparative advantage, apart from discussing

whether it is concerned only with a certain part of international trade or not, can
possibly straddle over different types of international trade theories, and can be
generally understood as a source that enables international trade to take place.
In this sense, Lafay (1992, p.210) shows a proper example of the broader notion
of comparative advantage, pointing out “the various economic conditions from
which firms can acquire the comparative advantage in a given territory: 1.
favorable natural resources endowment of the territory concerned; 2. lower
relative costs through the choice of segments that are best suited to the
macroeconomic factors of production; 3. lower relative costs through innovation
at microeconomic level in the production process; and 4. the acquisition of
monopoly elements through the microeconomic creation of new products.”

2. 2. Comparative Advantage 5 and ‘Revealed Comparative Advantage

Despite the powerful influence and usefulness of the concept of

comparative advantage, two major problems arise when applying this theoretical
concept in empirical analyses, especially when trying to measure the
comparative advantage in analyzing trade performance.

Ballance et al. (1987) provided a simple theoretical framework that shows

the relationship between the theoretical notion of comparative advantage and
the measures of comparative advantage we practically obtain. The following
diagram (1) allows us to look at the issue more clearly.

According to (1), economic conditions that vary across countries

determine the international pattern of comparative advantage , which lies
under the pattern of international trade, production and consumption .
The relationship between , and can be understood as what the
international trade theories have been trying to identify: what kind of economic
conditions determine comparative advantage that makes the trade to take place,
and how the trade is going to affect the economy.

5 The term „comparative advantage‟ from here is used in a broader sense. See Section 2.1 for
detailed discussion.
9
 The most preferable scenario would be to find the adequate variables that
determine comparative advantage and carry out the calculations that correspond
to the trade theories, e.g. calculating the difference between the autarkic and
free trade relative prices. The first problem arisen here is that the concept of
comparative advantage is in nature associated with unobservable autarkic
variables, such as autarkic prices or autarkic production costs. Hence,
empirical researchers have been looking for the second-best way to measure
comparative advantage by using available post-trade data such as exports and
imports, which is so-called „revealed comparative advantage indices.‟ The
RCA indices that have been suggested in order to identify the underlying pattern
of CA are constructed from and possibly other post-trade variables. That is,
due to these practical limitations, RCA indices are made to function as a tool to
trace back the pattern of CA by using the results assumably governed by
the pattern of CA. The implication of this RCA approach is that there can be as
many RCA indices as there are combinations and transformations of TPC
variables used to infer comparative advantage (Vollrath, 1991).

The second problem here is that determining comparative advantages in

the world with more than two countries, two commodities, and two factors is
often not as clear and straightforward as in a simple two-factor-cum-two-country
hypothetical economy. That is, some measure that can provide more information
than simply whether or not a given country has revealed comparative advantage
in a given commodity is needed. Empirical researchers thus cannot avoid a
question pertaining to „the degree of advantage exhibited by a particular country
over various goods,‟ or „the degree of advantage exhibited by various countries
with respect to a particular traded good‟ (Hillman, 1980).

At the same time, we can consider another issue here regarding the
relationship between measurement and theory. De Benedictis and Tamberi
(2004) bring up the issue of the independence of measurement from theory,
referring to the literature on oligopoly theory (Cordella, 1998), according to
which it is implied that even if autarkic relative prices were observable they
might not predict true comparative advantage.

Bearing in mind the issues so far discussed above, it seems that the
indices of RCA6 do not necessarily have an accurate and exclusive linkage with
the concept of CA. According to Drabicki and Takayama (1979), the deterministic
relationship between CA and TPC is not valid in a multi-country, commodity and
factor world. However, it is also demonstrated by Deardorff (1980) that there is a
negative correlation between net exports and relative autarkic prices under
relatively general conditions, which leads us to conclude that, while comparative
advantage may not be accurately measurable, indices based on post-trade
observations may „reveal‟ much more about the underlying pattern of

6 See Section 3 for full discussion on RCA indices.

10
comparative advantage (Ballance et al., 1987). That is, despite the lack of
theoretical accuracy of RCA indices, we still have a hope in that they can give us
proper information with respect to comparative advantage.

3. RCA INDICES AND THEIR USE

Conventionally, RCA indices have been used as measures, which can be
interpreted in three ways. First, they can provide a demarcation regarding
whether or not a given country enjoys a comparative advantage on a given
commodity (dichotomous measure); second, they can provide rankings of sectors
within a given country in cross-sector analysis, and rankings of countries with
respect to a given commodity in cross-country analysis (ordinal measure); lastly,
they can quantify the degree of comparative advantage enjoyed by a given
country with respect to a given commodity (cardinal measure) (Ballance et al.,
1987). On the other hand, there also have been scholars who challenged the use
of the RCA indices in econometric analysis, especially in analyzing structural
changes of trade specialization, for example, by using Galtonian regression
model. Here, we discuss all these issues by taking into account selected RCA
indices.

3.1. Measuring Comparative Advantage

As mentioned in Section 2, there can be as many RCA indices as there are

combinations and transformations of post-trade variables. (See Balance et al.
(1987); Memedovic (1994); and Vollrath (1991) for information on a variety of
RCA indices.) We introduce here the selected major RCA indices including the
famous Balassa‟s revealed comparative advantage index (henceforth BI), and the
subsequent attempts to overcome the shortcomings of BI.

Although BI has been a widely used index, it may not be the best trade
performance index. It has been accused of bearing technical problems when it
comes to ordinal and cardinal comparisons of BI values (Yeats, 1985). It is not
surprising that there have been numerous attempts of adjusting and
transforming BI, and suggestions of alternative indices for measuring
comparative advantage. (e.g. Bowen, 1983, Cai and Leung, 2008, Dalum et al.,
1998, Hillman, 1980, Hoen and Oosterhaven, 2006, Lafay, 1992, Proudman and
Redding, 2000, Yu et al., 2009)7

7 See Memedovic (1994) and Vollrath (1991) for an account of RCA indices.
11
3.1.1 Balassa’s Revealed Comparative Advantage Index

The major innovation with regard to how to measure the comparative

advantage was done by Balassa (1965). He adopted Liesner (1958)‟s idea of using
relative export performance, and proposed using the ratio of export shares as an
index for comparative advantage, which is so far one of the most widely used
trade performance indices.8 Using to denote exports, a specific country, a
specific commodity and
9 the world (or any reference group of countries taken
into account), the BI is:

It is noteworthy that: ; ; .

A given country is considered to have comparative advantage

(disadvantage) in commodity , when the commodity ‟s exports market size of
country in terms of its total national exports market size is greater (less) than
the commodity ‟s world exports market size in terms of the world total exports
market size, i.e. when is greater (less) than unity. Similarly interpreting, a
given country is considered to have comparative advantage (disadvantage) in
commodity , when country ‟s exports market size for commodity in terms of
commodity ‟s world export market size is greater (less) than the country ‟s
total national exports market size in terms of the world total exports market size.
Thus, the comparative advantage neutral (or, dichotomy) point is when is
equal to unity, i.e. when the size-wise importance of commodity ‟s market in the
country is as big as that in the world export market. The BI method is
basically comparing a given sector‟s export performance of a given country and
that “expected in a world in which world exports of the given sector are
distributed among countries in proportion to their share of world exports”
(Bowen, 1983, p.465).

As its name implies10 and as mentioned in Section 2, the BI is short of fully

representing its theoretical basis: it is likely that it does not fully deliver the
pattern of comparative advantage. Due to the unavailability of autarkic
variables and given the existence of numerous non-price factors that affect
comparative advantage, Balassa (1965) decided instead to use the consequences

8 He originally suggested two indices: one using relative export performance, and the other using
export-import ratio. Later he placed exclusive reliance on the former for the latter being affected
by tariffs and other protective measures (Balassa, 1977).
9 Balassa (1965) originally used the term „commodity.‟ Although, in empirical researches,

depends on the level of aggregation of data used in the analysis: „sector‟ or „industry‟ is more
frequently used since it is rarely the case where denotes a single commodity.
10 Balassa named the index „revealed comparative advantage‟ index, instead of just „comparative

advantage‟ index.
12
of trade flows that we can actually observe. Quoting Balassa himself:

“This [taking into account all influences that determine comparative advantage]
would be a rather laborious exercise and, in view of the difficulties of assigning
numerical values to these variables, it might bring disappointing results. Instead,
for purposes of indicating the possible consequences of trade liberalization, it
appears sufficient to provide information on “revealed” comparative advantage.”
(Balassa, 1965, p.103)

Nonetheless, the BI survived among a variety of trade performance indices

suggested so far: probably because of its simplicity of calculation and
convenience in dealing with data in empirical researches11, although there must
be many other reasons to use it (See Sanidas, 2007, 2009). Balassa originally
used the index in analyzing the manufacturing industries12; however, it has been
since then also used for other variety of industries, e.g. agriculture (Bojnec, 2001),
travelling (Peterson, 1988), services (Hisanaga, 2008, Seyoum, 2007). Also, he
ranked the industries in order of the calculated BI values, and used the ranks in
cross-commodity, cross-country and over-time comparisons (Balassa, 1965, 1977).

Hinloopen and Van Marrewijk (2001) point out that it is not possible to
theoretically derive the distribution of the BI, and thus corresponding empirical
studies have been carried out in the literature, e.g. De Benedictic and Tamberi
(2001, 2004), Dalum et al. (1998) and Laursen (1998). Accordingly, the BI has
been under critique for its incomparability across time and space, which
originates from its asymmetry (the BI values falling on between zero and infinity
with the comparative-advantage-neutral point being 1), unstable mean across
time and space, and aggregation effect (the BI varying depending on the
aggregation level of countries or reference groups). That is, the BI does not give
any significant information other than whether the comparative advantage
exists or not (Yeats, 1985), unless its distributions across space and time
coincidently close to being identical.

3.1.2 Alternative Revealed Comparative Advantage Indices

Due to the inconsistency and alleged incomparability of BI, there have

been numerous attempts to overcome the shortcomings of BI. The alternatives
can be classified in three classes, referring to Ballance et al. (1987): trade-cum-
production indices that contain both of trade and production variables; exports-
only indices that contain only exports variables; and indices using a hypothetical
situation such as comparative-advantage-neutral point.

Hoen and Oosterhaven (2006) suggested four properties which the ideal
trade performance index preferably should have, which are basically based on

11This issue is more discussed in Section 3.1.2.

12Balassa (1965) limited the inquiry to manufactured goods since a large number of primary
goods are subjected to subsidies, quotas and so on.
13
the shortcomings of BI that need to be overcome: a) a stable mean or median
across time and space; b) symmetry around the mean or median; c) independence
of classifications; and d) stable distribution across time and space. They did not
specifically mention the reasons why they suggested these four properties;
however it can be inferred by the literature that (a) is to meet the notion of
comparative advantage, the notion of zero-sum game embedded in the concept of
comparative advnatage; (b) is to consider the same weights on both sides of the
comparative-advantage-neutral point, especially when it comes to cardinally
using the RCA indices (De Benedictis and Tamberi, 2001, Laursen, 1998); (c) is
to obtain consistency over different levels of aggregation (Hoen and Oosterhaven,
2006, Yu et al., 2009); and (d) is to obtain the comparatbility over time and space.

(1) Trade-cum-production Indices

In a theoretical sense, trade-cum-production indices would be more

adequate and powerful measures in estimating comparative advantage, and thus
there have been many trade-cum-production indices suggested (See Ballance et
al., 1987, Memedovic, 1994, Vollrath, 1991). Bowen (1983) suggests an
alternative index including production variables, claiming that the BI index is
partly a “failure of the theoretical framework” since the BI separates exports and
imports when comparative advantage is properly a net trade concept. Also, Lafay
(1992) points out that the BI method eliminates the influence of macroeconomic
variables, and that the reason why Balassa chose only export variables and
excluded import variables can be applied in the same token to import variables.
That is, while tariffs and other protective measures on the import side are
accused of generating bias in trade performance measure, which has been
reduced through successive multilateral negotiations, the same argument can be
put on the subsidies or voluntary export restraint on the export side, which in
contrast has been increased (Lafay, 1992).

Furthermore, as Balassa (1965), Ballance et al. (1987) and Dowling and

Cheang (2000) point out, the major disadvantage of those indices is that trade
data and production data are usually collected at different point of time, using
different classifications and definitions13, which can cause unreliable inferences
in analyses. Concerning the limited information we have, the trade-only indices
provide a relatively more accurate indication of comparative advantage than
those of trade-cum-production in empirical tests do (Dowling and Cheang, 2000).

13 For example, trade data are collected by Standard International Trade Classification (SITC) or
Harmonized System (HS), while production data by International Standard Industrial
Classification (ISIC), Central Product Classification (CPC), etc. The concordance between the two
different classifications demand a very laborious exercise.
14
Lafay Index (henceforth LI)

However, Lafay (1992) avoided this problem by using only the total GDP,
which thus is not affected the problem of classification. Using to denote
country ‟s GDP, and imports, where , Lafay index (henceforth LI)
is:

Expressing it in a more condensed form,

Country is considered to have a comparative advantage (disadvantage) in a

given commodity when the balance in relation to GDP exceeds (is less than) the
attributed balance, i.e. exceeds (is less than) zero. The comparative
advantage neutral point is thus when the net exports marks zero, i.e. .

The algebraic sum of a country‟s LI scores is always zero, as shown in (10).

This implies that the LI‟s distribution of a country has the invariant mean value
over time, which puts more reliability on the over-time comparison of sectors
within a country.

As Alessandrini et al. (2007) point out, the LI can capture intra-industry

flows by using both of exports and imports variables, controlling the distortions
due to the macroeconomic factors with the GDP variable.14 It is also noteworthy
that the LI does not take into account world variables, which is a big difference
from other RCA indices introduced below.

(2) Export-only Indices

While trade-cum-production indices have more focused on how to better

link themselves with theory, attempts of exports-only indices have been more of
transforming and adjusting the existing BI index to overcome its disadvantages,
especially its asymmetric property, keeping at the same time its practical
easiness and simplicity.

