Assessing the Fragility of Global Trade:

The Impact of Localized Supply Shocks Using Network Analysis

Prepared by Yevgeniya Korniyenko, Magali Pinat, and

Brian Dew1

PRELIMINARY AND UNCOMPLETE DRAFT

Abstract Anecdotal evidence suggests the existence of specific choke points in the global trade network

revealed especially after natural disasters (e.g. hard drive components and Thailand flooding,

Japanese auto components post-Fukushima, etc.). Using a highly disaggregated international trade

database we assess the spillover effects of supply shocks from the import of specific goods. Our

goal is to identify inherent vulnerabilities arising from the composition of a country’s import basket

and to propose effective mitigation policies. First, using network analysis tools we develop a

methodology for evaluating and ranking the supply fragility of individual traded goods. Next, we

create a country-level measure to determine each country’s supply shock vulnerability based on the

composition of their individual import baskets. This measure evaluates the potential negative

supply shock spillovers from the import of each good.

JEL Classification Numbers: C38, F14, F10, F42, D85.

Keywords: international trade, supply shocks, spillovers, network analysis.

Authors’ E-Mail Address: ykorniyenko@imf.org; mpinat@worldbank.org; brian.dew@american.edu

_____________________ 1

The authors would like to thank Vikram Haksar, Tamim Bayoumi, Murtaza Husain Syed, Camelia Minoiu, Christian Henn, Tito

Cordella, Tatiana Didier, participants to the IMF SPR seminar and to XXXVI Sunbelt Conference for suggestions and comments.

The views expressed here are those of the authors and not necessarily those of the IMF. All errors and omissions are our own.

2 Contents

I. INTRODUCTION................................................................................................................ 3

II. SUPPLY SHOCKS AND THEIR CONSEQUENCES FOR GLOBAL TRADE: A

REVIEW OF THE LITERATURE .......................................................................................... 4

III. EMPIRICAL METHODOLOGY ................................................................................... 6

1. The four components of product fragility ........................................................................... 6

A. Presence of central players .............................................................................................. 6

B. Tendency to cluster ......................................................................................................... 8

C. Relevance ........................................................................................................................ 9

D. International substitutability ............................................................................................ 9

2. Classifying overall product fragility ................................................................................. 10

IV. RESULTS AND ANALYSIS ......................................................................................... 10

1. Descriptive statistics ......................................................................................................... 10

2. Countries’ fragility and origin of risk .............................................................................. 13

V. VALIDATION OF THE METHODOLOGY ................................................................. 17

1. 2011 Japanese earthquake and nuclear disaster ............................................................. 17

2. 2011 Thailand Floods ....................................................................................................... 19

VI. CONCLUSION AND POTENTIAL APPLICATIONS .............................................. 20

REFERENCES ......................................................................................................................... 22

Annex I. Technical details on the definition of components ................................................ 25

Annex I. Descriptive statistics................................................................................................. 28

Annex II. Fragility maps over time ......................................................................................... 33

Figures: Figure 1: Detection of the presence of central players using network analysis ........................... 8

Figure 2: Detection of the tendency to cluster using network analysis ........................................ 9

Figure 3. Industry classification of products traded in 2014 ...................................................... 12

Figure 4. Sectoral classification of products traded in 2014 ...................................................... 12

Figure 5. Importers of risky products, 2014 ............................................................................... 15

Figure 6. Exporters of risky products and Risk of Supply Shock in 2014 ................................. 16

Figure 7. Network analysis of diesel engines ............................................................................. 18

Figure 8. Macroeconomic spillovers of importing risky products from Japan .......................... 19

Tables:

Table 1. Top 10 risky import products by their value in trade ................................................... 11

Table 2. Importer of risky products from fragile countries ........................................................ 14

Table 3. Selected risky products exported by Japan in 2010 ...................................................... 17

Table 4. Selected risky products exported by Thailand in 2010. ............................................... 20

3

I. INTRODUCTION

The IMF review of the role of trade (2015) identifies gaps in the current understanding

of trade spillovers to the domestic economy: “The spillovers of trade examined are mostly

macroeconomic and on the demand side (…). Supply-side shocks and their welfare effects could

be studied.” The examination of global trade networks helps to fill this gap by explaining how

localized supply shocks can be transmitted to other countries through trade and how these shocks

can generate additional domestic disturbance for importer countries. Our methodology seeks to

identify and measure supply shock spillover potential arising from fragility in the trade networks

of individual goods.

Network analysis provides the foundation for our novel approach to studying supply

shock risks in the highly interconnected global trade system. Trade-network-level analysis offers

a more complete exploration of the consequences of potential shocks than is possible with either

classical trade theory analysis at a country level, or bilateral analysis, such as gravity models.

New evidence1 finds that highly interconnected countries and industries are more vulnerable to

economic shocks. When a shock hits, countries with the most connected industries (or those that

are heavily involved in the global value chain) are more likely to experience disruptions in

production, in particular without concerted government efforts. At the same time, highly

interconnected countries that produce easily substitutable goods are better positioned to

withstand disruptions in trade2.

We assess the sensitivity of import baskets to potential supply shocks3 based on

individual characteristics of the component imported goods using network analysis tools. In this

paper, we develop a methodology to evaluate the riskiness of more than 3340 products. In

contrast to existing assessments of vulnerabilities and risks in global trade based on individual

product characteristics (quality, cyclicality, complexity, etc.), our methodology breaks new

ground by examining the network properties of individual goods. In particular, it underscores

the riskiness arising from the presence of central players in the network of a product, from the

tendency to cluster, as well asfrom low international substitutability.

The methodology identifies the most vulnerable products in global trade and tracks top

exporters and importers of these products. The methodology also allows the benchmarking of

potential import basket vulnerabilities against different countries, country groups, and across

regions, and provides a new dataset that could be utilized for cross-country analysis.

1 IMF 2013, Acemoglu et al. 2015; Contreras and Fagiolo 2014. 2 OECD 2013; UNCTAD 2013 ; de la Torre et al. 2015. 3 At the moment, we conduct our analysis of localized supply shocks assuming constant demand. While it is a

suitable assumption, it is not always realistic. During global financial crisis, there were a number of examples (see

Baldwin (2009), Bems (2010), Freund (2009), WB Global Economic Prospects, Jan 2015), when exporters and

importers were facing collapse in external and domestic demand and had to adjust their trading flows accordingly.

This might complicate the analysis of supply-shocks propagation, as both demand and supply adjust simultaneously.

4

The paper proceeds as follows: section II provides a review of the literature on localized

supply shocks and their consequences for global trade, and on trade and network analysis.

Section III sets the analytical framework, describes the four components of product vulnerability,

and discusses the method of classifying a product’s overall fragility. Section IV describes and

analyzes the results. Section V provides a validation of the methodology based on two case

studies, while potential applications and extensions of the research are discussed, and concluding

remarks are presented, in section VI.

II. SUPPLY SHOCKS AND THEIR CONSEQUENCES FOR GLOBAL TRADE: A

REVIEW OF THE LITERATURE

This paper contributes to the literature on international trade and networks. Recent

research suggests that microeconomic shocks may be useful for explaining aggregate volatility.

Analogously to Carvalho (2014), our idea is that the structure of a production network is key in

explaining whether and how microeconomic shocks propagate throughout the economy and

impact output4,5. Understanding the trade network structure can better explain the origins of

aggregate fluctuations and how to prepare for and recover from adverse shocks that disrupt

production. We study network characteristics of individual products to identify critical goods

that can expose a country to a supply shock from abroad. Our results suggest that the transmission

of various types of supply shocks through economic networks and industry interlinkages could

have first-order implications for the macroeconomy.

