HOW SIMILAR ARE THE EAST ASIAN ECONOMIES? A CLUSTER ANALYSIS PERSPECTIVE

ON ECONOMIC COOPERATION IN THE REGION

May 4, 2012

Alexandre Repkine

INTERCONNECTION AND COOPERA-TION

1997 Asian financial crisis Japanese statement about interest rates Depreciation of Thai baht Capital flight and currency depreciation in

Asia Economic cooperation

Expanding the set of consumer choices Institutional framework helping to absorb

shocks

WHY ECONOMIC COOPERATION?

Comparative advantage Production possibility frontiers Gains from trade Examples of economic cooperation

Free Trade Agreement Regional Trade Agreement Customs Union Currency Union

ECONOMIC COOPERATION IN EAST ASIA

Growth in the number of FTAs from 2 in 1975 to 16 in 2000

By 2010: more than 45 FTAs concluded in the region

SCOPE AND SEQUENCING OF FTA

The scope of East Asian FTAs is con-stantly increasing

The problem of sequencing How do countries form economic coopera-

tion groups? How do smaller groups become larger

groups? What is the principle governing the process

of economic agglomeration?

ROLE OF ECONOMIC SIMILARITY

It makes sense for the economically simi-lar countries to form groups of eco-nomic cooperation prior to doing so with more different countries E.g. Philippines first forming alliance with

Thailand before concluding an FTA with EU How do we measure similarity Why are groups of similar countries better

off economically?

WHY ARE SIMILAR COUNTRIES BETTER OFF TOGETHER?

Gravity models Helpman and Krugman (1985) Geographically close countries trade more Countries with similar-sized GDPs trade more Similar countries in general trade more

(Bergstrand and Egger, 2007) Baier and Bergstrand (2004)

Cooperation and trade between economically simi-lar countries increases welfare

Similarity measured in terms of distance, GDPs, remoteness to ROW, and K/L ratios

DEFINING ECONOMIC SIMILARITY

What countries can be viewed as simi-lar and on what grounds? Language (Korea vs Japan) Historical heritage (Korea vs China) Current trade and political links (Korea vs

US) Once groups, or clusters, of similar

countries are identified, economic in-tegration can be based on those clusters

AN OUTLINE OF CLUSTER ANALYSIS

Every country is a collection of characteristics GDP size Population Human development index Trade openness

How does one compare collections of numbers? Cluster analysis employs a measure of

generalized distance based on several charac-teristics

AN EXAMPLE OF SIMILARITY MEASURE: EU-CLIDEAN DISTANCE

X and Y are any two economic parameters (GDP per capita, % urban popula-tion)

EUCLIDEAN DISTANCES IN ASIA

Real GDP per Capita, USD

Sh

are

of

Urb

an

Pop

ula

-ti

on

, %

• Number of distances grows quickly with more countries added

• Vietnam appears to be similar to Cambodia

• Should we include Malaysia in one group with China and Thailand?

• Dissimilarity matrix sum-marizes the informa-tion about economic distances between coun-tries

DATA SOURCES AND SUMMARY STA-TISTICS

  Mean Stan-dard

Devia-tion

Min Max Source Computa-tion

Structural Shares            

Agriculture, % 18.35% 0.13 1.46% 4.53% ADB SU501/SU499

Manufacturing, % 23.21% 0.12 5.96% 48.46% ADB SU504/SU499

Trading, % 13.98% 0.05 5.76% 21.66% ADB SU508/SU499

International Trade            

Trade Openness, % 106.59% 46.84 24.31% 213.75% Penn openk

Economic Develop-ment

           

GDP per Capita, $ $10167 11010 $1707 $34223 Penn rgdpl

Share of Urban Popu-lation

44.10% 19.82 12.52% 84.68% ADB, Trading Economics

SU1223

Human Development Index, %

0.62% 0.14% 0.44 0.89% ADB SU1023

Economic Size            

Population, mn people 61.2 65.6 2.9 240.3 Penn POP

GDP in constant 2005 prices, bn USD

1229 2195 7.46 9276 ADB SU499Note: ADB stands for the Asian Development Bank’s statistical database https://sdbs.adb.org/sdbs/index.jsp), Penn for the Penn World Table version 7 (http://pwt.econ.upenn.edu/php_site/pwt70/pwt70_form.php). Data on urban population shares in Papua New Guinea is taken from the Trading Economics In-dicators database (http://www.tradingeconomics.com/papua-new-guinea/urban-population-percent-of-total-wb-data.html). The “Computation” column is based on the variable names provided by the original databases. Population statistics are given for the subsample that excludes China.