14
Probably for this reason, ITC (International Trade Centre) adopted the LI index as an
indicator of trade specialization, together with the BI.
15
‘Symmetric’ Revealed Comparative Advantage Index (henceforth SI)

Vollrath (1991) made a suggestion of taking a log transformation of the BI

index as a solution to the asymmetry problem, 15 but it also added another
problem in that the BI could not be defined in the case of zero export, i.e. when
. However, another symmetrical index without the zero-exports issue was
provided by Dalum et al. (1998), which we call here symmetrical revealed
comparative advantage (SI):

The SI is an approximation of a log transformation of the BI (De Benedictis and

Tamberi, 2001), which ranges from -1 to +1 and is equal to 0 at the comparative-
advantage-neutral point, where . This index therefore is symmetric.

Being a simple transform of BI, the distribution of SI thus does not have
stable mean over space and time, which gives suspicion on its comparability.
However, it is considered that it has an advantage especially when econometric
analysis is taken place: Laursen (1998) shows that the distribution of SI meets
the normality assumption more frequently than that of BI. However, Benedictis
and Tamberi (2001) argue that reducing asymmetry16 of BI does not necessarily
imply obtaining normality.

‘Weighted’ Revealed Comparative Advantage Index (henceforth WI)

Proudman and Redding (1998) made an attempt to fix the mean of BI by

normalizing BI with its cross-section mean. Using to denote the number of
sectors,

Accordingly, the mean value of WI is constant and equal to one, which

enables the over-time comparison within a country. However, it does not solve
the problem of asymmetry: the WI ranges from zero to infinity. Added to that, as
Benedictis and Tamberi (2001) point out, the comparative-advantage-neutral
point is not generally determined and sensitive to the level of sectoral
aggregation.

15 That is, .
16 It should be noted that the symmetry used here indicates that the range where the RCA index
falls on is symmetric with respect to the comparative-advantage-neutral point, which is different
from the symmetry in distribution. For instance, the LI has a property of symmetry, but its
empirical distribution is right-skewed.
16
‘Additive’ Revealed Comparative Advantage Index (henceforth AI)

Hoen and Oosterhaven (2006) suggested an additive form of BI, pointing

out that the multiplicative form of BI causes the problematic properties. The
corresponding formula is the following:

The values of AI fall in between -1 and +1, with the comparative-

advantage-neutral point being zero, and thus the AI‟s distribution is symmetric.
Although its empirical distribution is bell-shaped and the mean value is centered
and constant, as shown in (14a), it does not follow normality due to far too high
kurtosis (Hoen and Oosterhaven, 2006). Yet, the AI has a big advantage on that
it is not affected by the level of sectoral aggregation, as shown in (14b).

Based on this additive property, Hoen and Oosterhaven (2006) further

suggest a regional specialization index in order to measure the degree of
inter/intra-industry trade.

While the AI has many advantages in cross-sector analysis, the

comparability in cross-country analysis is in doubt: the sum of AI values with
respect to a given sector is not stable, thus leads to variant mean value.

(3) Indices Using Hypothetical Situation

Some researchers expressed RCA using a hypothetical state: they used a

deviation of the actual data from the value that would have been in the
comparative-advantage-neutral (henceforth, CAN) point. Bowen (1983)
constructed an RCA index as a deviation between the actual and expected values
of trade, production, and consumption. The latter is based on a hypothetical
world of identical relative autarky prices across countries, thus a CAN world,
which is “one possible norm against which a country‟s actual trade could be
compared (Bowen, 1985)17” However, his index still has a problem of concordance
between production data and trade data, as discussed earlier with respect to the
trade-cum-production indices.

17 Re-quoted from Memedovic (1994, p.70).

17
 A recent attempt that brings in the hypothetical situation was done by Yu
et al. (2009), which is called the normalized revealed comparative advantage
(henceforth NI) shown in (15a) and (15b). The CAN point is based on the concept
of BI: the exports at the CAN point, , is derived from .
They normalized by the total world exports the deviation of the actual exports
with respect to the exports that would have been in the CAN point, as expressed
in (15a) and hence (15b).

The NI values range from -0.25 to 0.2518, with the CAN point being zero
when the actual exports is same as expected in the CAN world. Since it is
normalized by the size of world total exports, which is a very big number
compared to a country‟s exports on a sector, the numeric values of NI is usually
very small. Yu et al. (2009) therefore recommended to scale them by 10,000.

One big advantage of the NI is that the comparability across space and
time is well established: the sum of NI is stable and equal to zero across space
and time, hence so does the mean value. This explains well the notion of zero-
sum imbedded in comparative advantage: if a country gains comparative
advantage in one sector, then the country loses comparative advantage in other
sectors; and if one country gains comparative advantage in a sector, then other
countries lose comparative advantage in the sector.

Another advantage is that the NI is neither affected by the level of

sectoral aggregation nor by the choice of reference group of countries, while the
AI has this property only with the latter. This is shown mathematically in the
following:

where C and S denote a reference group of counties and a group of sectors,

respectively.

Although the NI obtains the symmetrical property, it still does not follow
the normal distribution. As will be seen in Section 4, its empirical distribution is

18 See Yu et al. (2009) for the proof.

18
very right-skewed and very leptokurtic.

3.1.3 Summary of the Six RCA Indices

Although the five indices introduced above overcome the shortcomings of

BI to some extent, yet none of them can be called „the perfect one.‟ Each one of
them has pros and cons, thus it should be important to well acknowledge the
properties of the indices and properly use them. The summary of the six indices
introduced above is presented in Tables 1 to 3.

Table 1 Formulae of the Six RCA Indices

Index Formula

BI

LI

SI

WI

AI

NI

Although, according to Tables 2 and 3, the NI seems to have more

favorable features as an RCA index than the others do. Especially, its stable
mean across space and time, and independence from aggregation level can be
very helpful in analyzing trade specialization.

Table 2 Statistical Properties of the Six Indices

BI LI SI WI AI NI
Comparative advantage
1 0 0 1 0 0
neutral point
Sum over sectors ∙ 0 ∙ ∙ 0 0
Sum over countries ∙ ∙ ∙ ∙ ∙ 0
Independence from aggregation
X X X X √ √
level
Independence from reference
X X X X X √
group of countries
Symmetry X √ √ X √ √
Normality X X X X X X

19
Table 3 Comparability of Six RCA Indices
Comparability BI LI SI WI AI NI
Cross-sector x ? x x(?) x √
Cross-country x ? x √(?) √ √
Over-time x √ x x(?) ? √

3. 2. Analyzing Structural Changes in Trade Specialization

Other than using RCA indices in non-econometric comparative analyses,

there have been studies that challenged the use of RCA indices in econometric
analysis. Hart and Praise (1956) first adopted „Galtonian regression,‟ which
originated Galton (1889)‟s methodology in examining the heights of fathers and
sons, to analyze the size-wise business concentration. This Galtonian regression
method has been also employed in a variety of areas, e.g. in examining
technological specialization patterns (Cantwell, 1989), changes in income
distribution (Hart, 1976), and convergence of productivity levels over time (Hart,
1995), as well as structural changes in trade specialization patterns using RCA
indices (e.g. Frantzen, 2008, Laursen, 1998, Sharma and Dietrich, 2007, Wörz,
2005).

The corresponding regression equation for a given country to test the

changes of trade specialization pattern, presuming that „nothing can be said on
these grounds about the determinants of the initial export specialization pattern‟
(Dalum et al., 1998), is:

where and indicates the earlier point of time and the later point of time,
respectively; and , standard regression coefficients; and , an error term.
It is assumed that the regression is linear, and that follows the normal
distribution and independent of . It is noteworthy that this method is
comparing two cross-sections at two different points of time: there is no element
of continuous time involved (Dalum et al., 1998).

When conducting the above regression method, using the BI is

problematic since it may not yield reliable estimates due to its asymmetric
property that leads to violating the normality assumption of error terms and
hence to invalid t-statistics. Cantwell (1989), who employed an index for
technological specialization, similarly built as BI 19 , dealt with this problem

19
He built an index called ‘revealed technological advantage (RTA)’, using the number of
technology patents.
20
practically by calculating the skewness and kurtosis for each distribution (Ibid.
p.31). In more recent literature, the SI was more frequently used (e.g. Dalum et
al., 1998, Laursen, 1998, Sharma and Dietrich, 2007) because of its symmetric
property, even though it does not always meet the normality assumption on error
terms. Laursen (1998) carried out Jarque-Bera test for normality of error term
and found out that normality assumption is met more frequently when using the
SI than using the unadjusted BI.

The underlying idea of Galtonian regression is basically to see how

similar or dissimilar the distributions of revealed comparative advantages at two
different points of time are to each other, hence the stability, or
convergence/divergence of the trade specialization patterns. To interpret the
corresponding results, Hart (1995) used the ratio of variances in two points of
time to measure the convergence of the labor productivity, as presented in (18),
where and are the regression coefficient and the coefficient of
determination between and , respectively. One advantage of examining (18)
is that the overall changes can be decomposed into regression effect, expressed
by , and mobility effect, expressed by .

20

If , there should be divergence of trade specialization since does

not exceed unity, that is, the pattern of trade specialization is considered to be
strengthened because industries with initial comparative advantage became
more advantaged, while those with initial comparative disadvantages became
more disadvantaged. If , it can be seen that there is no convergence or
divergence, thus the trade specialization pattern remained more or less stable.
However, if , it does not necessarily implies convergence. The convergence
happens when , where the pattern of trade specialization can be
considered to be weakened, that is, sectors with initial comparative advantage
improve their positions, while those with initial comparative advantage slip back.
However, there is another possible case of which implies divergence of
trade specialization pattern. Thus, when the case is , it is necessary to
avoid deriving an inference relying on only . In addition, there is another
possible case where : the sign of the index has been changed and hence the
ranking of sectors is reversed. However, this case is expected to rarely happen in
real world intuitively considering the path dependency of technology or trade
specialization (Dalum et al., 1998).

Therefore, the t-test of the null hypothesis can test the path
dependency of trade specialization against the randomly determined trade
specialization. Additionally, testing the null hypothesis can roughly test
convergence/divergence of the pattern of trade specialization. As explained above,

20 See Cantwell (1989) or Hart (1995) for the derivation of the formula.
21
if is significantly greater than 1, then it can be considered that there has been
a divergence. However, if is significantly less than 1, then we also need to
examine to determine convergence or divergence.

Furthermore, when comparing coefficients, we should also consider

whether or not the intercepts are different. Different intercepts can confuse us
when deriving inferences since the changes in intercepts affect those in slopes.
For this reason, Hart (1995) preferred the deviation form since the constant term,
, in (17) depends on units of measurement. In this study, thus, we suggest
further way of interpreting the regression results in Section 4.

Since this Galtonian regression is essentially the simple OLS regression,

the normally distributed error terms are assumed, which does not always the
case with respect to our RCA indices with non-normality and influential outliers.
This issue also will be discussed more with empirical example in Section 4,
where different methods that can resolve the normality issue will be suggested.

4. EMPIRICAL EXAMINATION: THE CASE OF EAST

ASIAN ECONOMIES

4.1. Data Description

Exports and imports data, the basic variables for calculating the
corresponding RCA indices, are retrieved from three PC-TAS (Personal
Computer Trade Analysis System) CD-ROMs (2000, 2003, 2008) published by
International Trade Centre (ITC). These CD-ROMs cover the periods of 1995-
1999, 1998-2002, and 2003-2007, respectively, and values for 2008 are obtained
from the website of ITC21. These trade data are based on the Harmonized System
(HS) 2-digit level of aggregation, which consists of 98 sectors (from HS 01 to HS
99 with HS 77 being empty as “reserved for possible future use.”). See Appendix
1 for the description of HS 2-digit codes. GDPs used here for the calculation of
Lafay index are obtained from World Economic Outlook (WEO) database
available on the website of International Monetary Fund (IMF)22.

The calculations of the six RCA indices, introduced in Section 3, for China,
Japan and South Korea are carried out by the author according to the formulae
presented in Table 1, except that we scaled up the NI by 10,000 as recommended
by Yu et al. (2009). Also, more South East Asian countries, namely, Hong Kong,
Indonesia, Malaysia, Philippines, Singapore and Thailand, are added in cross-
country analysis in order to conduct more appropriate comparison. In addition,

21 www.intracen.org
22 www.imf.org
22
for all sectors of the 3 countries, Hillman‟s condition is examined: no violation
was found, which implies that the calculated values of BI, and also the
transforms of the BI, i.e. the SI and WI, can adequately function in cross-country
analyses. The calculated values and cross-sector rankings of the six RCA indices
for the three countries are presented in Appendix 2. With regard to the
regression analysis, we used the statistical package STATA 10.0.

4.2. Comparison of RCA Indices in Measuring Comparative

Advantage

Here, we compare different RCA indices taking an example of East Asian

economies, namely, China, Japan and South Korea. The comparison is three-fold:
in terms of dichotomous assessment (whether the comparative advantage exists
or not); in terms of ordinal measurement (cross-sector rankings and cross-
country rankings); and in terms of cardinal measurement (degree of comparative
advantage).

4.2.1. RCA Indices as Dichotomous Assessment

Since the SI and AI are transforms of BI, and the NI adopted the basic
notion of BI, the SI, AI and NI lead to exactly the same results as those of BI in
determining whether or not a country has a comparative advantage in a given
sector with respect to their dichotomous thresholds. That is, the same sectors
show comparative advantage or disadvantage according to BI, SI, AI and NI.
However, LI and WI sometimes lead to results that are very contradictory to that
of the other four indices. See Appendix 2, where the sectors with comparative
advantage are tinted; those with comparative disadvantage are not. For example,
as shown in Table A1 in Appendix 2, South Korea has an evident comparative
disadvantage in HS 72 (iron and steel) according to LI method, while the other
indices tell us it does not. More examples of the “contradictions” between the six
indices can be found in Table A1. Although, the results for BI, SI, AI and NI
match 100% with each other in terms of dichotomous assessment according to
Table A3 in Appendix 2.

Furthermore, we can also examine how many sectors are considered to

have a comparative advantage according to each RCA index, which is shown in
Table A2 in Appendix 2. The same number of sectors is identified as having a
comparative advantage by the BI, SI, AI and NI, while the LI and WI show
different numbers, as expected. One interesting finding here is that all the RCA
indices except for the WI tell us that China has the most sectors with
comparative advantage among the three countries, and Japan has the least,
which possibly implies that Japan, the most economically developed country out
of the three, has specialized the most out of the three in terms of exports
performance, followed by South Korea and then China.