First, our work refers to the consequences of a temporary disruption of inputs for the

importing countries. Nguyen and Schaur (2012) find that firms that import goods channel foreign

input prices volatility to the domestic market. Bergin et al. (2011) show that offshoring industries

in Mexico experience fluctuations in employment that are twice as volatile as the corresponding

industries in the U.S6. Also, the development of Global Value Chains (GVCs) in the last two

decades is key in understanding the consequences of supply shocks, as countries become more

interconnected and dependent on each other. Carvalho, Nirei, and Saito (2014) examine firm

level data before and after the 2011 Great East Japan Earthquake to quantify the spillover effect

of exogenous shocks through the supply chain. Firms that were indirectly linked, even with two

or three degrees of separation, were found to be affected, establishing that supply-chain linkages

constitute a powerful transmission mechanism of otherwise localized shocks. The recent

contributions of di Giovanni and Levchenko (2010) and Johnson (2014) show that comovement

across countries is potentially the result of the international transmission of shocks through

4 Shocks that affect a particular firm or technology along the chain. 5 Similarly, Acemoglu, Carvalho, Ozdaglar, and Tahbaz-Salehi (2012) argue that in the presence of intersectoral I-

O linkages, microeconomic idiosyncratic shocks propagate through a network with higher-order interconnections

and cause “cascade effects” to the aggregate economy. The network’s structure determines the rate at which the

volatility decays. Acemoglu et al. (2015) find that the network-based propagation is larger than the direct effects

of the shocks. 6 In the macroeconomic literature, evidence on the relationship between trade and macro volatility is mixed: some

studies, most notably Easterly, Islam, and Stiglitz (2000), and Kose, Prasad, and Terrones (2003) argue that trade

openness increases volatility, while others, including Haddad, Lim and Saborowski (2010), Cavallo (2008), and

Bejan (2006) show that trade openness decreases volatility. Di Giovanni and Levchenko (2008) exploit variation

across countries and across sectors, concluding that trade openness leads to higher volatility of output. This result

is similar to Newbery and Stiglitz (1984), who showed that in an open economy, an industry is more vulnerable to

world supply and demand shocks.

5

vertical linkages and input-output networks. We aim to go deeper in the understanding of why

imports of specific goods from certain trading partners might be at the origin of volatility.

The second strand of the literature our work refers to is related to how trade is adversely

affected by temporary supply shocks. A natural disaster such as a hurricane or earthquake, an

armed conflict, or political turmoil can temporarily create negative supply shocks due to

destroyed production facilities, disruptions in transportation, or reduced human capital. Delays

to trade can additionally transmit negative supply shock effects7. Oh and Reuveny (2009) find a

large and persistent impact of climatic natural disasters on trade, national income, and global

economic welfare. The impact is particularly pronounced for small developing countries

according to Andrade da Silva and Cernat (2012). Similarly, Gassebner et al. (2011), using cross-

country analysis, find significant impact of natural disasters on trade, particularly for small

countries and autocracies. Besedes and Murshid (2014) examine the eruption of the Icelandic

volcano, Eyjafjallajökull, and find that it negatively impacted exports from the affected countries

to the U.S. and Japan. Volpe Martincus, and Blyde (2013) study the effect of a Chilean

earthquake in 2010 on export volumes and find that diminished transportation infrastructure had

a significant negative impact on firms' exports. At the same time, Escaith et al. (2011) found that

the effect of the earthquake in Japan in 2011 on global trade was relatively small and short-lived,

despite the devastation in Japan8.

In the same vein, conflicts and wars may reduce trade flows by raising costs to private

agents of engaging in international trade and investment. Long (2008) finds that external and

internal violent conflicts reduce trade. Blomberg and Hess (2006) estimates suggest that violent

conflict is a greater barrier to trade than traditional tariff barriers. Martin, Thoenig and Mayer

(2008) findings suggest that a year after a conflict a country's trade is reduced by 25 percent

relative to its trade in the absence of a conflict. Glick and Taylor (2010) study the effects of war

on bilateral trade with available data extending back to 1870 and find that in war losses to neutral

nations are of the same order of magnitude as losses to belligerents. Similarly, a number of

empirical studies find an impact of political instability on individual sectors. For example,

Muhammad et al. (2011) find evidence of a structural change in the import growth rate for

Kenyan roses to the EU, after the 2007/08 post-election violence and political instability in

Kenya, which is approximately equivalent to an 18.6% tariff. This paper aims to find a channel

through which localized temporary supply shocks spillover to the global economy.

Lastly, the novelty of our work is in the use of network analysis tools to assess the

fragility of global trade. Our approach based on seminal work of Newman (2005) that outlines

network analysis tools and techniques including the clustering coefficient and Jackson (2010)

who provides a useful guide to representing and measuring network structures and their

implications. Recently network analysis tools have been used to analyze global trade. For

example, De Benedictis, Nenci, Santoni, Tajoli, and Vicarelli (2014) use the BACI-CEPII

database to identify various measures of local and global centrality at a product level.

7 Djankov, Freund, and Pham (2010) estimate that an additional day spent prior to shipment reduces trade by more

than 1 percent. Similarly, Hummels and Schaur (2013) estimate that each day spent in transit costs a firm 0.6–2.1

percent of its shipment’s value. Such delays are extremely costly because they impose significant inventory-

holding costs, and may destroy perishable goods as well as goods that have seasonal demand. 8 For more research on the transmission of natural disasters, such as the 2011 Japanese earthquake, through the

global input-output network see Boehm, Flaaen and Pandalai-Nayar (2014), Barrot and Sauvagnat (2014), and

Carvalho, Nirei and Saito (2014).

6

Nevertheless, and to the best of our knowledge, no paper is assessing the systemic risk of the

country import basket using network analysis tools and detailed information on traded goods.

III. EMPIRICAL METHODOLOGY

Our research uses detailed UN COMTRADE and WITS bilateral trade data9, based on

the harmonized system 2002 (HS2002) classification at the 6-digit level, for the period 2003-

2014. There are in total 5223 products and an average of 146 countries per year in the database.

The bilateral export and import flows in the database are homogenized using the Feenstra et al.

(2005) methodology. Final goods and consumption goods, defined using the UN BEC

classification, are excluded from the dataset in order to focus exclusively on products that are

used by other industries and have the potential for negative spillovers10. In addition, from 2007

onwards some countries started submitting data using only the new HS2007 classification. We

used correspondence tables to merge the HS2002 with the HS2007 classification and excluded

all the products for which there were no perfect match11. Finally, we drop the two products that

refer to the crude oil and refined oil category. Our sample shrinks to 3340 products after the

cleaning process.

Using this database, we identify four characteristics for each product that contribute to

its vulnerability to potential supply shocks: (i) presence of central players; (ii) tendency to

cluster; (iii) relevance; and (iv) international substitutability. To our knowledge, we are the first

to suggest this approach to identify risky products in global trade and to measure the vulnerability

of country import basket to supply shocks based on its’ composition. Motivated by this gap in

the literature, the main aim of this paper is to construct a measure that is relevant for analyzing

inherent vulnerabilities of countries’ import baskets using individual product characteristics

1. The four components of product fragility

For each year and product in the sample, we calculate the fragility of the product based

on the following four components:

A. Presence of central players

The first characteristic identified as important for the analysis of risky products is the

presence of central players in the network of traded goods. The presence of large central players

has a role in the extent to which microeconomic shocks explain aggregate fluctuations (Gabaix

9 Data available at a highly disaggregated level are not split by value-added. This raises the possibility that the

network of a product is misrepresented because certain countries add very low value in the process of production

(for example, only pieces are assembled). This is a common criticism to work on trade networks, such as the seminal

work of Haussman and Hidalgo. Our approach is not totally subject to this criticism. Even if a country is only

assembling a good (and not producing it) it is still in the trade network of the product. If the “assembling” country

is hit by a temporary shock, it will (at least in the short term) lead to a supply shock for importing countries.

10 These are 652 goods consumer goods and 429 foods and beverage goods mainly for household consumption.

11 There are 753 products dropped during the process of merging HS2002 and HS2007 classifications; two

products are dropped due to their classification in miscellaneous products and nice categories 999999 and

9999XX.

7

(2011))12. Using network analysis measures of centrality we identify products with exporters so

integral that a shock to their supply may disrupt importers’ production. In network analysis

terms, we are looking for a network that represented by a star shape (Figure 1a), as opposed to a

fully connected network (Figure 1b), as star-shaped networks are riskier from the importer point

of view.

We use the standard deviation of outdegree centrality as our measure of this

characteristic13. First, we calculate the outdegree centrality of each player for each good network.

As detailed in Annex IA, outdegree centrality measures the number of partners a country exports

to as a share of the maximum number of partners in the product network. Countries with many

partners are more likely to generate negative spillovers in case of a negative supply shock. They

are often characterized as influential. Star shaped networks are characterized by the presence of

few central players. We then use the standard deviation of outdegree centrality to measure each

product’s risk of having a few very central exporters; the higher the standard deviation, more

likely the star shape and the higher the potential risk14.