EAST ASIAN COUNTRIES COVERED

1. Mongolia2. Korea3. China4. Taiwan (China)5. Cambodia6. Laos

7. Papua New Guinea8. Vietnam9. Indonesia10. Malaysia11. Philippines12. Thailand

Japan is excluded because of its special status of the second largest economy in the world (until recently) and its currency being the only hard currency in the region.

CLUSTERING BY K-MEANS

Specify the number of groups in advance

Make sure that the overall distance within each clus-ter of the individual observations from the cluster center (i.e. centroid) is minimized

Proceed in iteration so that each country may change its cluster several times in the process

Random assignment of group centers initially

Realistic group membership: clusters of 2, 3, and 4 countries

K-MEANS CLUSTERING, EUCLIDEAN DISTANCE MEASURE

2 groups 3 groups 4 groupsGroup 1

ChinaIndonesia

Korea Korea KoreaMalaysia Malaysia

PhilippinesTaiwan Taiwan Taiwan

Thailand

Group 2Cambodia Cambodia Cambodia

Laos Laos LaosMongolia Mongolia Mongolia

Papua New Guinea Papua New Guinea Papua New GuineaVietnam Vietnam

IndonesiaThailand

Philippines

Group 3China China

Group 4IndonesiaMalaysiaVietnam

Philippines

Thailand

• Results similar to the case when Man-hattan (c-ity block) measure is used

• China forms one –country cluster

• Korea and Taiwan

• Cambodia, Laos, Mongolia, Papua New Guinea

• Indonesia, Philip-pines, Thailand

K-MEDIAN CLUSTERING, EUCLIDEAN DISTANCE MEASURE

2 groups 3 groups 4 groupsGroup 1

ChinaIndonesia

Korea Korea KoreaMalaysia Malaysia

PhilippinesTaiwan Taiwan Taiwan

ThailandGroup 2

Cambodia Cambodia CambodiaLaos Laos Laos

Papua New Guinea Papua New GuineaPapua New

GuineaVietnam  Mongolia Mongolia

  

Group 3China  

    Indonesia Indonesia    Malaysia Mongolia    Philippines Philippines    Thailand Thailand    Vietnam Vietnam

Group 4  China

• Mongolia joins more advanced group with Thailand and In-donesia in 4-group solution

• China is still forming a separate cluster in 4-group so-lution

• Korea and Taiwan still stick together

• Indonesia, Philippines and Thailand

HOW MANY CLUSTERS TO CHOOSE?

Both K-means and K-median clustering pro-cedures need a priori the number of clusters

Hierarchical procedures start with each coun-try being its own cluster, then agglomerating up

Stopping rules Pseudo-F value Duda-Hart value

AGGLOMERATION INTO CLUSTERS

Based on dissimilarity matrices (normally Euclidean distances)

Merging clusters that are similar Single-linkage Complete linkage Average linkage Cluster centroid Ward’s method

OPTIMAL NUMBER OF CLUSTERS

  Average Linkage

Single Linkage

Complete Linkage

Centroid Ward’ sMethod

Number of Clusters

F DH F DH F DH F DH F DH

25.33 0.6 5.33 0.92 5.33 0.58 5.33 0.79 4.47 0.46

36.99 0.55 2.98 0.75 7.3 0.41 4.26 0.56 7.3 0.41

48.94 0.54 3.19 0.68 9.13 0.47 6.62 0 9.13 0.47

59 0.33 3.91 0.64 9 0.33 5.02 0.77 9 0.48

68.34 0 4.86 0.68 8.34 0.48 4.86 0.49 8.6 0.33

77.76 0.48 5.36 0.82 8.93 0 7.76 0 8.93 0

89.54 0 4.62 0.58 9.54 0.23 6.72 0.65 9.54 0.23

Four clusters appears to be the op-timal solu-tion

Single link-age is ex-ceptional

Average number of clusters 4.2

CLUSTER SEQUENCING: DENDRO-GRAMS

4-cluster group-ing coincides with K-means solution (Euclidean distance)China staying sepa-rately

Korea and Tai-wan

Cambodia, Laos, Papua New Guinea

Mongolia ambiguous

CONCLUSIONS

Economic theory suggests similar countries should trade and cooperate more since such cooperation increases their total welfare

Cluster analysis determines what countries are similar and suggests two approaches

Composing clusters if number of groups known

Hierarchical approach (dendrograms)

Results surprisingly stable over various procedures

Results could be used as background for future policy mak-ing on regional economic cooperation in East Asia