23
 To sum up, when using RCA indices as a dichotomous measure, i.e. when
the matter is whether a given country has a comparative advantage or
disadvantage in a given sector, making a choice of index among BI, SI, AI and NI
does not affect the inference since those four yields identical results. However,
one needs to be careful when using LI or WI due to their different content
included in the formulae. Additionally, as already mentioned, one should note
that the data used in this analysis are based on the 2-digit level of aggregation,
that is, the results can be different when less aggregated dataset are taken into
account.

4.2.2. RCA Indices as Ordinal Measures

In Cross-sector Analysis

In empirical researches, one of the most common features of RCA indices

is using ranking as an indicator as to roughly measuring the comparative
advantage. One way to do it is to rank sectors in order of calculated RCA values
with respect to a given country, basically to find out how important or promising
the given sector is in the given country.

Table A4 in Appendix 2 shows the cross-sector rankings of China, Japan

and South Korea for 2008. The first noticeable feature here is that different
indices yield very different results, which of course puts empirical researchers in
a difficult position as to tell whether the sectors of their interest are strong or
weak. For example, the AI and NI tell us that the strongest sector in China is HS
85 (electrical, electronic equipment), which ranked only 28th by the BI, SI and WI
and 88th by the LI. Other examples of difference between the indices can be
easily found in Japan and South Korea as well.

The pair-wise rank correlations between the six indices, averaged over
the three countries for 2008, are displayed in Table A5. As implied above, those
six indices can be classified into three groups: BI, SI and WI (Group 1); AI and
NI (Group 2); and LI (Group 3). Within each group, the rankings made by each
index are 100% correlated, i.e. choosing an index within each of these groups will
not make any difference when making cross-sector rankings in a given country.
This is because SI and WI are transformations of BI. The SI approximates the
logarithm of BI, which is a monotonic transformation of BI and thus results in no
change in ranks. The WI also is built by adding a scaling onto BI, as De
Benedictis and Tamberi (2001) point out. One noteworthy aspect of Table A5 is
that the six indices are not significantly independent, i.e. the null hypothesis of
pair-wise independence is not rejected.

Accordingly, choosing between each group will not make any difference in
conducting inference based on the rankings of RCA. However, as seen in Tables
24
A4 and A5, depending on different indices, one can derive varying inferences.
Nonetheless, it is not yet the right time to tell which one is the best performing
index: further researches will have to be conducted. At this point, one needs to be
careful and strategic in choosing an index when analyzing trade specialization,
and cannot be depending on only one index.

In Cross-country Analysis

The other way is to rank countries in order of calculated RCA values with
respect to a given sector, which is to find out how well a given country is
performing in the international (in relation to any reference group of countries
taken account in the analysis) export market for the given sector.

For convenience of analysis, we limit the reference group to Asian

countries, where more South East Asian countries (namely, Hong Kong,
Indonesia, Malaysia, Philippines, Singapore and Thailand) are added to our
three East Asian countries. Cross-country rankings of the nine Asian countries
are partly shown in Table A623. As seen, individual countries are ranked within
the reference group. It is noteworthy that the LI is not affected by the problem of
reference group since it does not take into account any reference (or world)
aggregate variables; and neither does the NI due to its independence from
aggregation effect.

In cross-country comparison, BI, SI and AI (Group 1) yield the same

results since SI is a monotonic transformation of BI, as discussed in cross-sector
comparison, and AI becomes to only a scaled form of BI given that the
denominator of BI and the negative part of AI are constant and equal with
respect to all countries. However, the LI, WI, NI do not always agree with the
Group 1. For example, WI tells us that South Korea has the strongest
comparative advantage in HS 76 (Aluminum and articles thereof), while the
other four say China. As for HS 85 (electrical, electronic equipment), the Group 1,
LI, WI and NI flag different countries as the number one: Philippines, Singapore,
South Korea, and Hong Kong, respectively.

The Spearman rank correlation coefficients between indices are presented

in Table A7. As expected, the indices in the Group 1 are 100% correlated to each
other. The noticeable aspect of Table 10 is that the LI and the WI are the most
different indices from the rest: especially, the LI is correlated with the others by
less than 50%, particularly only 27% with the WI. In addition, the LI is
significantly independent from the other five, and so is the WI. That is, one
needs to be more careful in choosing an RCA index in cross-country analyses.

23 The whole results are not presented in this paper due to the space limitation.
25
4.2.3 RCA Indices as Cardinal Measures: Over-time Analysis

There are several ways of cardinally using RCA indices. They can be used
to examine how much more of comparative advantage a given sector has in
comparison to other sectors; or how much more of comparative advantage a given
country has in comparison to other countries; or how much more of comparative
advantage a given sector of a given country gained through time. Here, we
provide an example of the third way, which is usually the interest of empirical
researchers; rankings are more frequently used in the cross-sector and cross-
country comparisons, where, according to De Benedictis and Tamberi (2004),
using rankings has more benefits than simply using numeric values.
Additionally, we chose three indices, namely, the BI, LI and NI, which are
thought to be enough to cover the relevant issues.

As discussed in Section 3, the NI and LI are the only indices among the
six that have comparability over time. The sum of all of the variations is zero for
LI and NI, while those of the other indices are not. Table A8 below shows
variations of the BI, LI and NI values between 1995 and 2008 in case of South
Korea. The sum of variations when using the BI is -38, whereas those of the LI
and NI are zero, as expected. The two indices do not all the time agree with each
other in whether or the comparative advantage is gain during 1995 and 2008.

We need to note that all the differences across indices so far are possibly
due to the aggregation level of the data used here. As more detailed level of
aggregation is used, e.g. 4-digit or 6-digit of HS classification, the situation
would be expected to be different since the higher level of aggregation is adopted,
the bigger role the intra-industry trade plays in international trade. More
precisely, the larger the number of commodities covered by a single
nomenclature, the more likely it is that there are flows in both directions. (Lafay,
1992)

4.3. Comparison of RCA Indices in Econometric Analysis

4.3.1. Preliminary Data Analysis

The Galtonian method is the linear simple regression with respect to the
two cross-sections of two different time periods. Rewriting the corresponding
model, we have:

where the normality is assumed on the error terms. Thus, we first investigate
whether or not the normality assumption is met. We use the average over 1995-

26
1997 for , and the average over 2006-2008 for . 24 First, we investigate
whether or not the normality assumption is met. The Central Limit Theorem
says that estimates of any random variable approach normal as the sample size
increases, and it is generally known that the large enough sample size is 30 or
more. Yet, we still examine the normality issue in this study following previous
researches (e.g. Dalum et al., 1998, Laursen, 1998) which took this issue into
account.

To have a preliminary idea on the distributions of the six indices, we

examine the scatter plots and histograms of each index of both time periods for
each country, shown in Appendix 4. The reference line indicates the threshold,
i.e. comparative-advantage-neutral point.

Through the scatter plots, we can roughly see the linear relationship
between the two points of time and also the presence of outliers25. Taking a
glance of histograms, we can infer that all relevant data do not look normal. The
distributions of the BI and WI seem to be far from being symmetric: they are
very right-skewed. Those of the LI, AI and NI seem to possibly symmetric and
bell-shaped without the outliers; however, they are very concentrated around the
threshold point, which leads to a high kurtosis. The distribution of the SI seems
to have the most spread-out distribution. Another noteworthy aspect is that, in
the distributions of all indices except for the SI, we can easily witness the
existence of outliers, which we actually detected by using Cook‟s D26 (Cook, 1977)
(the results are not shown here). The existence of outliers puts us in a difficult
position since we cannot simply ignore them because the extreme values in trade
specialization are usually within our interest.

Table 4 is the descriptive statistics of the RCA indices averaged over

2006-2008 in the case of South Korea. As discussed in Section 2, we can
reconfirm the zero mean of LI, AI and NI. For all of the indices, the mean is
greater than the median, which implies the right-skewness. All of the indices
except for the SI seem to be very far from the normal distribution looking at the
summary statistics, especially in terms of kurtosis. As mentioned above in
Section 3, the symmetry as a property of an index has a different implication
from the symmetry of the index‟s empirical distribution: for example, the NI has
the symmetric property, implying that the upper bound and the lower bound are
symmetrical with respect to the CAN point. The distributions for the other time
period, the average over 1995-1997, and for the other two countries are more or
less similar (not shown here).

24 We took the average into account here to control the effect from short-term fluctuation.
25 Not shown here due to space limitation.
26 An observation with Cook‟s D greater than N/4 is considered to be an influential outlier. N

denotes the sample size.

27
Table 4 Descriptive Statistics of RCA Indices (averaged over 2006-2008,
South Korea)
BI LI SI WI AI NI
mean 0.64 0.00 -0.44 0.99 0.00 0.00
median 0.35 -0.21 -0.48 0.55 0.00 -0.43
s.d. 0.95 8.48 0.43 1.47 0.02 5.07
skewness 4.04 -1.46 0.62 3.98 5.27 5.21
kurtosis 22.57 29.89 -0.52 21.98 43.90 42.77
max. 7.04 40.19 0.71 10.89 0.15 40.31
min. 0.00 -58.04 -1.00 0.00 -0.06 -15.96
CAN point 1 0 0 1 0 0

upper bound. ∞. ∞ -1.00 ∞ 1.00 2500.00

Lower bound 0.00 -∞ 1.00 0.00 -1.00 -2500.00
Note: The skewness and kurtosis are 0 and 3, respectively, in the normal distribution.; „CAN
point‟ denotes comparative advantage neutral point.

4.3.2 OLS Galtonian Regression and Normality Issue

As a first step, we carry out the OLS Galtonian regressions. Then we

examine the corresponding residuals by some normality tests such as Shapiro-
Wilk W test (henceforth S-W test), Shapiro-Francia test (henceforth S-F test),
and Skewness-Kurtosis test (henceforth S-K test). The results are shown in
Table 5.

Table 5 Normality Tests for OLS Residuals

Country Index S-K test S-F test S-W test
chi(2) z z
BI 7.39 ** 3.02 *** 3.2 ***
LI 46.90 *** 6.76 *** 7.76 ***
SI 1.90 0.51 0.63
China
WI 7.63 ** 3.10 *** 3.29 ***
AI 45.79 *** 7.32 *** 8.54 ***
NI 45.87 *** 7.33 *** 8.55 ***
BI 49.19 *** 5.97 *** 6.70 ***
LI . *** 7.11 *** 8.29 ***
SI 10.29 *** 3.74 *** 3.8 ***
Japan
WI 49.44 *** 6.00 *** 6.74 ***
AI 51.32 *** 7.25 *** 8.47 ***
NI 48.86 *** 7.26 *** 8.46 ***
BI 55.64 *** 6.45 *** 7.46 ***
LI 55.17 *** 6.68 *** 7.68 ***
SI 1.62 1.35 * 1.05
S.Korea
WI 55.42 *** 5.74 *** 6.38 ***
AI 35.79 *** 6.55 *** 7.51 ***
NI 36.95 *** 6.55 *** 7.52 ***
Note: * (significant at 10% level) ** (significant at 5% level) *** (significant at 1% level)

The null hypothesis of normality is accepted only in the cases of the SI for
China and South Korea: an insignificant result implies that the distribution
follows normality, which implies the validity of t-statistics. In terms of using
28
RCA indices in econometric analysis, the SI is considered so far to be the most
adequate alternative of BI with respect to normality issue (Dalum et al., 1998).
The relevant regression results are shown in Table 6.27

Table 6 Results of OLS Galtonian Regression

China Japan South Korea
ß 0.40 (20.39)*** 1.02 (22.83)*** 0.47 (10.92)***
BI R 0.190 0.150 0.440
α 0.479 (5.81)*** 0.06(1.63) 0.016 (2.1)**
ß 0.24 (1.6)*** 2.06 (22.46)*** 1.35 (18.43)***
LI R 0.970 0.160 0.220
α 0.0002 (0.00)*** 0.00 (0.00) 0.00 (0.00)
ß 0.82 (16.47)*** 0.98 (33.97)*** 0.65 (11.28)***
SI R 0.260 0.080 0.430
α -0.11 (-4.05)*** 0.04(1.97)* -0.023 (-6.77)***
ß 0.66 (20.58)*** 0.89 (22.99)*** 0.73 (11.03)***
WI R 0.180 0.150 0.440
α 0.33 (5.78)*** 0.105(1.64) 0.256 (2.14)**
ß 0.36 (3.34)*** 0.97 (17.21)*** 0.87 (15.14)***
AI R 0.900 0.240 0.300
α 0.00 (0.00) 0.00(0.03) 0.00(0.00)
ß 0.96 (3.37)*** 0.57 (17.10)*** 0.85 (15.09)***
NI R 0.890 0.250 0.300
α 0.003 (0.00) 0.00(0.00) 0.00(0.00)
Note: * (significant at 10% level) ** (significant at 5% level) *** (significant at 1% level); the t-
statistics are shown in parentheses; the regressions that yield a reliable t-statistics due to the
non-normality of residuals are tinted; See the Section 4.3.3 for the discussion on these results.

To conclude so far, there are not many cases where the OLS Galtonian
regression can be effectively used due to the normality issue. Therefore, we apply
the log transformation and the Box-Cox transformation (Box and Cox, 1964)28
that allow us to decrease the skewness of the distributions of the indices, re-run
the OLS regression29, and then conduct the normality tests on the corresponding
residuals again. The selected results are shown in Table 7, where it turned out
that only the BI, SI and WI for China and South Korea could obtain the
normality of residuals through the transformations. The other cases still showed
high kurtosis after the transformations.

27 These results are also presented in Table 14 for convenience of comparison with the other
results from the robust regression and the quantile regression, and will be discussed later.
28 The log transformation and the Box-Cox transformation are , ,
respectively. These processes are basically to find the value of and that make the skewness
close to zero.
29 The results of the OLS regression after the transformations of indices are presented in Table

14 for convenience of comparison with the other results from different methods.
29
Table 7 Normality Tests for the Residuals after Transformations
S-K test S-F test S-W test
chi(2) z z
ln BI 2.46 0.93 1.12
ln SI 2.05 0.57 0.64
ln WI 2.78 1.08 1.27
China
bc BI 2.01 0.67 0.81
bc SI 2.05 0.565 0.63
bc WI 1.83 0.6 0.69
ln BI 0.4 -0.42 -0.49
ln SI 0.72 0.125 -0.016
ln WI 0.29 -0.59 -0.69
S.Korea
bc BI 1.08 0.05 -0.03
Bc SI 0.77 0.122 -0.012
bc WI 0.96 -0.02 -0.13
Note: „ln‟ denotes the log transformation of the index; „bc‟ denotes the Box-Cox transformation of
the index; with respect to Japan, the null hypothesis of normality was rejected in all cases thus
not shown here.