12 In networks where the largest firms contribute disproportionately to aggregate output, shocks to these firms

contribute to aggregate fluctuations. Similarly, Carvalho (2014) uses the Katz–Bonacich measure of centrality,

which assigns to each sector a centrality score that is the sum of some baseline centrality level (equal across sectors),

and the centrality score of each of its downstream sectors, defined in the same way. 13 Variants of this measure have been deployed in the sociology literature, notably Bonacich (1972) and Katz (1953),

in computer science with Google’s PageRank algorithm (Brin and Page, 1998), or in social networks literature

within economics (for example, Ballester, Calvo-Armengol, and Zenou, 2006). 14Due to data availability, we are not able to use more disaggregated information (such as, for example, HS 10-digit

classification or firm level data). A potential concern is the loss of precision in identifying products, as many very

similar products would be aggregated to the 6-digit level. There is a possibility that two products at 10-digit

disaggregation level would have very different networks (for instance, one being very risky and the other not), and

that the aggregation at 6-digit would be misinterpreted. Nevertheless, this would only underestimate the risk

associated to the centrality component. Two 10-digit good networks from the same 6-digit category: one star-shaped

and the other fully connected, will be fully connected at the 6-digit network. Star-shaped networks that we detect at

the 6-digit level effectively only include star-shaped networks at the 10-digit level. We can then affirm that all the

categories that are detected as risky contain actually risky products. Case studies reinforce this view.

8

Figure 1: Detection of the presence of central players using network analysis

Panel A. Star-shape network Panel B. Fully connected network

Notes: Nodes with letters represent countries. The ties that link the nodes show the direction of a trade

of a product (either exports/imports, or both). In panel A, the outdegree centrality of node A is equal

to one, as country A is exporting to all the countries in the sample. In contrast, the outdegree centrality

of countries B, C, D, E, and F is equal to zero, as they are exporting to zero countries out of the five

possible. The standard deviation of this network is 0.45. In panel B, all of the countries are trading

with all other countries in the network. All countries in this fully connected network have an outdegree

centrality of one, so the standard deviation is zero.

B. Tendency to cluster

Another characteristic of a product network that increases potential spillover risk is the

tendency of countries to cluster—the tendency of groups of countries to trade more among each

rather than with the rest of the world15. Figure 2 demonstrates a network with the tendency to

cluster. Risk emerges if a cluster is destabilized (for example, after a supply shock to its most

central country) as the probability of importers in the cluster finding a new supplier is lower than

in product networks where all countries are highly connected (networks with only one cluster).

As detailed in Annex IB, standard algorithms to detect clusters in the network analysis

literature are not applicable to trade data. We use two characteristics from cluster analysis to

detect products for which countries have tendency to cluster: the weighted average local

clustering coefficient and the network diameter. The weighted average local clustering

coefficient quantifies how close the partners of a country are to others. In other words, what is

the likelihood of the trade partners of a particular country for a particular good also trading the

same good among each other. The higher the clustering coefficient the higher the tendency of

countries to cluster.

We multiply this measure with the diameter of the product network. The diameter of a

network is the size of the longest path–the maximum number of steps that separate the two most

distant countries. If a country belonging to a cluster needs to find a new provider, it will be easier

to connect to a country in a close cluster (i.e. a cluster already connected to other countries in

the clusters).

15 For the relevance and presence of clusters in trade networks, see Fagiolo et al. (2009), Fagiolo, Reyes and Schiavo

(2010), Ward, Ahlquist and Rozenas (2013); for the presence of cluster in finance networks, see Hattori and Suda

(2007) for the network of international bank exposures, Kubelec and Sá (2010) and Sá (2010) for different asset

class and Miniou and Reyes (2011) for syndicated loans network.

A

F

E

D

C

B

A

F

E

D

C

B

9

Figure 2: Detection of the tendency to cluster using network analysis

Notes: This network is a typical representation of a tendency to cluster. The weighted average of the

local cluster is high (equal to one on a scale going from zero to one) and the diameter is equal to 11.

Our measure for tendency to cluster for this particular product is then 1x11=11 (on a scale going from

0 to the maximum value of the parameter16).

C. Relevance

The third component in our methodology is the relevance of the product to global trade

and production. We look for systemically relevant products, with a potential supply shock effect

on many countries. To characterize relevance, we use the average share of a product in world

imports (measured as the average share of all countries’ imports). Products that comprise a larger

share of trade are considered risky due to the increased magnitude of effects from a trade

disruption. Annex IC provides details of the equation.

D. International substitutability

The final component is the degree of international substitutability of the product. The

idea is based on the assumption made by Paul Armington in 1969 that products traded

internationally are differentiated by country of origin. As such, when a shock hits major suppliers

the extent of spillovers will depend on the availability of goods’ substitutes on international

markets. If there are no close substitutes in the short run, every user is affected by the

disturbances at the source country17. We do not have exact information on the Armington

elasticity18 for each product, therefore we proxy it with an indicator inspired by the Revealed

16 The maximum value of the diameter in our case is 17 - the max of the value of diameter of a network for each

good between 2003 and 2014. 17 For example, Tanaka (2012) finds that some Japanese auto parts are less substitutable, which led to the disruptions

throughout the global supply chain for the auto industry after the earthquake in Japan in 2011. 18 The estimates vary significantly depending on the method of estimation and data used. Aspalter (2016), Feenstra

(2014), Saito (2004) provide some estimates of simple Armington elasticity using both bilateral and multilateral

trade data. Additionally, pioneering work by Feenstra, Luck, Obstfeld, and Russ (2014) allows further

differentiation between a macro Armington elasticity of substitution between domestic and imported goods and a

micro Armington elasticity between different import sources. Their empirical work highlights differences in these

micro and macro elasticities. In particular, they find that the macro elasticity is significantly lower than the micro

elasticity for up to one-half of the goods considered, relying on both simulation studies and highly disaggregated

U.S. data.

10

Factor Intensity (RFI) developed by Shirotori, Tumurchudur, and Cadot (2010). We are

particularly interested in the level of human capital of each exporter country and its’ distribution

for each product. In the case of a temporary supply shock, the importing country will look for

alternative suppliers with similar characteristics to those who provided the temporally

unavailable good. The “wider” the distribution of human capital of exporting countries, the more

heterogeneous the available production methods are for a product. This heterogeneity

complicates international substitutability, as a country’s substitute supplier must comply with its

standard of production. Combined with the three other components (influential players,

clustering, and relevance), low international substitutability adds to the vulnerability of

imports19.

2. Classifying overall product fragility

Identifying which products are risky at 6-digit disaggregation level can help to track

importers’ vulnerability to supply shocks from abroad and exporters’ potential to originate

important negative spillovers from natural disasters, political instability, and conflicts. The

methodology classifies a product as risky if it scores high in each of the four components

described in the previous section. To classify products in different groups we use cluster analysis

(the k-median procedure) which is applied to the standardized scores to group the products into

risk categories. From the partition exercise we obtain a cluster for which the value of each

component is high: this is the group of products that we will consider as risky20,21. After

categorizing products, we can track the importers and exporters of risky products by looking at

the risky-product share of total imports or exports in a county’s trade basket.

IV. RESULTS AND ANALYSIS

1. Descriptive statistics

We applied the methodology described in the previous section to the bilateral trade

database for each year from 2003–2014 for 147 countries22. We obtained for each year 7 product

groups of different levels of risk (see Annex II Table A.1. Group 7 is considered to be the

riskiest.). Over 2003-2014, an average of 359 products are classified in Group 7, and 158 are

consistently classified in Group 7 in each of the twelve years of the sample (see Annex II Table

A.2.). In Table 1, we present the ten risky products with the highest share of trade by value of

imports. Products identified as risky belong mainly to two broad sectors: machinery and

mechanical appliances (6-digit HS codes starting with 84 and 85) and transport equipment (HS

codes 87 and 88). Other sectors that are overrepresented in Group 7 are plastic articles (HS codes

starting with 39), articles of iron and steel (73), and rubber articles (40).

19 For more details see Annex I.D. 20 Note that the partition is not hierarchized, but one group emerges naturally maximizing the value of each of the

component. 21 More details are in Annex I.E. 22 The full HS2002 6-digit classification dataset contained bilateral trade data for 187 different countries for at least

one year during the period 2003–2014. Countries with fewer than six years of available data also contain, on average,

fewer than 1100 products per year, compared with an average of more than 2700 products per year for countries

with six or more years of data. To improve consistency over time, 26 countries with fewer than 6 years of data have

been removed from the dataset.

11

Table 1. Top 10 risky import products by their value in trade

Notes: The products shown in the table are consistently classified as risky (cluster Group 7) over 2003-

2014. The ranking is by their value in imports. The full list of products constantly classified as the risky

over time can be found in Annex Table A.2.

To better understand the products that are defined as risky we group them by industry

and by sector (see Figure 3 and Figure 4). Top bar in Figure 3 shows the total number of products

for each industry for the full sample of 3360 products, while the bottom bar shows only the

products belonging to the risky group. Looking at Figure 3, we observe that processed industrial

supply and capital goods are the categories of products most represented in global trade.

Interestingly, the comparison between the top and bottom bars shows a different picture of the

relative importance. The accessories of capital goods represent only 15% of products in the full

sample (top bar), whereas their share more than doubles in the risky group of products,

representing a 31% of the total. Parts and accessories of capital goods represent 6% of the full

sample and more than 22% of the products in the risky group. In contrast, products in the

processed industry category are under-represented in the risky group; the category comprises

almost 60% of the full sample but only 34% of the risky group.

HS2002_6d

Product Description Share

Value of

Imports

1 870323 Cars # Of a cylinder capacity exceeding 1,500 cc but not exceeding 3,000 cc 2.777%

2 870332 Cars # Of a cylinder capacity exceeding 1,500 cc but not exceeding 2,500 cc 1.485%

3 847170 Computers # Storage units 1.033%

4 870322 Cars # Of a cylinder capacity exceeding 1,000 cc but not exceeding 1,500 cc 0.791%

5 870829 Vehicle Parts # Other 0.678%

6 880330 Aircraft Parts # Other parts of aeroplanes or helicopters 0.671%

7 854140 Semiconductor Devices # Photosensitive semiconductor devices, including photovoltaic cells

whether or not assembled in modules or made up into panels; l ight emitting diodes

0.647%

8 850440 Electrical Transformers # Static converters 0.591%

9 870421 Delivery Trucks # g.v.w. not exceeding 5 tonnes 0.521%

10 853400 Printed Circuit Boards # Printed circuits. 0.503%

12

Figure 3. Industry classification of products traded in 2014

Notes: Classification corresponds to the Broad Economic categories (BEC).

Another way to analyze what kind of products are defined as risky is to group them by

sector. Figure 4 shows sectoral classification of the products. Similar to Figure 3, the top bar of

Figure 4 presents the sectoral composition of goods in the full sample of 3360 products while

the bottom bar shows the sectoral composition of only the risky goods. Comparing both panels,

we observe the overrepresentation of mechanical appliances and electrical equipment in the risky

category. While their share is around 18% of the full sample, mechanical appliances and

electrical equipment comprise more than 41% of the risky group.

Figure 4. Sectoral classification of products traded in 2014

Notes: Classification corresponds to the HS 2002 2-digit sector classification.

13

2. Countries’ fragility and origin of risk

After identifying risky products from a supply shock perspective, we are able to

determine which countries are importing and exporting them. Figure 5 shows the share of total

imports that is potentially vulnerable to supply shocks for countries in 201423. On average, 21%

of total imports are risky, but this figure hides an important variance across countries. Oman has

the highest share (33%) of risky products in its import basket. Eastern Europe is also relatively

vulnerable with Slovakia, Hungary, and the Czech Republic holding import baskets comprised

in 2014 of more than 30% risky products. North America is particularly vulnerable according

this ranking; Mexico is the 3rd most vulnerable importer of risky products in the world (in term

of its share of imports), Canada the 6th, and the US 17th. Zimbabwe (with 30.6%) and Australia

(with 30.1%) are also disproportionally exposed. At the other end of the spectrum, many East

Asian countries appear relatively less exposed to spillovers from risky products. China, Japan,

Korea, the Philippines, and Indonesia all have a below-average risk of importing fragile products.

Figure 6 shows the intensity of exports of risky products by countries. On average, risky

products comprise 12.5% of total exports, however, variance by country in risky products’ share

of exports is more than twice as wide as in risky products’ share of imports. Much of Eastern

Europe is in the top 15 risky goods exporters: Hungary is 1st, Czech Republic 2nd, Slovakia 5th,

Slovenia 7th, Romania 9th, Austria 10th and Poland 13th. Associated with Germany, they belong

to the central-eastern Europe value chain and appear to be key players in the exports of risky

products. Many advanced economies are at the top of this ranking. After Germany (3rd), we find

Japan (4th), United Kingdom (11th), Spain (12th), the USA (15th), Italy (16th), Sweden (18th), and

France (19th). The top 20 countries, on average, have a share of risky products exports of more

than 31%. At the other extreme, we find less open markets such as Sub-Saharan Africa and Latin

America with the share of exports of risky products close to zero. Exports of risky products

appear to be highly correlated to economic development: 22 out of the 30 top exporters of risky

products are classified as high-income countries24.

As discussed in the previous section, if there is no supply shock, the importation of risky

products does not present an issue for countries. To further examine spillover potential, we can

consider information on the potential for localized supply shocks within each country. In

Figure 6, we combine the share of risky goods in each country’s exports with external data from

Maplecroft on political risk and risk of natural disaster. A higher intensity of exports of risky

products (darker blue area) associated with a risk of supply shock (red for political risk and/or

green for natural disaster risk) moves the supplier country into a higher risk category from the

perspective of importer countries.

For example, the Philippines face a political and natural disaster risk, Thailand a political

risk, and both have an above average share of exports of risky products. Table 2 shows the top

ten partner importers of risky products from the two countries. Following our methodology, these

countries are particularly exposed to a temporary disruption of supply. For instance, 2.7% of

total imports of Australia are risky products imported from Thailand. That is, a political or natural

disaster shock in Thailand may result in domestic spillovers for Australia. Note that Australia is

23 Annex III present maps over time. 24 The World Bank definition, high-income economies corresponding to countries with GNI per capita of $12,736 or more.

14

additionally exposed to imports of risky products from the Philippines, another exporter of risky

products with high political and natural disaster risk.

Table 2. Importer of risky products from fragile countries

Thailand Philippines

Australia 2.7% Japan 0.7%

Philippines 2.2% Thailand 0.4%

New Zealand 1.9% Palau 0.4%

Nicaragua 1.7% Hong Kong 0.3%

New Caledonia 1.5% Australia 0.3%

Moldavia 1.2% China 0.3%

Vietnam 1.2% Central African Rep. 0.2%

French Polynesia 1.2% Mexico 0.2%

Yemen 1.0% Singapore 0.2%

Pakistan 1.0% USA 0.2%

Notes: Figures show the share of risky products imports of countries from Thailand and the

Philippines.

15

Figure 5. Importers of risky products, 2014

Note

s: L

ate

st d

ata

ava

ila

ble

is

use

d f

or

this

map, 2014.

Sourc

e: A

uth

ors

’ ca

lcula

tion b

ase

d o

n U

NC

om

tra

de

da

tab

ase

.

16

Figure 6. Exporters of risky products and Risk of Supply Shock in 2014

Note

s: L

ate

st d

ata

ava

ila

ble

is

use

d f

or

this

map:

2014,

for

the

share

of

import

s of

fragil

e pro

du

cts

an

d 2

01

5Q

3 f

or

cou

ntr

y ri

sks.

Ris

ks a

re

dis

pla

yed

only

if

the

info

rma

tion

on

the

data

for

export

s of

fragil

e pro

duct

s is

ava

ilable

. T

he

risk

is

con

sid

ered

hig

h (

an

d d

isp

laye

d o

n t

he

map)

if i

n t

he

pro

du

ct i

s in

th

e to

p t

hir

d o

f th

e sa

mple

. Sourc

e: a

uth

ors

calc

ula

tion b

ase

d o

n U

NC

om

trad

e da

tab

ase

an

d M

aple

croft

da

ta.

17

V. VALIDATION OF THE METHODOLOGY

The methodology is validated with two case studies based on recent adverse events:

2011 Japanese earthquake and nuclear disaster

2011 Thailand floods

1. 2011 Japanese earthquake and nuclear disaster

On March 11, 2011, a magnitude 9.0 earthquake struck 70 km off the eastern coast of

Japan. The earthquake and resultant tsunami killed and injured more than 21,000 people.

Property destruction was enormous with 125,000 buildings totally collapsed and over a million

damaged. Manufacturing facilities in many parts of the country were damaged or destroyed. The

natural disasters were followed by electricity shortages which increased the affected zone and

further exacerbated the effect on manufacturing. The economic toll was steep, with 2011 GDP

growth figures 2 percentage points below their March 2011 forecast.