4.3.3 Robust and Quantile Galtonian Regressions

As shown above, using the transformation techniques could only save the
three out of the six indices for only two countries among three. Thus, we here
suggest trying different methods instead of the OLS regression. The first one is
the robust regression. Regression outliers pose a serious threat to standard least
squares analysis, and the robust regression tries to devise estimators that are
not so strongly affected by outliers (Rousseeuw, 1987). As can be seen in the
histograms presented in Appendix 4 and also as detected by Cook‟s D, the most
noticeable characteristic of the RCA indices‟ distributions, except for the SI‟s, is
the existence of outliers, which possibly results in non-normality and unreliable
regression results. Yet, there is one disadvantage in using the robust regression
in our analysis: the way the robust regression deals with outliers is essentially
down-weighting the extreme values, which we may not want because those
values, indicating sectors with extremely high or low performance, are important
in analyzing trade performance. Also, the R-squared, which is one of the
important features in Galtonian regression, may not be a proper indicator for the
interpretation (Street et al., 1988).

The second method we suggest is the quantile regression, which estimates

the median or other different quantiles. One of the advantages of using the
quantile regression is that it out-performs the least-square estimator over a wide
class of non-normal error distributions by minimizing the sum of absolute
residuals (Koenker and Basset, 1978). De Benedictis and Tamberi (2004, p.334)
argue the usefulness of using the median instead of the mean: quoting him,

“The median comes out as a useful synthetic locational index, in particular

because it is not influenced by extreme values, a typical characteristic of the BI
distribution. Moreover, consider that, because the BI is a relative index, if some
sectors inside a country have a BI above 1, other sectors (at least one) must

30
 necessarily be below unity; as a consequence there is no real statistical meaning in
saying that the μ(BI) is above or below unity. On the contrary, the median of
sectoral BI has an immediate meaning: a low median means that a country has a
large share of sectors with comparative disadvantage; a high median means that a
country has a large share of sectors with comparative advantage.”

Accordingly, the robust regression and the quantile regression are conducted
on our dataset: each RCA index averaged over 2006-2008 is regressed on that
over 1995-1997 for China, Japan and South Korea. The relevant results are
shown in Tables 8 and 9. Table 9 contains the results from the OLS regression as
well for the convenience of comparison.

Firstly, we conducted t-tests on whether the regression coefficient is

significantly different from unity, and from the coefficient of determination, R,
which will tell us whether the pattern of trade specialization has strengthened
(See Table 10 for the relevant results). Following the way of interpretation
mentioned in Section 3.2, we face confusing results since different results are
yielded across indices as well as regression methods, even if we take into
consideration the fact that each index is constructed based on different idea and
purpose thus some difference should be acceptable to some extent. For example,
according to the results of OLS regression using the BI, we can infer that the
trade specialization pattern of China and South Korea have been weakened
while that of Japan strengthened; however, by using the NI, we get an inference
that the trade specialization pattern of China has been strengthened, that of
Japan weakened, and that of South Korea remained stable. At the same time,
with respect to the case of China using the AI, the OLS and quantile methods
tell us that the trade specialization pattern of China has remained stable, while
the robust regression method say it has been weakened; in case of South Korea,
the OLS shows that the trade specialization pattern of South Korea has been
stable, the robust regression says it has been weakened, and the quantile
regression method says strengthened. Examining other cases as well, we see that
different regression methods and different indices do not agree on whether the
countries' patterns of trade specialization has been strengthened or weakened.

31
Table 8 Results of Robust and Quantile Galtonian Regressions

Robust Regression Quantile (Median) Regression

China Japan S.Korea China Japan S.Korea
0.98 0.32 0.40 0.99
0.40 (20.17)*** 0.36 (16.67)***
(60.34)*** (11.99)*** (28.65)*** (66.83)***
BI R 0.901 0.987 0.779 0.509 0.682 0.309
0.026 0.154 0.31
0.38 (4.59)*** 0.016 (1.36) 0.09 (2.35)**
(2.22)** (4.01)*** (2.29)**
1.31 0.78 1.38
0.78 (25.42)*** 1.43 (172)*** 1.34 (141.8)***
(67.85)*** (40.78)*** (193.2)***
LI R 0.934 0.998 0.990 0.117 0.571 0.353
0.54 0.035 -0.16 -0.003 -0.05
0.025 (1.36)
(0.67) (2.87)*** (-2.01)** (-0.04) (-0.95)
0.71 0.91 1.02
0.85 (16.97)*** 1.02 (41.3)*** 0.72 (9.97)***
(12.46)*** (26.02)*** (41.79)***
SI R 0.866 0.973 0.786 0.558 0.770 0.397
-0.09 0.464 -0.21 -0.09 -0.23
0.037 (2.20)**
(-3.58)*** (2.76)*** (-6.19)*** (-4.73)*** (-5.43)***
0.49 0.64 0.87
0.65 (20.49)*** 0.85 (67.1)*** 0.57 (23.24)***
(12.13)*** (16.01)*** (76.98)***
W
R 0.902 0.990 0.781 0.512 0.683 0.311
I
0.27 0.023 0.22
0.041 (2.24)** 0.026 (1.39) 0.015 (3.44)***
(4.66)*** (3.96)*** (2.63)***
0.53 0.50 0.70
0.39 (28.65)*** 0.73 (43.6)*** 1.06 (32.37)***
(39.10)*** (61.73)*** (213.6)***
AI R 0.947 0.977 0.971 0.241 0.551 0.315
-0.0005 -0.0008 -0.0005 0.00 0.00
0.0001 (1.08)
(-2.66)*** (-5.73)*** (-3.86)*** (0.00) (0.00)
1.04 0.43 0.54 1.35 0.41 0.88
(28.92)*** (51.61)*** (44.93)*** (57.42)*** (299.7)*** (159.28)***
NI R 0.948 0.983 0.977 0.243 0.557 0.314
-0.46 -0.19 -0.47 -0.015
0.016 (0.34) 0.004 (0.16)
(-2.5)** (-5.46)*** (-3.39)*** (-0.59)
Note: * (significant at 10% level); ** (significant at 5% level); *** (significant at 1% level); the t-
statistics are shown in parentheses.

32
Table 9 Results of OLS / Robust / Quantile Regression for China,
Japan and South Korea

China

OLS after Log OLS after B-C

OLS Robust Reg. Quantile Reg.
Transformation Transformation

0.40 (20.39)*** 0.69 (16.93)*** 0.77 (17.21)*** 0.40 (20.17)*** 0.40 (28.65)***

BI R 0.9 0.750 0.870 0.901 0.71

-0.09 -0.2 0.38 0.31
0.479 (5.81)***
(-1.81)* (-3.27)*** (4.59)*** (2.29)**
0.24 (1.6)*** 0.78 (25.42)*** 0.78 (40.78)***

LI R 0.16 0.934 0.34

0.54 -0.003
0.0002 (0.00)***
(0.67) (-0.04)
0.82 (16.47)*** 1.76 (16.58)*** 0.89 (16.55)*** 0.85 (16.97)*** 0.91 (26.02)***

SI R 0.86 0.860 0.861 0.866 0.75

-0.11 -2.77 0.064 -0.09 -0.09
(-4.05)*** (-9.85)*** (0.98) (-3.58)*** (-4.73)***
0.66 (20.58)*** 0.67 (17.50)*** 0.75 (17.26)*** 0.65 (20.49)*** 0.64 (16.01)***

WI R 0.9 0.230 0.871 0.902 0.72

0.33 0.08 0.27 0.22
0.137 (2.65)***
(5.78)*** (1.30) (4.66)*** (2.63)***
0.36
0.39 (28.65)*** 0.50 (61.73)***
(3.34)***
AI R 0.32 0.947 0.49
0.00 -0.0005 -0.0005
(0.00) (-2.66)*** (-3.86)***
0.96 (3.37)*** 1.04 (28.92)*** 1.35 (57.42)***

NI R 0.33 0.948 0.49

0.003 -0.46 -0.47
(0.00) (-2.5)** (-3.39)***
Note: In the results of OLS, the case where the normality assumption on the error terms is met is
tinted; * (significant at 10% level); ** (significant at 5% level); *** (significant at 1% level); the t-
statistics are shown in parentheses; with respect to the OLS regression results after
transformation, only the cases where the error terms are significantly normal are presented.

33
Table 9 (cont.) Results of OLS / Robust / Quantile Regression
for China, Japan and South Korea

Japan
OLS Robust Quantile
1.02 0.98 0.99
(22.83)*** (60.34)*** (66.83)***
BI R 0.92 0.987 0.83
0.06 0.026 0.016
(1.63) (2.22)** (1.36)
2.06 1.43 1.38
(22.46)*** (172.34)*** (193.24)***
LI R 0.92 0.998 0.76
0.00 0.035 0.025
(0.00) (2.87)*** (1.36)
0.98 1.02 1.02
(33.97)*** (41.33)*** (41.79)***
SI R 0.96 0.973 0.88
0.04 0.464 0.037
(1.97)* (2.76)*** (2.20)**
0.89 0.85 0.87
(22.99)*** (67.10) (76.98)***
WI R 0.92 0.990 0.83
0.105 0.041 0.026
(1.64) (2.24)** (1.39)
0.97 0.73 0.70
(17.21)*** (43.57) (213.64)***
AI R 0.87 0.977 0.74
0.00 0.0001 0.00
(0.03) (1.08) (0.00)
0.57 0.43 0.41
(17.10)*** (51.61)*** (299.71)***
NI R 0.87 0.983 0.75
0.00 0.016 -0.015
(0.00) (0.34) (-0.59)
Note: * (significant at 10% level); ** (significant at 5% level); *** (significant at 1% level); the t-

statistics are shown in parentheses.

34
Table 9 (cont.) Results of OLS / Robust / Quantile Regression
for China, Japan and South Korea

South Korea
OLS after Log OLS after B-C
OLS Robust Quantile
Transformation Transformation
0.47 (10.92)*** 0.72 (12.23)*** 0.68 (12.36)*** 0.32 (11.99)*** 0.36 (16.67)***

BI R 0.75 0.782 0.785 0.779 0.309

0.016 -0.54 -0.54 0.09
0.154 (4.01)***
(2.1)** (-6.2)*** (-6.36)*** (2.35)**
1.35 (18.43)*** 1.31 (67.85)*** 1.34 (141.82)***

LI R 0.88 0.990 0.353

0.00 -0.16 -0.05
(0.00) (-2.01)** (-0.95)
0.65 (11.28)*** 0.975 (11.88)*** 0.59 (11.82)*** 0.71 (12.46)*** 0.72 (9.97)***

SI R 0.76 0.771 0.770 0.786 0.397

-0.023 -0.402 -0.21 -0.23
0.08 (3.01)***
(-6.77)*** (-10.53)*** (-6.19)*** (-5.43)***
0.73 (11.03)*** 0.72 (12.61)*** 0.73 (12.3)*** 0.49 (12.13)*** 0.57 (23.24)***

WI R 0.75 0.790 0.784 0.781 0.311

-0.083 -0.12
0.256 (2.14)** 0.023 (3.96)*** 0.015 (3.44)***
(-0.97) (-1.30)
0.87 (15.14)*** 0.53 (39.10)*** 1.06 (32.37)***

AI R 0.84 0.971 0.315

0.00 -0.0008 0.00
(0.00) (-5.73)*** (0.00)
0.85 (15.09)*** 0.54 (44.93)*** 0.88 (159.28)***

NI R 0.84 0.977 0.314

0.00 -0.19 0.004
(0.00) (-5.46)*** (0.16)
Note: In the results of OLS, the case where the normality assumption on the error terms is met is
tinted; * (significant at 10% level); ** (significant at 5% level); *** (significant at 1% level); the t-
statistics are shown in parentheses.

35
Table 10 Results of t-test on the regression coefficients

China
OLS Robust Reg. Quantile Reg.
Ho: b=1 897.7 *** b<1 888.8 *** b<1 441.5 *** b<1
BI convergence convergence convergence
Ho: b=R 621.3 *** b<R 615.7 *** b<R 118.3 *** b<R
Ho: b=1 25.09 *** b<1 52.0 *** b<1 137.2 *** b<1
LI stable convergence divergence
Ho: b=R 0.29 b=R 25.6 *** b<R 525.8 *** b>R
Ho: b=1 13.02 *** b<1 8.81 *** b<1 6.2 ** b<1
SI stable stable divergence
Ho: b=R 0.64 b=R 0.09 b=R 21.5 *** b>R
Ho: b=1 117.2 *** b<1 118.0 *** b<1 78.0 *** b<1
WI convergence convergence convergence
Ho: b=R 59.1 *** b<R 59.7 *** b<R 3.525 * b<R
Ho: b=1 35.5 *** b<1 2061.7 *** b<1 3811.1 *** b<1
AI stable convergence stable
Ho: b=R 0.13 b=R 1720.7 *** b<R 1.52 b=R
Ho: b=1 0.02 b=1 1.36 b=1 222.7 *** b>1
NI divergence divergence divergence
Ho: b=R 4.88 ** b>R 6.8 * b>R 1339.3 *** b>R

Japan
OLS Robust Reg. Quantile Reg.
Ho: b=1 0.21 b=1 0.99 b=1 0.38 b=1
BI divergence stable divergence
Ho: b=R 5.04 ** b>R 0.04 b=R 125.1 *** b>R
Ho: b=1 133.9 *** b>1 2727.0 *** b>1 2838.9 *** b>1
LI divergence divergence divergence
Ho: b=R 154.8 *** b>R 2752.1 *** b>R 7668.6 *** b>R
Ho: b=1 0.56 b=1 0.48 b=1 0.56 b=1
SI stable divergence divergence
Ho: b=R 0.41 b=R 3.2 * b>R 33.14 *** b>R
Ho: b=1 8.37 *** b<1 142.9 *** b<1 123.8 *** b<1
WI stable convergence divergence
Ho: b=R 0.77 b=R 124.6 *** b<R 17.7 *** b>R
Ho: b=1 0.26 b=1 251.4 *** b<1 8655.8 *** b<1
AI divergence convergence convergence
Ho: b=R 3.22 * b>R 210.0 *** b<R 193.6 *** b<R
Ho: b=1 162.4 *** b<1 4859.4 *** b<1 190000 *** b<1
NI convergence convergence convergence
Ho: b=R 78.5 *** b<R 4576.2 *** b<R 60781 *** b<R

South Korea
OLS Robust Reg. Quantile Reg.
Ho: b=1 149.9 *** b<1 653.5 *** b<1 486.1 *** b<1
BI convergence convergence divergence
Ho: b=R 40.4 *** b<R 298 *** b<R 76.1 *** b>R
Ho: b=1 23.2 *** b>1 261.7 *** b>1 1474.0 *** b>1
LI divergence divergence divergence
Ho: b=R 41.6 *** b>R 278.7 *** b>R 7131 *** b>R
Ho: b=1 35.9 *** b<1 25.5 *** b<1 15.4 *** b<1
SI convergence stable stable
Ho: b=R 3.1 * b<R 1.7 b=R 1.48 b=R
Ho: b=1 16.1 *** b<1 155.2 *** b<1 320.3 *** b<1
WI stable convergence stable
Ho: b=R 0.06 b=R 50 *** b<R 0.04 b=R
Ho: b=1 4.8 ** b<1 1227.3 *** b<1 0 b=1
AI stable convergence divergence
Ho: b=R 0.34 b=R 1081.3 *** b<R 201 *** b>R
Ho: b=1 6.8 ** b<1 1447.6 *** b<1 484.5 *** b<1
NI stable convergence divergence
Ho: b=R 0.05 b=R 1306 *** b<R 3335.8 *** b>R
Note: *** 1% significant; ** 5% significant; * 10% significant

36
 Additionally we also carried out a t-test30 for two on whether the regression
coefficients yielded by different regression methods are significantly different
from each other (at 5% significance level). The relevant results are shown in
Table 11. It is found that the two coefficients yielded by the OLS and the robust
regressions are significantly unequal, except for when using the BI and WI for
China; that the two coefficients yielded by the OLS and the quantile regression
are significantly unequal in all cases; and that the two coefficients yielded by the
robust and quantile regressions are significantly unequal except for when using
the BI and LI for China, the SI for Japan, and South Korea.

It seems that it is not the right time yet to say which regression method
or which index is superior to the others. However, here we cannot avoid casting
doubts on the regression coefficient of OLS method that are significantly
different from those by the other methods, considering the presence of influential
outliers and non-normality. Thus, one should be careful when making a decision
on which index and regression method to use in order to examine the structural
change of trade specialization pattern

Table 11 Results of t-test on the equality of coefficients

China
mean s.d. H0 t-statistic p-value
OLS 0.405 0.020 H0: ß o=ß r 0.035 0.486
BI Robust 0.404 0.381 H0: ß r=ß q 0.125 0.450
Quantile 0.399 0.029 H0: ß q=ß o -1.760 0.041**
OLS 0.242 0.151 H0: ß o=ß r -34.432 0.000***
LI Robust 0.779 0.031 H0: ß r=ß q 0.575 0.283
Quantile 0.777 0.019 H0: ß q=ß o 34.720 0.000***
OLS 0.820 0.050 H0: ß o=ß r -4.311 0.000***
SI Robust 0.851 0.050 H0: ß r=ß q -9.979 0.000***
Quantile 0.913 0.035 H0: ß q=ß o 15.032 0.000***
OLS 0.655 0.032 H0: ß o=ß r 0.378 0.353
WI Robust 0.654 0.032 H0: ß r=ß q 1.740 0.042**
Quantile 0.644 0.040 H0: ß q=ß o -2.073 0.020***
OLS 0.359 0.108 H0: ß o=ß r -2.556 0.006***
AI Robust 0.387 0.014 H0: ß r=ß q -71.114 0.000***
Quantile 0.500 0.008 H0: ß q=ß o 12.947 0.000***
OLS 0.957 0.284 H0: ß o=ß r -2.941 0.002***
NI Robust 1.042 0.036 H0: ß r=ß q -71.119 0.000***
Quantile 1.351 0.024 H0: ß q=ß o 13.702 0.000***

30 The t-test for two groups with equal sample sizes and equal variances is carried out here.
37
 Japan
mean s.d. H0 t-statistic p-value
OLS 1.020 0.045 H0: ß o=ß r 7.609 0.000***
BI Robust 0.984 0.016 H0: ß r=ß q -3.146 0.001***
Quantile 0.991 0.015 H0: ß q=ß o -6.215 0.000***
OLS 2.063 0.092 H0: ß o=ß r 67.396 0.000***
LI Robust 1.435 0.008 H0: ß r=ß q 48.752 0.000***
Quantile 1.381 0.007 H0: ß q=ß o -73.274 0.000***
OLS 0.978 0.029 H0: ß o=ß r -10.085 0.000***
SI Robust 1.017 0.025 H0: ß r=ß q -0.356 0.361
Quantile 1.018 0.024 H0: ß q=ß o 10.454 0.000***
OLS 0.886 0.039 H0: ß o=ß r 9.105 0.000***
WI Robust 0.849 0.013 H0: ß r=ß q -14.517 0.000***
Quantile 0.874 0.011 H0: ß q=ß o -3.053 0.001***
OLS 0.971 0.056 H0: ß o=ß r 40.030 0.000***
AI Robust 0.733 0.017 H0: ß r=ß q 21.099 0.000***
Quantile 0.697 0.003 H0: ß q=ß o -48.098 0.000***
OLS 0.573 0.034 H0: ß o=ß r 42.361 0.000***
NI Robust 0.425 0.008 H0: ß r=ß q 19.063 0.000***
Quantile 0.409 0.001 H0: ß q=ß o -48.337 0.000***

South Korea
mean s.d. H0 t-statistic p-value
OLS 0.472 0.043 H0: ß o=ß r 29.706 0.000***
BI Robust 0.319 0.027 H0: ß r=ß q -12.923 0.000***
Quantile 0.364 0.022 H0: ß q=ß o -21.939 0.000***
OLS 1.354 0.073 H0: ß o=ß r 5.276 0.000***
LI Robust 1.313 0.019 H0: ß r=ß q -11.823 0.000***
Quantile 1.338 0.009 H0: ß q=ß o -2.019 0.023**
OLS 0.653 0.058 H0: ß o=ß r -7.112 0.000***
SI Robust 0.712 0.057 H0: ß r=ß q -0.646 0.260
Quantile 0.718 0.072 H0: ß q=ß o 6.904 0.000***
OLS 0.733 0.067 H0: ß o=ß r 30.476 0.000***
WI Robust 0.493 0.041 H0: ß r=ß q -14.951 0.000***
Quantile 0.565 0.024 H0: ß q=ß o -23.549 0.000***
OLS 0.874 0.058 H0: ß o=ß r 57.856 0.000***
AI Robust 0.527 0.013 H0: ß r=ß q -138.759 0.000***
Quantile 1.000 0.031 H0: ß q=ß o 19.075 0.000***
OLS 0.853 0.056 H0: ß o=ß r 53.336 0.000***
NI Robust 0.541 0.012 H0: ß r=ß q -251.794 0.000***
Quantile 0.879 0.006 H0: ß q=ß o 4.522 0.000***
Note: ß o, ß r and ß q indicate the regression coefficients of OLS, robust and quantile regressions,
respectively.

In addition, we also cast some doubts in that the way of interpreting

Galtonian regression results is mainly dependent on the regression coefficient,
which possibly is affected by the intercept. That is, comparing the ‟s or ‟s
across countries or indices is only effective when the same intercepts are
guaranteed, which may be the reason why the results across indices in Tables 8

38
to 10 are confusing. Hart (1995) preferred the deviation form instead since the
constant term depends on units of measurement.

Thus, we suggest another way of interpreting the results, by the fitted

value of dependent variable together with Spearman rank correlation. One
advantage of this way of interpretation considering both slopes and intercepts is
that the problem of different intercepts hindering the comparison across
countries or indices does not exist anymore.

Accordingly, we presented in Table 12 the fitted values of the cutoff point

(i.e. comparative advantage neutral point) and the median. In addition, we
examine Spearman correlation coefficients between the index values of the two
points of time, namely, the average over 1995-1997 and 2006-2008, which are
also shown Table 12. As discussed earlier, the median can be an important
indicator in examining the change of trade pattern: as mentioned earlier, it is a
very good proxy of the share of sectors that show a comparative advantage (De
Benedictis and Tamberi, 2004). The sign of the variation between the
comparative-advantage neutral point of the independent variable and its fitted
value shown in Table 18 can give us information about whether or not the
country gained overall comparative advantage. Also, that between the median
point of the independent variable and its fitted value can tell us whether more or
fewer sectors have comparative advantage in later period of time than in earlier
period of time. At the same time, the Spearman correlation coefficient will tell us
how much the rankings across sectors within a given country have changed, i.e.
how stable or unstable the pattern of comparative advantage between the two
points of time is.

Examining Table 12, when using the BI, SI and AI for the robust
regression and the quantile regression, in case of China, we can witness that the
variation between the median of the independent variable and the fitted value
value is negative, that is, China has less sectors with comparative advantage (or
more sectors with comparative disadvantage) in 2007 than in 1996. Similarly,
South Korea also has fewer sectors with comparative advantage in 2007 than in
1996, while Japan has more. Especially in case of South Korea, except for when
using the NI for the quantile regression, the choice of the index does not make
any difference in terms of deciding whether more or fewer sectors have
comparative advantage than before. At the same time, the Spearman correlation
coefficients calculated with respect to Japan is on average 0.94, that is, there has
not been much change in terms of rankings of sectors during the period of 1996-
2007; in case of South Korea, it is 0.7 on average; and China, 0.78.

In other words, according to these results, Japan has gained overall

comparative advantage with stability of the trade specialization pattern, while
South Korea and China lost overall comparative advantage with less stability of
trade specialization pattern than that of Japan. This agrees with our intuitive
knowledge on dynamically changing Chinese and South Korean economies and

39
relatively stabilized Japanese economy; yet it is quite surprising that, despite the
dynamism of Chinese and Korean economies, they might still have lost overall
comparative advantage. However, we need to note that this analysis is based on
the 2-digit level of aggregation of exports data, which implies that different
results can be possibly expected when using data of more detailed level of
aggregation.

Another interesting finding according to these results is that, in empirical

studies that adopt the robust regression and the quantile regression using the
RCA indices, even though the BI has revealed many flaws and the NI seemed to
be having the most favorable features among the six indices, the BI might be „not
that bad‟, and at the same time, the NI be „not that good‟, considering their
concordant results.

40
Table 12 Fitted value of the Comparative-Advantage-Neutral Points and
the Medians

China
OLS Robust Reg. Quantile Reg.
Spearman
indep. est. indep. est. indep. est.
c.a.n. 1 0.879 _ 1 0.78 _ 1 0.71 _
BI 0.85
med. 0.91 0.843 _ 0.91 0.744 _ 0.91 0.674 _
c.a.n. 0 0 0 0.54 + 0 -0.003 _
LI 0.55
med. 0.14 0.034 _ 0.14 0.649 + 0.14 0.106 _
c.a.n. 0 -0.11 _ 0 -0.09 _ 0 -0.09 _
SI 0.86
med. -0.05 -0.151 _ -0.05 -0.133 _ -0.05 -0.136 _
c.a.n. 1 0.99 _ 1 0.92 _ 1 0.86 _
WI 0.86
med. 0.4 0.594 + 0.4 0.53 + 0.4 0.476 +
c.a.n. 0 0 0 -0.001 _ 0 -0.001 _
AI 0.78
med. 0 0 0 -0.001 _ 0 -0.001 _
c.a.n. 0 0.003 + 0 -0.46 _ 0 -0.47 _
NI 0.78
med. 0.012 0.015 + 0.012 -0.448 _ 0.012 -0.454 _
Japan
OLS Robust Reg. Quantile Reg.
Spearman
indep. est. indep. est. indep. est.
c.a.n. 1 1.08 + 1 1.006 + 1 0.45 +
BI 0.96
med. 0.171 0.234 + 0.171 0.194 + 0.171 0.252 +
c.a.n. 0 0 0 0.035 + 0 -0.05 +
LI 0.96
med. -0.093 -0.192 _ -0.093 -0.098 _ -0.093 -0.18 _
c.a.n. 0 0.04 + 0 0.464 + 0 -0.23 _
SI 0.96
med. -0.709 -0.655 + -0.709 -0.259 + -0.709 -0.506 +
c.a.n. 1 0.995 _ 1 0.891 _ 1 0.585 _
WI 0.96
med. 0.35 0.417 + 0.35 0.339 _ 0.35 0.273 _
c.a.n. 0 0 0 0 0 0
AI 0.9
med. -0.002 -0.002 -0.002 -0.001 + -0.002 -0.001 +
c.a.n. 0 0 0 0.016 + 0 0.004 +
NI 0.9
med. -1.494 -0.852 + -1.494 -0.626 + -1.494 -0.309 +
South Korea
OLS Robust Reg. Quantile Reg.
Spearman
indep. est. indep. est. indep. est.
c.a.n. 1 0.486 _ 1 0.474 _ 1 0.45 _
BI 0.74
med. 0.45 0.228 _ 0.45 0.298 _ 0.45 0.252 _
c.a.n. 0 0 0 -0.16 _ 0 -0.05 _
LI 0.61
med. -0.097 -0.131 _ -0.097 -0.287 _ -0.097 -0.18 _
c.a.n. 0 -0.023 _ 0 -0.21 _ 0 -0.23 _
SI 0.75
med. -0.384 -0.273 + -0.384 -0.483 _ -0.384 -0.506 _
c.a.n. 1 0.986 _ 1 0.513 _ 1 0.585 _
WI 0.75
med. 0.453 0.587 + 0.453 0.245 _ 0.453 0.273 _
c.a.n. 0 0 0 -0.001 _ 0 0
AI 0.63
med. -0.001 -0.001 -0.001 -0.001 -0.001 -0.001
c.a.n. 0 0 0 -0.19 _ 0 0.004
NI 0.62
med. -0.356 -0.302 _ -0.356 -0.382 _ -0.356 -0.309 +
Note: c.a.n, med., indep., est. denote comparative advantage neutral point, median, independent
variable, and estimation, respectively.