The economic effects of the disasters spread throughout the world through trade and

global supply chains in the period following the earthquake, particularly impacting the Asia

region. Damage to manufacturing in Japan had been amplified by the riskiness of several key

products for which Japan plays a central role in world production. The following products were

affected strongly by the disasters: diesel engines, power supply and aluminum capacitors, and

LCD screens used in TV sets, notebook computers, smartphones, and tablets (See Table 3).

Table 3. Selected risky products exported by Japan in 2010

Note: Column 1 of Table 2 present the HS2002-6digit classification of products identified as having

disrupted production in other countries after the 2011 disasters. The category of the product is

described in columns 3 and the precise description of the good is shown in italics. Column 2 presents

the level of risk found with our methodology, 7 being the highest risk category. Out of eight products

identified, the methodology categorizes six in the highest risk group one year before the event. Columns

4-7 detail the standardized value of each component.

Exports of diesel engines by Japan decreased by nearly 20 percent in 2011, as

manufacturers were not able to supply parts. A major French automobile manufacturer, for

example, in turn delayed the launch of two car models and was eventually forced to source from

another supplier. Diesel engines are considered a risky product by our methodology. They are

produced by “central players,” have clusters in their trade network (as shown in Figure 7, the

Japanese cluster for diesel engines disappears entirely in 2011), are highly systemically relevant,

and are not easily substitutable on international markets. The resultant choke point predicted by

our methodology proved problematic for importers following the supply shock.

HS2002 6-digit

Risk

category

in 2010

Products Description

(sometimes shortened)

z-score

Comp. 1

z-score

Comp. 2

z-score

Comp. 3

z-score

Comp. 4

Japan Case Study - 2011 Earthquake and Nuclear Disaster

840890 7 Combustion Engines # Other engines 1.05 1.26 1.20 1.44includes diesel engines

853229 7 Electrical Capacitors # Other 1.44 -0.14 0.73 1.37includes power supply capacitors and aluminium capcacitors

901380 4 LCDs # Other devices, applicances and instruments 1.30 1.37 0.79 0.71includes LCD screens in TV sets, notebook computers, smartphones, and tablets

18

Figure 7. Network analysis of diesel engines

2010 2011 2012

Note: This Figure shows trade networks for product 840890 from the HS2002 classification – diesel

engines. We use the community detector of Rosvall and Bergstrom (2008) to show the evolution of the

network for this product between 2010 and 2012. Countries are grouped by similarity of their trade

matrix and the country with the highest Page Rank centrality is displayed below the node25. The bigger

the share of trade of the countries in each group, the bigger is the node. Above some links, the share

of total imports of the destination country from the country at the origin of the arrow is displayed (e.g.,

in 2010, the USA imports 46% of their diesel engines from Japan). On the top of some nodes, the value

of imports is displayed (e.g., in 2010, the US had imported 2.8 billion US of diesel engine). The figure

shows only the links that reflect the structure of the network. In 2010, Japan is a key player, exporting

to both the US and Chinese clusters. In 2011, the year of the Fukushima accident, the Japanese cluster

disappears. Korean cluster seems to reinforce exports to China, but not to the US. In 2012, the

Japanese cluster is back and the network is more connected than before the earthquake. In 2011, a

Belgium and Netherlands simple connection and Iran are omitted from the algorithm-generated

graphic for simplicity of presentation.

Small parts can also cause disruption in production and carry outsized trade risks. Tiny

capacitors and resistors are critical to global electronics supply chains and Japan is a major

producer of these products. Following the earthquake, prices of the tiny inputs increased, and in

importer countries, production of various electronics and automotive parts that used the

capacitors slowed. Aluminum capacitors, which were highlighted in the literature, are included

in a product grouping with very similar risk characteristics to the diesel engines. An additional

risky product, the LCD screens used in many modern devices, was affected by the disaster and

has similar risk characteristics, but was not categorized as high-risk.

The 2011, the Japanese earthquake had significant negative consequences on its trade partners. We show in Figure 8, the correlation between the reliance on risky products from Japan and the negative revisions of GDP growth. Figure 8 panel A plots the value of actual GDP growth in 2011 versus its latest WEO forecast in 2010 on the y-axis and the share of imports from Japan in 2010 on the x-axis. The relation is slightly negative. In panel B, the x-axis is replaced by the share of imports of risky goods from Japan, keeping y-axis unchanged. The relation is even more negative and significant, which is confirmed by the regression presented Table 3.

25 The PageRank algorithm defines the centrality as the popularity of a node, i.e. the more central is a country the

higher likelihood a trade connection goes through it. Compared to the outdegree centrality we are using in the paper,

this algorithm uses recursive equation to compute the centrality.

19

Figure 8. Macroeconomic spillovers of importing risky products from Japan

Panel A. share of imports from Japan and the revisions of 2011 GDP forecast

Panel B. share of imports of risky products from Japan and the revisions of GDP

forecast in 2011

Notes: Equation of the fitted line is 𝑦 = −5.5𝛽 − 3.8∗∗∗ for panel A and 𝑦 = −44.9∗∗𝛽 − 3.4∗∗∗ for

panel B.

2. 2011 Thailand Floods

The 2011 monsoon season in Thailand brought severe flooding to 65 of Thailand’s 77

provinces, causing more than 815 deaths and 45 billion USD of property damage. Triggered by

a tropical storm at the end of July, floodwaters spread throughout the northern parts of the

country and through the Mekong and Chao Phraya rivers, eventually reaching Bangkok. Efforts

made to protect the capital city from the floodwaters were successful to varying degrees.

Industrial estates and manufacturing facilities were badly flooded in many parts of the

country, causing production and exports to be adversely affected. Literature on the economic

effects of the floods mentions some specific products which had seemingly outsized

consequences on global trade and supply chains. Three such products were hard disk drives,

semiconductors, and pick-up trucks. All three of these products are listed as risky according to

our methodology; two of which display all four of the risky characteristics.

Thailand is the world’s second largest producer of hard disk drives, which serve as the

“long-term” memory and file storage in desktop and laptop computers, tablets, and smart phones.

When factories which produce these hard drives were flooded, exports decreased, prices

increased (almost doubling and remaining elevated for two years), and production of the

electronic devices which use these intermediate goods slowed in many countries. Hard disk

drives are a risky good according to our methodology, which exhibits particularly high levels of

out degree centrality and clustering.

Flooding also damaged the manufacturing equipment used to produce semiconductors

and pick-up trucks. Suspension of the production of pick-up trucks in Thailand has economic

impact in Japan, whose automotive companies produce in Thailand, and in the countries from

which source component orders are suspended. Similarly, suspended semiconductor production

in Thailand slowed other countries’ production of goods for which semiconductors are a

20

subcomponent. Semiconductors and integrated circuits are valuable exports for which Thailand

is a fairly central player, and also register above average in three areas of trade risk (see Table 3

for the correspondence with our methodology).

Table 4. Selected risky products exported by Thailand in 2010.

Note: Column 1 of Table 2 present the HS2002-6digit classification of products identified as having

disrupted production in other countries after the 2011 event in Thailand. The category of the product

is described in column 3 and the precise description of the good is shown in italic. Column 2 presents

the level of risk found with our methodology, 7 being the highest risk category. Out of eight products

identified, the methodology categorizes six in the highest risk group one year before the event. Columns

4-7 detail the standardized value of each component.

VI. CONCLUSION AND POTENTIAL APPLICATIONS

Applying network analysis tools to evaluate and compare the supply fragility of

individual traded goods at a global scale generates new insight into the supply side risks of

modern international trade. Anecdotal evidence of choke points in the global trade network

corresponds well with the risky products predicted by network analysis tools. Case study analysis

provides some evidence of outsized domestic effects from import supply shocks to risky products

(those with the most fragile networks). Using a highly disaggregated international trade database

we examine variation in trade networks structure and use these differences in structure to identify

the most risky products globally. We assess the results with a country-level measure of potential

supply shock vulnerability based on the composition of their individual import basket. This

measure can be used to assess potential spillover effects of supply shocks from the importation

of specific goods (from specific countries). The methodology additionally can be applied to

predict exporters’ potential to originate negative spillovers from natural disasters, political

instability, and conflicts.

By exploring individual characteristics of the riskiness of individual goods, in addition to the overall measure, researchers and policy makers can break down and investigate different dimensions of an imports’ fragility. Likewise, a country-level indicator can be an useful starting point for undertaking nuanced policy-relevant diagnostics and analysis and identification of specific areas for reforms/interventions. Over time, the methodology could be used to evaluate ongoing efforts to improve the resilience of trade to global shocks.

The suggested methodology has a number of potential applications, including the following: (i) as a vigilance tool; (ii) a tool to evaluate spillovers; and (iii) assess policies. To evaluate respective vulnerabilities of countries over time, one can use the country-level share of risky products in import basket, supplemented by information on main exporters of risky products and the structure of domestic economy. The data can be used to assess the potential

HS2002 6-digit

Risk

category

in 2010

Products Description

(sometimes shortened)

z-score

Comp. 1

z-score

Comp. 2

z-score

Comp. 3

z-score

Comp. 4

Thailand Case Study - 2011 Floods

847170 7 Computers # Storage units 2.43 2.81 1.94 1.42includes computer hard disk drives

854121 7 Semiconductor Devices # with a dissipation rate of less than 1 W 0.70 -0.07 1.82 0.88includes semiconductors used in microprocessors

870421 7 Delivery Trucks # g.v.w. not exceeding 5 tonnes 1.40 6.01 1.21 1.77includes pick up trucks

21

impact of natural disasters or political shocks globally and by country. Maps and network graphs can be helpful in visualizing the spillovers.

Additionally, by using information on all goods, this methodology can be useful for identifying the potential impact of natural disasters on the supply and prices of consumer goods and raw materials, which is particularly important for low income countries and island countries.

This research offers several potential extensions, such as cross-country analysis (for example, the analysis of business cycles and localized supply shocks), in an interaction with the global value chain, analysis of permanent supply shocks (technological progress). The same methodology can be used for the analysis of trade in services, FDI, and other financial instruments. Additionally, it can be a powerful tool for micro level research of firms or industries interconnectedness globally (through, for example, input-output tables).

A number of policy implications emerge from the analysis of risky products and countries

vulnerabilities from importing them. As we show, better monitoring and more-detailed data are

needed to adopt a more robust approach to understanding risks inherent in the modern global

trade system. Indeed, such risks can be foreseen by taking network effects of trade into account.

For example, as discussed in this paper, shock spillovers can be mitigated by macroeconomic

policies that influence the properties of the export-import matrices of individual countries by

changing their in- and out- degrees, the centrality measure, tendency to cluster, and other network

properties. Efforts towards the diversification of suppliers of risky products might be desirable

for some countries (i.e. imports remains to be highly concentrated). Countries may consider

policy instruments capable of mitigating systemic trade risk arising from importation of risky

goods, such as building up strategic physical reserves of certain risky products (at a country or

firm level), trade regulations (for example, tax incentives), and trade promotion agencies, each

with the intention of mitigating market volatility while ensuring sufficient supply.

22

REFERENCES

Abbassi, P., Gabrieli, S., and C.P. Georg, 2013, “A network view on money market freezes,”

mimeo.

Acemoglu, D., Akcigit U., and W. Kerr, 2015, “Networks and the Macroeconomy: An Empirical

Exploration,” NBER Working Paper Series 21344.

Acemoglu, D., Carvalho, V.M., Ozdaglar, A., and A. Tahbaz‐Salehi, 2012, “The network origins

of aggregate fluctuations,” Econometrica, 80(5):1977-2016.

Allen, F., Babus, A., and E. Carletti, 2012, “Asset commonality, debt maturity and systemic

risk,” Journal of Financial Economics, 104(3), pp.519-534.

Andrade da Silva, J., and L. Cernat, 2012, “Coping with loss: the impact of natural disasters on

developing countries' trade flows,” Directorate General for Trade, European Commission

Working Paper No. 1.

Armington, P., 1969, “A Theory of Demand for Products Distinguished by Place of Production,”

International Monetary Fund Staff Papers, XVI (1969), 159-78.

Aspalter L., 2016, “Estimating Industry-level Armington Elasticities for EMU Countries,”

Working Paper N217, Department of Economics, Vienna University of Economics and

Business.

Baldwin, R., ed. 2009, “The Great Trade Collapse: Causes, Consequences and Prospects,”

London: Centre for Economic Policy Research.

Ballester, C., Calvó‐Armengol, A., and Y. Zenou, 2006, “Who's who in networks. wanted: the

key player,” Econometrica, 74(5), pp.1403-1417.

Barrat, A., Barthelemy, M., Pastor-Satorras, R., and A. Vespignani, 2004, “The architecture of

complex weighted networks,” Proceedings of the National Academy of Sciences of the

United States of America, 101(11), pp.3747-3752.

Barrot, J., and J. Sauvagnat, 2014, “Input specificity and the propagation of idiosyncratic shocks

in production networks,” Working Paper, University of Bocconi.

Bejan, M., 2006, “Trade Openness and Output Volatility,” manuscript SSRN

Bergin, P.R., Feenstra, R.C., and G.H. Hanson, 2011, “Volatility due to offshoring: Theory and

evidence,” Journal of International Economics, 85(2), pp.163-173.

Bems, R., Robert C. J., and K. Yi. 2010, “Demand Spillovers and the Collapse of Trade in the

Global Recession,” IMF Economic Review 58 (2): 295–326.

———, 2013, “The Great Trade Collapse,” Annual Review of Economics 5 (1): 375–400.

Besedes, T., and A.P. Murshid, 2014,“The Effects of Airspace Closures on Trade in the

Aftermath of Eyjafjallajökull,” Georgia Institute of Technology, mimeo.

Blomberg, S.B., and G.D. Hess, 2006, “How Much Does Violence Tax Trade?” The Review of

Economics and Statistics, 88(4): 599-612.

Boehm, C., Flaaen A., and N. Pandalai-Nayar, 2014, “Complementarities in Multinational

Production and Business Cycle Dynamics,” Working paper, University of Michigan.

Bonacich, E., 1972, “A theory of ethnic antagonism: The split labor market,” American

sociological review, pp.547-559.

Brin and L. Page, 1998, “Computer Networks and ISDN Systems,” N30, 107.

Carvalho, V. M., Nirei, M., Saito, Y., and A. Tahbaz-Salehi, 2014, “Supply chain disruptions:

evidence from the Great East Japan Earthquake,”.

Carvalho, V.M., 2014, “From micro to macro via production networks,” The Journal of

Economic Perspectives, 28(4), pp.23-47.

Cavallo, E., 2008, “Output Volatility and Openness to Trade: a Reassessment,” Journal of

LACEA Economia, Latin America and Caribbean Economic Association.

23

Contreras, M.G.A., and G. Fagiolo, 2014, “Propagation of economic shocks in input-output

networks: A cross-country analysis,” Physical Review E, 90(6), p.062812.

De Benedictis, L., Nenci, S., Santoni, G., Tajoli, L., and C. Vicarelli, 2014, “Network Analysis

of World Trade using the BACI-CEPII dataset,” Global Economy Journal, 14(3-4): 287-

343.

De la Torre, A., Didier, T., Ize, A., Lederman, D., and S. Schmukler, 2015, Latin America and

the Rising South: Changing World, Changing Priorities, Washington, DC: World Bank.

Di Giovanni, J., and A. Levchenko, 2008, “Trade Openness and Volatility,” Review of

Economics and Statistics, 91(3): 558-585.

Di Giovanni, J., and A. Levchenko, 2010, “Putting the parts together: trade, vertical linkages,

and business cycle comovement,” American Economic Journal: Macroeconomics 2(2):

95-124.

Djankov, S., Freund, C., and C. Pham, 2010, “Trading on Time,” The Review of Economics and

Statistics, 92(1): 166-173.

Easterly, W., Islam R., and J. Stiglitz, 2001, “Shaken and Stirred: Explaining Growth Volatility,”

Annual World Bank Conference on Development Economics, p. 191-212. World Bank,

July, 2001.

Escaith, H., Keck A., Nee C., and R. The, 2011, “Japan's earthquake and tsunami: International

trade and global supply chain impacts,” VoxEU. org, last modified April 28 (2011).

Fagiolo, G., Reyes, J., and S. Schiavo, 2009, “World-trade web: Topological properties,

dynamics, and evolution,” Physical Review E, 79(3), p.036115.

Fagiolo, G., Reyes, J., and S. Schiavo, 2010, “The evolution of the world trade web: a weighted-

network analysis,” Journal of Evolutionary Economics, 20(4), pp.479-514.