41
5. SUMMARY AND CONCLUSION
In the present study, we explored six important RCA indices and
suggested and showed a strategy when using these indices in analyzing trade
performance.

In Section 2, the relationship between the theoretical concept of

comparative advantage and the RCA indices that we practically obtain was
studied. Since the concept of comparative advantage is essentially based on
unobservable autarkic variables, the empirical researchers have looked for a
proper measure for comparative advantage, which is our interest here, the RCA
indices, using available variables from post-trade equilibria, such as exports and
imports.

In Section 3, an in-depth examination was made with respect to the

selected six RCA indices in terms of statistical properties and their comparability,
through which we find that the NI has more favorable features that an RCA
index should have than the other indices. This will be good information when
conducting non-econometric comparative analyses on trade specialization. We
also examined the Galtonian regression, an econometric method that uses RCA
indices for analyzing structural changes of trade specialization pattern between
two different points of time, and how to interpret the relevant results.

Then, in Section 4, we empirically explored the RCA indices taking an

example of mainly, China, Japan and South Korea, and a few more South East
Asian countries when needed, and derive a strategy of using the RCA indices in a
non-econometric comparative analysis as well as in regression analysis for
structural changes between two different points of time.

Regarding non-econometric comparative analysis, we find some useful

information: when using RCA indices as a dichotomous assessment, i.e. when the
matter is whether or not a given country has a comparative advantage in a given
sector, the BI, SI, AI and NI yield the same results, while the LI and the WI do
differently; when using cross-sector rankings of RCA indices, the BI, SI, AI and
NI yield the same results, while the LI and the WI do respectively differently;
when using cross-country rankings of RCA indices, the BI, SI and AI yield the
same results, while the LI, the WI and the NI do respectively differently; when
cardinally using RCA indices, especially in over-time analysis, each index yield
different results. Thus, in empirical researches, we recommend to be aware of
these findings and not to depend on only one index when drawing an inference.
In addition, when adopting the BI into cross-country analysis, the Hillman‟s
condition needs to be also met in order to assure the validity of the BI values.31

31
Although Hillman (1980) derived the condition only with respect to the BI, it may also be valid
for other indices that are transforms of the BI.
42
 With respect to econometric analysis for structural changes of trade
specialization, i.e. the OLS Galtonian regression, the normality assumption of
error terms and the existence of outliers have to be checked beforehand, by
running normality tests and calculating Cook‟s D, respectively. We found that,
among the six RCA indices that we examined, the SI seems to have the closest
distribution to normality, although the normality is not always guaranteed, as
also found by Laursen (1998) and Dalum et al. (1998). However, the other RCA
indices are far from being normally distributed: they have high skewness due to
the existence of outliers and high kurtosis as well. Thus, we suggest using
different methods: the log transformation and the Box-Cox transformation in
order to resolve the normality issue to some extent; and the robust regression
and the quantile regression that will yield more effective results when the case is
with outliers (and non-normality).

Now, the issue is how to interpret the regression results. Since the results
yielded from the OLS are mostly significantly different from those by the robust
or quantile regressions, we cast doubts on the appropriateness of the OLS
regression coefficient. Also, Hart (1995) who first used Galtonian regression in
analyzing the business concentration only focuses on the coefficient of
independent variable, , and the coefficient of determination when comparing
the results across countries, which we believe to be valid only when the same
intercepts are guaranteed. Therefore, we suggest a different way of interpreting
the results by taking into consideration intercepts, i.e. by using the fitted value
of the dependent variable together with the Spearman correlation coefficient
between the two time periods. One interesting finding is that, with the way of
interpretation that considers both the slopes and intercepts, which is suggested
in this study, the results yielded by different indices agree to each other
substantially. For example, when using the robust regression and the quantile
regression in case of South Korea, by all of the RCA indices, we could infer that
South Korea gain overall comparative advantage during the period of 1996-2007
and the stability of cross-sector rankings is lower than Japan‟s. The whole
process so far is presented in Table 13.

The hypotheses we provided earlier in the introduction can be checked at

this point. a) The problems inherent in the RCA indices can be resolved by being
informed of pros and cons of these indices and adequately using them. b) The
usefulness of the RCA indices can be achieved by formulating a proper strategy
of using them. c) A proper strategy can be formulated by applying the methods
suggested in this study.

There can be many other ideas of how to effectively use the RCA indices
other than those introduced and suggested in the present study. For instance, De
Benedictis and Tamberi (2004) suggested a research strategy using kernel
density estimation, Markov stationary methods, and so on. In addition, further
research can be undertaken on expanding the idea of Galtonian regression,
43
which takes into account only two different points of time, to the fixed effect
model within more continuous time basis. Added to that, as already pointed out
several times, one should note that the aggregation level of data affects the
results to some extent, and the present study used the data aggregated at 2-digit
level of HS classification. The situation would be expected to be different when
the lower aggregation level, e.g. 4-digit or 6-digit of HS classification, which is
not covered in the present study mainly due to time limitation. Further research
regarding this issue also can be done based on this study.

44
Table 13 Strategies of Using RCA Indices

1 Calculate RCA indices.

Dichotomous The BI, SI, AI and NI yield the same results, while the
assessment LI and the WI do respectively differently.
The BI, SI, AI and NI yield the same results, while the
Cross-sector
LI and the WI do respectively differently. Although,
analysis
the LI is highly correlated with the AI and NI after
(ordinal)
eliminating a few outlying sectors.

Non- Cross- Check Hillman's condition for the BI and eliminate

econometric country the data that violate the condition. The BI, SI and AI
2 analysis yield the same results, while the LI, the WI and the
Comparative
Analysis (ordinal) NI do respectively differently.
Over-time
analysis Each index yield different results.
(cardinal)
The NI has more favorable features as an RCA index than other
indices do. Particularly, it is not affected by the level of aggregation
and bears comparability in cross-country, cross-sector and over time
analyses.

a. Conduct a preliminary examination on RCA indices‟ distributions,

e.g. examining summary statistics and histograms.
b. Run OLS regression, and run the normality tests and calculate
Cook‟s D with respect to the corresponding residuals.

c. If normality is met, interpret the regression results. Compare the

slopes if the intercepts are close to each other; if not, calculate the
fitted values of median and mean, which considers both slopes and
intercepts, and Spearman rank correlation coefficient. (end)
d. If normality is not met, carry out log-transformation and Box-Cox
transformation.
Econometric
Analysis e. Run OLS regression on the transformed data, and run the
3 normality tests on the corresponding residuals.
(Galtonian
Regression) f. If normality is met, interpret the regression results. Compare the
slopes if the intercepts are close to each other; if not, calculate the
fitted values of median and mean, which considers both slopes and
intercepts, and Spearman rank correlation coefficient. (end)
g. If normality is not met, run the robust regression and the quantile
regression, which do not assume normality of error terms and yield
effective results with outliers. Compare the slopes if the intercepts
are close to each other; if not, calculate the fitted values of median
and mean, which considers both slopes and intercepts, and Spearman
rank correlation coefficient. (end)
Here, the BI might be „not that bad‟, and at the same time, the NI be
„not that good‟, considering their concordant results.

45
References

ALESSANDRINI, M., FATTOUH, B. & SCARAMOZZINO, P. 2007. The changing pattern of

foreign trade specialization in Indian manufacturing. Oxford Review of Economic Policy,
23, 270-291.
BALASSA, B. 1965. Trade liberalisation and "revealed" comparative advantage. The Manchester
school of economic and social studies, 33, 99-123.
BALASSA, B. 1977. 'Revealed' comparative advantage revisited: an analysis of relative export
shares of the industrial countries, 1953-1971. The Manchester school of economic and
social studies, 45, 327-344.
BALLANCE, R. H., FORSTNER, H. & MURRAY, T. 1987. Consistency tests of alternative
measures of comparative advantage. Review of Economics & Statistics, 69, 157-161.
BOJNEC, Š. 2001. Trade and revealed comparative advantage measures regional and Central
and East European agricultural trade. Eastern European Economics, 39, 72-97.
BOWEN, H. 1983. On the theoretical interpretation of indices of trade intensity and revealed
comparative advantage. Review of World Economics, 119, 464-472.
BOWEN, H. 1985. On measuring comparative advantage: a reply and extension.
Weltwirtschaftliches Archiv, 121, 351-354.
BOX, G. E. P. & COX, D. R. 1964. An Analysis of Transformations. Journal of the Royal
Statistical Society. Series B (Methodological), 26, 211-252.
CAI, J. & LEUNG, P. 2008. Towards a more general measure of revealed comparative advantage
variation. Applied Economics Letters, 15, 723-726.
CANTWELL, J. 1989. Technological innovation and multinational corporations, Oxford, Basil
Blackwell.
COOK, R. D. 1977. Detection of influential observation in linear regression. Technometrics, 19,
15-18.
CORDELLA, T. 1998. Patterns of trade and oligopoly equilibria: an example. Review of
International Economics, 6, 554-563.
DALUM, B., LAURSEN, K. & VILLUMSEN, G. 1998. Structural change in OECD export
specialisation patterns: de-specialisation and 'stickiness'. International Review of
Applied Economics, 12, 423-443.
DAVIS, D. R. 1997. Critical evidence on comparative advantage? North-North trade in a
multilateral world. Journal of Political Economy, 105, 1051-1060.
DE BENEDICTIS, L. & TAMBERI, M. 2001. A note on the Balassa index of revealed
comparative advantage. Working Papers #158. Dipartimento di Economia, Universita'
Politecnica delle Marche (Ⅰ), doi:10.2139/ssrn.289602.
DE BENEDICTIS, L. & TAMBERI, M. 2004. Overall Specialization Empirics: Techniques and
Applications. Open Economics Review, 15, 323-346.
DEARDORFF, A. V. 1980. The general Validity of the law of comparative advantage. Journal of
Political Economy, 88, 941-957.
DOWLING, M. & CHEANG, C. T. 2000. Shifting comparative advantage in Asia: new test of the
'flying gese' model. Journal of Asian Economics, 11, 443.
DRABICKI, J. Z. & TAKAYAMA, A. 1979. An antinomy in the theory of comparative advantage.
Journal of International Economics, 9, 211-223.
FAGERBERG, J. 1988. International Competitiveness. The Economic Journal, 98, 355-374.
FRANTZEN, D. 2008. Technology, competitiveness and specialisation in OECD manufacturing.
Journal of Economic Studies, 35, 44-68.
GALTON, F. 1889. Natural Inheritance, London, Macmillan.
HAMILTON, L. C. 2009. Statistics with Stata : updated for version 10, Belmont, CA, Brooks/Cole.
HART, P. E. & PRAIS, S.J. 1956. The analysis of business concentration: a statistical approach.
Journal of the Royal Statistical Society. Series A (General), 119, 150-191.
HART, P.E. 1995. The convergence of labour productivity in British and German industries.
International Journal of the Economics of Business, 2, 453-464.
HART, P. E. 2000. Theories of firms' growth and the generation of jobs. Review of Industrial
46
 Organization, 17, 229.
HILLMAN, A. L. 1980. Observations on the relation between 'revealed comparative advantage'
and comparative advantage as indicated by pre-trade relative prices.
Weltwirtschaftliches Archiv, 116, 315-321.
HINLOOPEN, J. & VAN MARREWIJK, C. 2001. On the empirical distribution of the Balassa
index. Weltwirtschaftliches Archiv, 137, 1-35.
HINLOOPEN, J. & VAN MARREWIJK, C. 2008. Empirical relevance of the Hillman condition
for revealed comparative advantage: 10 stylized facts. Applied Economics, 40, 2313-2328.
HISANAGA, M. 2008. Revealed specialization: evidence on U.S. international services. The
International Trade Journal, 22, 378-414.
HOEN, A. & OOSTERHAVEN, J. 2006. On the measurement of comparative advantage. Annals
of Regional Science, 40, 677-691.
HUMMELS, D. & LEVINSOHN, J. 1993. Product differentiation as a source of comparative
advantage. American Economic Review, 83, 445.
INTERNATIONAL TRADE CENTRE 2000. PC-TAS HS Rev.0 1995-1999. Geneva: International
Trade Centre.
INTERNATIONAL TRADE CENTRE 2003. PC-TAS HS Rev.1 1998-2002. Geneva: International
Trade Centre.
INTERNATIONAL TRADE CENTRE 2008. PC-TAS HS Rev.2 2003-2007. Geneva: International
Trade Centre.
KRUGMAN, P. 1979. Increasing returns, monopolistic competition, and international trade.
Journal of International Economics, 9, 469-479.
LAFAY, G. 1992. The measurement of revealed comparative advantages. In: DAGENAIS, M. G.
& MUET, P.-A. (eds.) International Trade Modelling. London: Chapman & Hall.
LAURSEN, K. 1998. Revealed comparative advantage and the alternatives as measures of
international specialisation. DRUID Working Paper No.98-30.
LIESNER, H. H. 1958. The European Common Market and British Industry. The Economic
Journal, 68, 302-316.
MARCHESE, S. & DE SIMONE, F. 1989. Monotonicity of indices of "revealed" comparative
advantage: empirical evidence on Hillman's condition. Review of World Economics, 125,
158-167.
MEMEDOVIC, O. 1994. On the theory and measurement of comparative advantage : an empirical
analysis of Yugolslav trade in manufactures with the OECD countries 1970-1986,
Amsterdam, Thesis Publishers.
OHLIN, B. G. 1933. Interregional and international trade, Cambridge : (1957 printing), Harvard
Univ. Press.
PALLEY, T. I. 2008. Institutionalism and New Trade Theory: Rethinking Comparative
Advantage and Trade Policy. Journal of Economic Issues (Association for Evolutionary
Economics), 42, 195-208.
PARK, H. M. 2008. Univariate analysis and normality test using SAS, Stata, and SPSS. Working
paper. The University Information Technology Services Center for Statistical and
Mathematical Computing, Indiana University.
PETERSON, J. 1988. Export shares and revealed comparative advantage: a study of
international travel. Applied Economics, 20, 351-365.
PROUDMAN, J. & REDDING, S. 1998. Openness and growth. The Bank of England.
PROUDMAN, J. & REDDING, S. 2000. Evolving patterns of international trade. Review of
International Economics, 8, 373-396.
RICARDO, D. 1817/1951. On the principles of political economy and taxation. In: SRAFFA, P.
(ed.) The works and correspondence of David Ricardo, Vol.1. Cambridge, UK: Cambridge
University Press.
ROUSSEEUW, R. & LEROY, A. 1987. Robust Regression and Outlier Detection. New York: Wiley
SANIDAS, E. 2007. Revealed comparative advantage and economic development with emphasis
on East Asia. Proceedings of the Korea in the World Economy, VI Conference, 2, 445-463.
SANIDAS, E. 2009. Patterns and distances of catch-up in trade: China and East Asia. China
Economic Journal, 2, 105-118.

47
SEYOUM, B. 2007. Revealed comparative advantage and competitiveness in services: A study
with special emphasis on developing countries. Journal of Economic Studies, 34, 376-
388.
SHARMA, A. & DIETRICH, M. 2007. The Structure and Composition of India's Exports and
Industrial Transformation (1980-2000). International Economic Journal, 21, 207-231.
STREET, J. O., CARROL, R. J. & RUPPERT, D. 1988. A note on computing robust regression
estimates via iteratively reweighted least squares. The American Statistician, 42, 152-
154.
TYBOUT, J. 1993. Internal returns to scale as a source of comparative advantage: the evidence.
The American Economic Review, 440-444.
UNIDO. 1982. Changing patterns of trade in world industry: an empirical study on revealed
comparative advantage, New York.
VOLLRATH, T. 1991. A theoretical evaluation of alternative trade intensity measures of
revealed comparative advantage. Weltwirtschaftliches Archiv, 127, 265-280.
WÖRZ, J. 2005. Dynamics of Trade Specialization in Developed and Less Developed Countries.
Emerging Markets Finance & Trade, 41, 92-111.
YEATS, A. 1985. On the appropriate interpretation of the revealed comparative advantage index:
implications of a methodology based on industry sector analysis. Weltwirtschaftliches
Archiv, 121, 61-73.
YU, R., CAI, J. & LEUNG, P. 2009. The normalized revealed comparative advantage index.
Annals of Regional Science, 43, 267-282.

48
Appendix 1 Description of HS 2-digit Codes
HS Description
1 Live animals
2 Meat and edible meat offal
3 Fish, crustaceans, molluscs and other aquatic invertebrates
4 Birds' eggs; natural honey; edible products of aminal origina, n.e.s.
5 Products of animal origin, n.e.s.
6 Live trees and other plants; bulb, roots and the like; cut flowers and ornamental foliage
7 Edible vegetables and certain roots and tubers
8 Edible fruit and nuts; peel of citrus fruit or melons
9 Coffee, tea, mate and spices
10 Cereals
11 Products of the milling industry; malt; starches; inulin; wheat gluten
Oil seed and oleaginous fruits; miscellaneous grains, seeds and fruit; industrial or medicinal plants; straw and
12
fodder
13 Lac; gums, resins and other vegetable saps and extracts
14 Vegetable plaiting materials; vegetable products n.e.s.
15 Animal or vegetable fats and oils and their cleavage products; prepared edible fats; aminal or vegetable waxes
16 Preparations of meat, of fish or of crustaceans, molluscs or other aquatic invertebrates
17 Sugars and sugar confectionery
18 Cocoa and cocoa preparations
19 Preparations of cereal, flour, starch or milk; pastry cooks' products
20 Preparations of vegetables, fruit, nuts or other parts of plants
21 Miscellaneous edible preparations
22 Beverages, spirits and vinegar
23 Residues and wastes from the food industry; prepared animal fodder
24 Tobacco and manufactured tobacco substitutes
25 Salt; sulphur; earth & stone; plastering materials; lime & cement
26 Ores, slag and ash
27 Mineral fuels, mineral oils and products of their distillation; bituminous substances; mineral waxes
Inorganic chemicals; organic or inorganic compounds of precious metals, of rare-earth metals, of radioactive
28
elements or of isotopes
29 Organic chemicals
30 Pharmaceutical products
31 Fertilizers
Tanning or dyeing extracts; tannins and their derivatives; dyes, pigments and other coloring matter; paints and
32
varnishes; putty and other mastics; Inks
33 Essential oils and resinoids; perfumery cosmetics or toilet preparations
Soap, organic surface-active agents, washing preparations, lubricating preparations, artificial waxes, prepared
34 waxes polishing or scouring preparations, candles and similar articles, modelling pastes, 'dental waxes' and dental
preparations
35 Albuminoidal substances; modified starches; glues; enzymes
36 Explosives; pyrotechnic products, matches; pyrophoric alloys; certain combustible preparations
37 Photographic or cinematographic goods
38 Miscellaneous chemical products
39 Plastics and articles thereof
40 Rubber and articles thereof
41 Raw hides and skins (other than furskins) and leather
Articles of leather; saddlery and harness; travel goods, handbags and similar containers; articles of animal gut
42
(other than silk-worm gut
43 Furskins and artificial fur; manufactures thereof
44 Wood and articles of wood; wood charcoal
45 Cork and articles of cork
46 Manufactures of straw, of esparto or of other plaiting materials; basketware and wickerwork
47 Pulp of wood or of other fibrous cellulosic material; recovered (waste and scrap) of paper or paperboard
48 Paper and paperboard; articles of pulp, of paper or of paperboard
49 Printed books, newspapers, pictures and other products of the printing industry; manuscript, typescripts and plans
50 Silk
51 Wool, fine or coarse animal hair, horse hair yarn and woven fabric
52 Cotton
53 Other vegetable textile fibres; paper yarn and woven fabrics of paper yarn
54 Man-made filaments
55 Man-made staple fibres
56 Wadding, felt and nonwovens; special yarns; twine, cordage, ropes and cables and articles thereof
57 Carpets and other textile floor coverings
58 Special woven fabrics; tufted textile fabrics; lace; tapestries; trimmings; embroidery
59 Impregnated, coated, covered or laminated textile fabrics; textile articles of a kind suitabe for industrial use
60 Knitted or crocheted fabrics
61 Articles of apparel and clothing accessories, knitted or crocheted

49
62 Articles of apparel and accessories, not knitted or crocheted
63 Other made up textile articles; sets; worn clothing and worn textile articles; rags
64 Footwear, gaiters and the like; parts of such articles
65 Headgear and parts thereof
66 Umbrellas, sun umbrellas, walking-sticks, seat-sticks, whips, riding-crops, and parts thereof
67 Prepared feathers and down and articles made of feathers or of down; artificial flowers; articles of human hair
68 Articles of stone, plaster, cement, asbestos, mica or similar materials
69 Ceramic products
Glass and glassware
70
Natural or cultured pearls, precious or semi-precious stones, precious metals, metals clad with precious metal, and
71
articles thereof; imitation jewellery; coin
72 Iron and steel
73 Articles of iron or steel
74 Copper and articles thereof
75 Nickel and articles thereof
76 Aluminium and articles thereof
77 Reserved for Possible Future Use
78 Lead and articles thereof
79 Zinc and articles thereof
80 Tin and articles thereof
81 Other base metals; cermets articles thereof
82 Tools, implements, cutlery, spoons and forks, of base metal; parts thereof of base metal
83 Miscellaneous articles of base metal
84 Nuclear reactors, boilers, machinery and mechanical appliance; parts thereof
Electrical machinery and equipment and parts thereof; sound recorders and reproducers, television image and
85
sound recorders and reproducers and parts and accessories of such articles
Railway or tramway locomotives, rolling-stock and parts thereof; rail-way or tramway track fixtures and
86
fittingsand parts thereof; mechanical (including electro-mechanical) traffic signalling equipment of all kinds
87 Vehicles other than railway or tramway rolling stock, and parts and accessories thereof
88 Aircraft, spacecraft, and parts thereof
89 Ships, boats and floating structures
Optical, photographic, cinematographic, measuring, checking, precision, medical or surgical instruments and
90
apparatus; parts and accessories thereof
91
Clocks and watches and parts thereof
92 Musical instruments; parts and accessories or such articles
93 Arms and ammunition; parts and accessories thereof
Furniture; bedding, mattresses, mattress supports, cushions and similar stuffed furnishings; lamps and lighting
94
fittings, n.e.s; illuminated signs, illuminate name-plates and the like; prefabricated building
95 Toys, games and sports requisites; parts and accessories thereof
96 Miscellaneous manufactured articles
97 Works of art, collectors' pieces & antiques
98 Commodies specified at chapter level only
99 Commodities n.e.s.

Appendix 2 Comparison of RCA Indices

Table A1 ‘Apparent’ contradiction between RCA indices (2008)

Sectoral
HS Rank BI LI SI WI AI NI
share
S.Korea 72 0.048 7 1.499 -20.543 0.2 2.5 0.016 3.923
84 0.116 2 0.947 -5.158 -0.027 1.579 -0.007 -1.606
Japan 81 0.002 29 1.656 -0.278 0.247 2.745 0.001 0.408
28 0.007 16 0.888 -0.727 -0.059 1.472 -0.001 -0.402
China 85 0.23 1 1.892 -1.452 0.308 1.299 0.109 96.365
72 0.037 4 1.172 4.62 0.079 0.804 0.005 4.865
Note: „Sectoral share‟ indicates the export share of the corresponding sector with respect to the
nation‟s total export market. Sectors with comparative advantage are tinted. See Appendix 2 for
all sectors.

50
Table A2 Number of Sectors with Comparative Advantage and
Comparative Disadvantage

China Japan South Korea

1995- 2006- 1995- 2006- 1995- 2006-
1997 2008 1997 2008 1997 2008
Number of sectors with Comparative
47 45 14 18 28 21
B Advantage
I Number of sectors with Comparative
51 52 84 79 70 76
Disadvantage
Number of sectors with Comparative
57 60 29 31 36 25
Advantage
LI
Number of sectors with Comparative
41 37 68 66 62 72
Disadvantage
Number of sectors with Comparative
47 45 14 18 28 21
Advantage
SI
Number of sectors with Comparative
51 53 84 80 70 77
Disadvantage
Number of sectors with Comparative
30 35 34 32 28 26
W Advantage
I Number of sectors with Comparative
68 63 64 66 70 72
Disadvantage
Number of sectors with Comparative
47 45 14 18 28 21
A Advantage
I Number of sectors with Comparative
51 52 84 79 70 76
Disadvantage
Number of sectors with Comparative
47 45 14 18 28 21
N Advantage
I Number of sectors with Comparative
51 52 84 79 70 76
Disadvantage
Note: CAN denotes „comparative advantage neutral.‟; note that the total number of sectors taken
into account in this analysis is 98 in some cases, and 97 in other cases since HS 98 is not recently
used in some countries

Table A3 Percentage of sectors that show the same results by pair-

wise comparison between indices (averaged over China, Japan and
South Korea, 2008)
BI LI SI WI AI NI
BI 100%
LI 80% 100%
SI 100% 80% 100%
WI 90% 76% 90% 100%
AI 100% 80% 100% 90% 100%
NI 100% 80% 100% 90% 100% 100%

51
Table A4 Cross-sector rankings of principal sectors32: China, Japan
and South Korea for 2008
a) China
HS BI LI SI WI AI NI
46 1 31 1 1 29 29
66 2 35 2 2 34 34
67 3 32 3 3 30 30
50 4 39 4 4 35 35
65 5 29 5 5 31 31
58 6 22 6 6 19 19
63 7 10 7 7 9 9
42 8 11 8 8 10 10
61 9 2 9 9 3 3
95 10 6 10 10 6 6
64 11 7 11 11 7 7
62 12 3 12 12 4 4
96 13 15 13 13 15 15
86 14 12 14 14 13 13
60 15 27 15 15 18 18
94 16 4 16 16 5 5
92 17 40 17 17 36 36
89 27 9 27 27 11 11
85 28 88 28 28 1 1
84 33 1 33 33 2 2
b) Japan
HS BI LI SI WI AI NI
37 1 14 1 1 9 9
89 2 6 2 2 4 4
87 3 1 3 3 1 1
92 4 26 4 4 19 19
81 5 65 5 5 17 17
84 6 2 6 6 2 2
90 7 7 7 7 6 6
38 8 11 8 8 7 7
72 9 4 9 9 5 5
70 11 16 11 11 11 11
85 12 3 12 12 3 3
96 13 22 13 13 18 18
82 14 17 14 14 14 14
32 15 15 15 15 15 15
29 16 13 16 16 10 10
74 18 12 18 18 13 13
55 19 19 19 19 21 21
99 20 5 20 20 12 12
39 21 8 21 21 16 16
73 24 9 24 24 57 57
c) South Korea
HS BI LI SI WI AI NI
89 1 5 1 1 6 6
60 2 9 2 2 10 10
85 3 1 3 3 1 1

32These sectors are chosen because they are in general considered to be important in each
country, and can well reveal the differences across the six RCA indices.
52
 79 4 11 4 4 14 14
54 5 10 5 5 9 9
90 6 3 6 6 2 2
59 7 12 7 7 13 13
29 8 6 8 8 4 4
55 9 13 9 9 12 12
39 10 4 10 10 3 3
72 11 97 11 11 7 7
40 12 7 12 12 8 8
87 13 2 13 13 5 5
73 14 8 14 14 11 11
92 15 27 15 15 18 18
74 16 81 16 16 15 15
58 17 17 17 17 17 17
41 18 32 18 18 16 16
56 19 20 19 19 22 22
84 22 95 22 22 88 88

Table A5 Pair-wise Spearman rank correlation coefficients between

RCA indices, averaged over China, Japan and South Korea for 2008
BI LI SI WI AI NI
BI 1
LI 0.598 1
SI 1 0.598 1
WI 1 0.598 1 1
AI 0.704 0.782 0.704 0.704 1
NI 0.704 0.782 0.704 0.704 1 1
Note: All of the indices are not significantly independent for each other, i.e. the null hypothesis of
pair-wise independence is not rejected for all cases.