Feenstra, R. C., Lipsey R. E., Deng H., Ma A.C., and H. Mo, 2005, “World trade flows: 1962-

2000,” National Bureau of Economic Research, No. w11040.

Freund, C., 2009, “The Trade Response to Global Downturns: Historical Evidence,” Policy

Research Working Paper 5015, World Bank, Washington, DC.

Gabaix, X., 2011, “The granular origins of aggregate fluctuations,” Econometrica, 79(3),

pp.733-772.

Gassebner, M., Keck A., and R. Teh, 2010. “Shaken, not Stirred: The Impact of Disasters on

International Trade,” Review of International Economics, 18(2):351-368.

Glick, R., and A.M. Taylor, 2010, “Collateral Damage: Trade Disruption and the Economic

Impact of War,” The Review of Economics and Statistics, 92(1): 102-127.

Hummels, D., and G. Schaur, 2013, “Time as a Trade Barrier,” American Economic Review,

103(7): 2935-59.

IMF, 2013, “Trade interconnectedness: The World with Global Value Chains,” Washington, DC:

International Monetary Fund.

IMF, 2015, “Review of the Role of Trade in the Work of the Fund,” Washington, DC:

International Monetary Fund.

Haddad, M., Lim J., and C. Saborowski, 2010, “Trade Openness Reduces Growth Volatility

When Countries Are Well Diversified,” The World Bank, WPS5222.

Hattori, M., and Y. Suda, 2007, “Developments in a cross-border bank exposure network,”

In proceedings of a CGFS workshop held at the Bank for International Settlements.

Jackson, M., 2010, Social and Economic Networks, NJ: Princeton University Press

Johnson, R.C., 2014, “Trade in Intermediate Inputs and Business Cycle Comovement,”

American Economic Journal: Macroeconomics, 6 (4), 2014.

Katz, L., 1953, ”A new status index derived from sociometric analysis,” Psychometrika 18(1):

39-43.

24

Kose, A., and K. Yi, 2001, “International Trade and Business Cycles: Is Vertical Specialization

the Missing Link?,“ American Economic Review, 91(2):, 371- 375.

Kubelec, C., and F. Sá, 2012, “The Geographical Composition of National External Balance

Sheets: 1980–2005,” Working paper No. 384. Bank of England.

Long, A.G., 2008 “Bilateral trade in the shadow of armed conflict,” International Studies

Quarterly, 52(1): 81-101.

Martin, P., Mayer, T., and M. Thoenig, 2008,“Civil Wars and International Trade,” Journal of

the European Economic Association, 6(2-3): 541-550.

Minoiu, C., and Reyes, J.A., 2013, “A network analysis of global banking: 1978–2010,” Journal

of Financial Stability, 9(2), pp.168-184.

Muhammad, A., D'Souza, A., W. Amponsah, and W. W., 2011, “Violence, Political Instability,

and International Trade: Evidence from Kenya’s Cut Flower Sector,” SSRN working

paper.

Newbery, D. MG, and J. E. Stiglitz, 1984, "Pareto inferior trade," The Review of Economic

Studies 51(1): 1-12.

Newman, M.E.J., 2005, “Networks: An Introduction,” UK: Oxford University Press.

Nguyen, D. X., and G. Schaurb, 2012. “Import and Export Linkages Transmit Volatility Across

Markets,” Working paper, University of Copenhagen.

OECD, 2013, Interconnected Economies: Benefiting from Global Value Chains. Paris: OECD.

Oh, C.H., and R. Reuveny, 2010, “Climatic natural disasters, political risk, and international

Trade,” Global Environmental Change 20 (2): 243-254.

Sá, F., 2010, “Evidence on financial globalization and crisis: Geographic/bilateral external

balance sheets,” mimeo.

Saito, M., 2004, “Armington elasticities in intermediate inputs trade: a problem in using

multilateral trade data,” Canadian Journal of Economics, 37(4): 1097-1117.

Shirotori, M., Tumurchudur, B., and O. Cadot, 2010, “Revealed Factor Intensity Indices at the

Product Level,” United Nations Conference on Trade and Development Working Paper.

Tabak B., Takami Y., Rocha J., and D. Cajueiro, 2014, “Directed Clustering Coefficient as a

Measure of Systemic Risk in Complex Banking Networks,” Physica A: Statistical

Mechanics and its Applications. Vol. 394: 211–16.

Tanaka, A., 2012, “The effects of FDI on domestic employment and workforce composition.

Research Institute of Economy,” Trade and Industry (RIETI) Working Paper.

UNCTAD, 2013, World Investment Report: Global Value Chains: Investment and Trade for

Development. Geneva: UNCTAD.

Volpe Martincus, C., and J. Blyde, 2013, “Shaky roads and trembling exports: Assessing the

trade effects of domestic infrastructure using a natural experiment,” Journal of

International Economics 90(1): 148-161.

Ward, M.D., Ahlquist, J.S., and A. Rozenas, 2013, “Gravity's rainbow: a dynamic latent space

model for the world trade network,” Network Science,1(01), pp.95-118.

Watts, D.J., and S.H. Strogatz, 1998, “Collective dynamics of ‘small-world’

networks,” Nature, 393(6684): 440-442.

WB, 2015, Global Economic Prospects, Chapter 4, “What Lies Behind the Global Trade

Slowdown?”, Washington DC, World Bank.

25

Annex I. Technical details on the definition of components and the overall product

fragility measure

Our methodology explores the trade network of individual goods. For each of network, we use

information on which countries export and import the good, and the annual value of exports for

each resultant pair of countries. Using network analysis terminologies, countries are represented

by nodes, and exports will be represented by directed ties linking a pair of nodes. Three network

analysis measures are used in the paper:

A. Outdegree centrality

Outdegree centrality is a network analysis tool to identify the most influential nodes within a

graph. It is defined as the sum of ties that a node directs outward to other nodes as a share of the

total number of other nodes. This measure is unweighted (the value of the ties is not taken into

account). The mathematical formulation for each product network is:

𝐶𝑖𝑘𝑜𝑢𝑡 =

∑ 𝑇𝑖𝑗𝑘𝑛𝑗=1

𝑛 − 1

where 𝐶𝑖𝑜𝑢𝑡 is the outdegree centrality of country 𝑖, 𝑇𝑖𝑗 equals to 1 if there is a directed tie linking

𝑖 to 𝑗 (equal to zero otherwise), and 𝑛 is the total number of nodes in the network.

We use the standard deviation of outdegree centrality to measure each product’s risk arising from

having a few very central exporters. Formally:

𝐶𝑒𝑛𝑡𝑟𝑎𝑙𝑖𝑡𝑦𝑘 = √𝐶𝑖𝑘𝑜𝑢𝑡 − 𝐶𝑘

𝑜𝑢𝑡̅̅ ̅̅ ̅̅

𝑛 − 1

where 𝐶𝑘𝑜𝑢𝑡̅̅ ̅̅ ̅̅ is the average centrality of countries for product 𝑘.

B. Tendency to Cluster

To assess the tendency to cluster of a network of goods, we use two complementary measures:

B.1. Weighted average of local cluster coefficient

The clustering coefficient measures the degree to which nodes tend to cluster together. The local

cluster coefficient in the sense of Watts and Steven Strogatz (1998) quantifies the tendency of

the connected nodes of a country to form a clique, i.e. to trade together.

The local clustering coefficient 𝐶𝐶𝑖 for a node 𝑖 is given by the proportion of ties between 𝑖’s

neighbor, divided by the maximal number of possible connections. In the non-weighted version,

i i i

26

possible outcomes range from 0 (no connection among the partners of a country) to 1 (all the

neighbor countries are connected). We first use the weighted extension proposed by Barrat et al.

(2004). A value is assigned to each triplet in the network based on the arithmetic mean. Next,

the sum of the value of each closed triplet in the neighbor of each i is calculated and divided by

the sum of the value of the triplets.

Formally:

𝐶𝐶𝑖𝑤 =

1

𝑘𝑖(𝑘𝑖 − 1)∑

1

⟨𝑤𝑖⟩

𝑤𝑖𝑗 + 𝑤𝑖𝑘

2𝑗,𝑘

𝑇𝑖𝑗𝑇𝑖𝑘𝑇𝑗𝑘

Where 𝑘𝑖 is the number of nodes connected to 𝑖, 𝑤𝑖𝑗 the value of the tie between 𝑖 and 𝑗, and

⟨𝑤𝑖⟩ the average weight of ties connected with 𝑖:

⟨𝑤𝑖⟩ =∑ 𝑇𝑖𝑗𝑗

𝑘𝑖

The weighted local cluster coefficient calculates the contribution of each triangle, weighted by

the arithmetic average of the two adjacent ties, to the average weight of all the connections of

node 𝑖.