Table A6 Cross-country rankings of selected sectors, Asian Countries,

2008
HS Country BI LI SI WI AI NI
72 CHINA 3 2 3 3 3 2
(Iron and steel) HONGKONG 9 3 9 7 9 9
INDONESIA 5 5 5 9 5 5
JAPAN 1 1 1 1 1 1
KOREA REP. 2 7 2 2 2 3
MALAYSIA 6 8 6 4 6 7
PHILIPPINES 8 4 8 8 8 4
SINGAPORE 7 6 7 5 7 8
THAILAND 4 9 4 6 4 6
76 CHINA 1 1 1 2 1 1
(Aluminum and articles thereof) HONGKONG 8 2 8 8 8 8
INDONESIA 3 3 3 7 3 3
JAPAN 6 5 6 4 6 9
KOREA REP. 5 6 5 1 5 5
MALAYSIA 4 7 4 3 4 4
PHILIPPINES 9 4 9 9 9 6
SINGAPORE 7 9 7 5 7 7
THAILAND 2 8 2 6 2 2
84 CHINA 1 4 1 5 1 1
(Nuclear reactors, boilers, etc.) HONGKONG 7 7 7 7 7 7
INDONESIA 9 9 9 9 9 8
JAPAN 2 6 2 1 2 2
KOREA REP. 8 8 8 3 8 9
MALAYSIA 6 5 6 6 6 6

53
 PHILIPPINES 3 3 3 4 3 3
SINGAPORE 5 2 5 2 5 5
THAILAND 4 1 4 8 4 4
85 CHINA 5 6 5 6 5 7
(Electrical, electronic equipment) HONGKONG 2 4 2 3 2 1
INDONESIA 9 8 9 9 9 8
JAPAN 6 5 6 5 6 9
KOREA REP. 3 2 3 1 3 2
MALAYSIA 7 9 7 7 7 5
PHILIPPINES 1 3 1 4 1 4
SINGAPORE 4 1 4 2 4 3
THAILAND 8 7 8 8 8 6

Table A7 Spearman Rank Correlation Coefficients between Indices,

averaged over HS 72, 76, 84 and 85
BI LI SI WI AI NI
BI 1
LI 0.40 1
SI 1* 0.40 1
WI 0.63 0.27 0.63 1
AI 1* 0.40 1* 0.63 1
NI 0.84* 0.45 0.84* 0.47 0.84* 1
Note: Asterisk indicates that the null hypothesis of independence between the two indices is
rejected in all four sectors.

Table A8 Variation between 1995 and 2008, South Korea

1995 2008 1995 2008 1995 2008
HS BI BI Variation NI NI Variation LI LI Variation
1 0.04 0.01 -0.03 -0.6 -0.26 0.34 -0.04 -0.03 0
2 0.1 0 -0.09 -2.2 -1.48 0.73 -1.12 -2.29 -1.17
3 1.24 0.62 -0.62 0.5 -0.41 -0.91 0.99 -1.14 -2.13
4 0 0 0 -1.88 -1.07 0.81 -0.22 -0.5 -0.28
5 0.29 0.22 -0.07 -0.14 -0.09 0.05 -0.23 -0.08 0.15
6 0.05 0.06 0.02 -0.49 -0.25 0.24 -0.04 -0.03 0.01
7 0.2 0.08 -0.12 -1.01 -0.69 0.31 0.02 -0.4 -0.42
8 0.2 0.06 -0.14 -1.23 -1.02 0.21 0.01 -0.39 -0.4
9 0.02 0.03 0.01 -0.87 -0.48 0.39 -0.45 -0.23 0.23
10 0 0 0 -2.34 -1.58 0.76 -3.31 -3.87 -0.56
11 0.07 0.05 -0.02 -0.35 -0.23 0.13 -0.11 -0.27 -0.16
12 0.51 0.11 -0.4 -0.52 -0.87 -0.36 -0.65 -1.28 -0.63
13 1.37 0.46 -0.91 0.04 -0.04 -0.08 0.03 -0.05 -0.09
14 0.18 0.04 -0.15 -0.02 -0.01 0.01 -0.02 0 0.02
15 0.03 0.01 -0.02 -1.62 -1.33 0.29 -0.7 -0.99 -0.28
16 1.06 0.14 -0.92 0.05 -0.5 -0.55 0.56 -0.47 -1.03
17 0.53 0.19 -0.34 -0.46 -0.39 0.07 -0.54 -0.21 0.33
18 0.1 0.04 -0.07 -0.59 -0.48 0.11 -0.09 -0.18 -0.09
19 0.55 0.26 -0.29 -0.39 -0.52 -0.12 0.26 -0.06 -0.32
20 0.16 0.13 -0.02 -1 -0.65 0.35 -0.36 -0.53 -0.17
21 0.28 0.36 0.08 -0.72 -0.43 0.3 -0.14 -0.34 -0.21
22 0.12 0.15 0.03 -1.77 -1.15 0.62 -0.2 -0.57 -0.37
23 0.05 0.06 0.01 -1.14 -0.76 0.38 -0.89 -1.82 -0.93
24 0.08 0.23 0.15 -1.18 -0.39 0.79 -0.58 -0.18 0.4
25 0.49 0.46 -0.03 -0.5 -0.4 0.1 -0.71 -0.23 0.49
26 0.03 0.07 0.04 -1.24 -2.09 -0.85 -2.88 -8.26 -5.38
27 0.36 0.47 0.11 -9.62 -22.06 -12.44 -29.15 -31.31 -2.17
28 0.32 0.83 0.51 -1.34 -0.3 1.03 -1.73 -2.43 -0.7
29 0.88 1.82 0.93 -0.82 4.52 5.34 -4.41 3.04 7.45
30 0.06 0.06 0 -3.32 -5.62 -2.3 -0.47 -2.98 -2.51
31 0.65 0.42 -0.23 -0.3 -0.63 -0.32 0.21 0.16 -0.05

54
32 0.51 0.66 0.15 -0.9 -0.35 0.56 -0.84 -0.86 -0.02
33 0.07 0.14 0.07 -1.35 -1.12 0.23 -0.54 -1.2 -0.66
34 0.29 0.33 0.04 -0.63 -0.47 0.16 -0.4 -0.41 -0.01
35 0.35 0.56 0.21 -0.31 -0.15 0.16 -0.32 -0.25 0.07
36 0.68 0.23 -0.45 -0.02 -0.04 -0.02 0.04 0 -0.03
37 0.3 0.54 0.23 -0.74 -0.14 0.61 -0.93 -0.86 0.07
38 0.27 0.49 0.22 -2.02 -1.2 0.82 -2.5 -4.4 -1.9
39 1.27 1.7 0.43 2.43 5.12 2.69 5.55 10.54 4.98
40 1.18 1.48 0.29 0.58 1.12 0.54 1.34 2.78 1.43
41 3.08 1.07 -2.01 2.33 0.03 -2.3 0.25 -0.05 -0.3
42 2.36 0.12 -2.24 1.32 -0.67 -1.98 1.73 -1.33 -3.06
43 0.52 0.07 -0.45 -0.1 -0.1 0 -1.02 -0.14 0.88
44 0.08 0.02 -0.06 -3.38 -1.75 1.63 -4.39 -2.18 2.21
45 0.01 0.05 0.04 -0.07 -0.03 0.04 -0.01 0 0.01
46 0.4 0.03 -0.38 -0.04 -0.04 0 -0.03 -0.05 -0.01
47 0.02 0.08 0.06 -1.64 -0.57 1.06 -3.28 -1.59 1.69
48 0.46 0.51 0.05 -3.19 -1.31 1.87 0.76 0.54 -0.22
49 0.18 0.29 0.12 -1.19 -0.52 0.67 -0.08 -0.06 0.01
50 3.92 0.96 -2.96 0.5 0 -0.51 0.13 -0.08 -0.21
51 0.41 0.27 -0.14 -0.52 -0.15 0.37 -0.92 -0.31 0.61
52 0.88 0.49 -0.39 -0.21 -0.41 -0.2 -1.25 -0.62 0.63
53 0.74 0.24 -0.5 -0.04 -0.04 0 -0.14 -0.07 0.08
54 8.36 2.36 -6 12.09 0.86 -11.23 9.9 1.37 -8.53
55 3.27 1.78 -1.49 3.5 0.38 -3.13 2.26 0.55 -1.71
56 1.83 0.98 -0.86 0.45 -0.01 -0.46 0.56 0.05 -0.51
57 0.11 0.17 0.07 -0.45 -0.18 0.27 -0.05 -0.04 0.02
58 3.7 1.1 -2.6 0.93 0.02 -0.91 0.9 0.24 -0.67
59 4.06 2.03 -2.03 1.7 0.33 -1.37 1.68 0.57 -1.11
60 5.48 3 -2.48 2.24 0.73 -1.51 2.27 1.54 -0.73
61 1.63 0.22 -1.41 1.99 -2.13 -4.13 3.86 -1.38 -5.25
62 0.98 0.08 -0.9 -0.1 -2.54 -2.44 3.04 -2.36 -5.4
63 1.21 0.34 -0.87 0.16 -0.45 -0.61 0.69 -0.06 -0.76
64 1.5 0.09 -1.41 1.12 -1.27 -2.39 2.27 -0.69 -2.97
65 2.95 0.92 -2.03 0.21 -0.01 -0.22 0.25 0.08 -0.18
66 0.08 0.05 -0.03 -0.05 -0.04 0.01 0 -0.07 -0.07
67 2.37 0.61 -1.75 0.1 -0.02 -0.12 0.09 -0.01 -0.1
68 0.6 0.49 -0.11 -0.37 -0.33 0.04 -0.03 -0.94 -0.91
69 0.15 0.12 -0.02 -1.03 -0.54 0.49 -0.27 -0.83 -0.56
70 0.29 0.66 0.37 -1.12 -0.34 0.78 -1.18 -1.94 -0.76
71 1.29 0.3 -0.99 1.42 -3.78 -5.2 -0.09 -0.88 -0.79
72 1.43 1.5 0.07 3.15 3.92 0.78 -3.23 -20.54 -17.31
73 1.57 1.13 -0.43 2.67 0.62 -2.05 3.93 2 -1.94
74 0.54 1.11 0.57 -1.02 0.25 1.27 -2.73 -1.49 1.24
75 0.19 0.32 0.13 -0.29 -0.31 -0.02 -0.58 -0.72 -0.14
76 0.54 0.58 0.03 -1.47 -1.06 0.41 -2.7 -2.95 -0.25
78 0.38 0.96 0.57 -0.06 0 0.05 -0.17 -0.09 0.08
79 0.53 2.68 2.14 -0.12 0.32 0.44 -0.01 0.63 0.64
80 0.25 0.42 0.16 -0.07 -0.06 0.01 -0.11 -0.24 -0.13
81 0.17 0.6 0.43 -0.21 -0.13 0.08 -0.37 -0.91 -0.54
82 0.97 0.92 -0.05 -0.04 -0.07 -0.03 0.28 0.36 0.08
83 0.52 0.72 0.2 -0.49 -0.24 0.26 0.18 0.39 0.21
84 0.66 0.95 0.29 -14.17 -1.61 12.56 -18.67 -5.16 13.52
85 2.35 2.69 0.34 48.73 50.58 1.85 38.91 55.8 16.89
86 3.46 0.46 -3 1.49 -0.31 -1.8 1.25 -0.24 -1.5
87 0.78 1.23 0.45 -6.02 4.2 10.22 14.34 26.54 12.2
88 0.15 0.08 -0.07 -3.57 -2.8 0.77 -3.87 -3.04 0.83
89 5.7 3.34 -2.36 10.17 3.99 -6.18 7.89 7.47 -0.42
90 0.41 2.35 1.93 -4.67 9.21 13.87 -7.86 11.17 19.03
91 0.53 0.11 -0.42 -0.52 -0.49 0.03 0.02 -0.31 -0.34
92 4.01 1.13 -2.88 0.61 0.01 -0.6 0.52 -0.02 -0.54
93 0.33 0.35 0.02 -0.2 -0.08 0.12 -0.28 -0.28 0
94 0.2 0.16 -0.04 -2.62 -2.24 0.38 -0.08 -1 -0.93
55
 95 1.09 0.18 -0.91 0.14 -1.24 -1.38 0.42 -0.87 -1.3
96 2.02 0.64 -1.38 0.73 -0.14 -0.87 0.87 0 -0.86
97 0.17 0.2 0.04 -0.29 -0.26 0.04 -0.09 -0.36 -0.26
98 0 -1.15 0 1.15 -0.03 0 0.03
99 0 0.25 0.25 -7.25 -8.34 -1.1 -0.2 -2.77 -2.57
Positive sum 11.86 (40) 70.65 (59) 86.52 (32)
Negative sum -50.57(54) -70.65 (36) -86.51 (64)
Sum -38.69 0 0
Note: The positive variations are tinted.; the number of sectors that shows positive/negative
variations are shown in the parentheses.

Appendix 4 Histograms of the 6 RCA Indices for South Korea (averaged

over 1995-1997, and 2006-2008)
* The reference lines indicate comparative advantage neutral points.

40
30
15

Percent
Percent

20
10

10
5

0
0

0 2 4 6 8 -40 -20 0 20 40
BI averaged over 1995-1997, S.Korea LI averaged over 1995-1997, S.Korea
40
20

30
15

Percent
Percent

20
10

10
5

0
0

0 2 4 6 8 -60 -40 -20 0 20 40

BI averaged over 2006-2008. S.Korea LI averaged over 2006-2008, S.Korea
6

15
4
Percent

Percent

10
2

5
0

-1 -.5 0 .5 1 0 2 4 6 8
SI averaged over 1995-1997. S.Korea WI averaged over 1995-1997, S.Korea

56
 20
15
4
Percent

Percent
3

10
2

5
1
0

0
-1 -.5 0 .5 1 0 2 4 6 8 10
SI averaged over 2006-2008. S.Korea WI averaged over 2006-2008. S.Korea
30

30
Percent

Percent
20

20
10

10
0

-.05 0 .05 .1 .15 -20 0 20 40

AI averaged over 1995-1997, S.Korea NI averaged over 1995-1997. S.Korea
40

40
30

30
Percent

Percent
20

20
10

10
0

-.05 0 .05 .1 .15 -20 0 20 40

AI averaged over 2006-2008, S.Korea NI averaged over 2006-2008, S.Korea

57