Note that the direction of the ties is not taken into account in this measure.

B.2. Diameter

The diameter of a network is the length of the shortest path between the most distant nodes, i.e.

the length of the longest geodesic path. It calculates the number of steps necessary for a node to

reach the furthest node in the network. This measure is directed (the direction of the ties matter)

but not weighted.

B.3. Value of the component

The value of the component of tendency to cluster is then equal to:

𝐶𝑙𝑢𝑠𝑡𝑒𝑟𝑘 = 𝐶𝐶𝑖𝑤. 𝐷𝑖𝑎𝑚𝑒𝑡𝑒𝑟

C. Systemic Relevance

The third component of the index of vulnerability is the relevance of each product to countries’

imports. We average the share of countries’ total import of each good and produce a world

average. Formally,

𝑆𝑦𝑠𝑡𝑒𝑚𝑖𝑐 𝑅𝑒𝑙𝑒𝑣𝑎𝑛𝑐𝑒𝑘 =∑ 𝑀𝑖𝑘/𝑀𝑖

𝑛𝑖=1

𝑛

Where 𝑀𝑖𝑘 the value of import of country 𝑖 in product 𝑘, and 𝑀𝑖 is the value of total imports of

country 𝑖.

27

D. International substitutability

The last component calculates the dispersion of human capital levels of countries exporting a

good. The formulation is the following:

𝐼𝑛𝑡′𝑙 𝑆𝑢𝑏𝑠𝑡𝑖𝑡𝑢𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑘 = √𝐿𝑖

𝑘 − 𝐿𝑖𝑘̅̅ ̅

𝑛 − 1

where 𝐿𝑖𝑘 is the level of human capital of country 𝑖 exporter of product 𝑘. The “wider” the

distribution of human capital of exporting countries, the more difficult it will be for a country to

find a substitute supplier that corresponds to its standard.

E. Classifying overall product fragility

To classify products in different groups (risky/not risky) we first normalize the values of

the four components described in the Section III.1 by calculating z-scores for each product and

year:

𝑧𝑖𝑡 =𝑥𝑖𝑡 − �̅�•𝑡

𝜎(𝑥𝑖𝑡)

( 1)

where the z-score, z, for each product, i, and year t , is calculated as the raw score for each

product and each year, 𝑥𝑖𝑡, minus the average score for all products in that year, �̅�•𝑡, divided by

the standard deviation of the index component 𝜎(𝑥𝑖𝑡).

In the next step, we use cluster analysis (the k-median procedure) in order to partition the

products into mutually excluding groups. The algorithm seeks to maximize the variation between

clusters and minimize the variation inside. To reach this goal, the algorithm iterates the

minimization of the following equation:

∑ ∑|𝑋𝑐𝑘 − �̅�𝑐|2

𝑐𝑘

( 2)

where 𝑋𝑐𝑘 is the value of the component 𝑐 of product 𝑘, and |𝑋𝑐𝑘 − �̅�𝑐| is the distance

between each product and the “center” of the cluster (i.e. the median of the current product in

the cluster).

After categorizing products, we can track the importers and exporters of risky products

by looking at the risky-product share of total imports or exports in a county’s trade basket.

28

Annex II. Descriptive statistics

A. Frequency of products

Table A1. Number of products in each category over time

Notes: The table summarizes the frequency of the products over time in seven Groups. Group 1 is the

least-risky group while Group 7 the riskiest. The methodology implies certain constancy in the number

of products by groups of riskiness. On average across years, 359 products are categorized in the riskier

group (Group 7). Products which the algorithm fails to associate to a group are shown in the “non-

classified” column.

Group 1 Group 2 Group 3 Group 4 Group 5 Group 6 Group 7Non-

classified

2003 371 519 444 453 592 420 382 204

2004 469 410 572 454 458 632 386 5

2005 432 454 616 380 517 564 416 7

2006 358 529 445 628 520 534 366 6

2007 500 646 393 526 678 78 435 130

2008 359 394 585 554 638 495 357 4

2009 447 478 655 506 422 524 350 4

2010 456 571 572 608 348 471 356 4

2011 308 469 642 435 482 466 460 124

2012 310 589 569 645 323 596 348 6

2013 351 537 532 566 416 580 382 22

2014 318 601 482 665 390 523 382 25

29

B. List of risky products

Table A2. Products with highest risk average during 2003-2014

HS2002_6d

Product Description Share

Value of

Imports

1 870323 Cars # Of a cylinder capacity exceeding 1,500 cc but not exceeding 3,000 cc 2.777%

2 870332 Cars # Of a cylinder capacity exceeding 1,500 cc but not exceeding 2,500 cc 1.485%

3 847170 Computers # Storage units 1.033%

4 870322 Cars # Of a cylinder capacity exceeding 1,000 cc but not exceeding 1,500 cc 0.791%

5 870829 Vehicle Parts # Other 0.678%

6 880330 Aircraft Parts # Other parts of aeroplanes or helicopters 0.671%

7 854140 Semiconductor Devices # Photosensitive semiconductor devices, including photovoltaic cells

whether or not assembled in modules or made up into panels; l ight emitting diodes

0.647%

8 850440 Electrical Transformers # Static converters 0.591%

9 870421 Delivery Trucks # g.v.w. not exceeding 5 tonnes 0.521%

10 853400 Printed Circuit Boards # Printed circuits. 0.503%

11 401110 Rubber Tires # Of a kind used on motor cars (including station wagons and racing cars) 0.413%

12 840820 Combustion Engines # Engines of a kind used for the propulsion of vehicles of Chapter 87 0.374%

13 732690 Other Iron Products # Other 0.371%

14 853710 Electrical Control Boards # For a voltage not exceeding 1,000 V 0.368%

15 853690 Low-voltage Protection Equipment # Other apparatus 0.362%

16 840991 Engine Parts # Suitable for use solely or principally with sparkignition internal combustion

piston engines

0.350%

17 853890 Electrical Power Accessories # Other 0.334%

18 840999 Engine Parts # Other 0.332%

19 940190 Seats # Parts 0.292%

20 901839 Medical Instruments # Other 0.266%

21 843149 Excavation Machinery # Other 0.265%

22 730890 Iron Structures # Other 0.264%

23 841480 Air Pumps # Other 0.261%

24 401120 Rubber Tires # Of a kind used on buses or lorries 0.230%

25 850300 Electric Motor Parts # Parts suitable for use solely or principally with the machines of

heading 85.01 or 85.02.

0.221%

26 870190 Tractors # Other 0.215%

27 848340 Transmissions # Gears and gearing, other than toothed wheels, chain sprockets and other

transmission elements presented separately; ball or roller screws; gear boxes and other speed

changers, including torque converters

0.213%

28 842139 Centrifuges # Other 0.209%

29 852190 Video Recording Equipment # Other 0.192%

30 853650 Low-voltage Protection Equipment # Other switches 0.191%

31 841490 Air Pumps # Parts 0.186%

32 850780 Electric Batteries # Other accumulators 0.184%

33 841430 Air Pumps # Compressors of a kind used in refrigerating equipment 0.183%

34 843143 Excavation Machinery # Parts for boring or sinking machinery of subheading 8430.41 or

8430.49

0.177%

35 848210 Ball Bearings # Ball bearings 0.175%

36 870870 Vehicle Parts # Road wheels and parts and accessories thereof 0.174%

37 840890 Combustion Engines # Other engines 0.169%

38 848190 Valves # Parts 0.167%

39 841391 Liquid Pumps # Of pumps 0.166%

40 731815 Iron Fasteners # Other screws and bolts, whether or not with their nuts or washers 0.166%

30

41 853120 Audio Alarms # Indicator panels incorporating l iquid crystal devices (LCD) or l ight emitting

diodes (LED)

0.165%

42 854129 Semiconductor Devices # Other 0.164%

43 841330 Liquid Pumps # Fuel, lubricating or cooling medium pumps for internal combustion piston

engines

0.163%

44 850110 Electric Motors # Motors of an output not exceeding 37.5 W 0.162%

45 853669 Low-voltage Protection Equipment # Other 0.156%

46 841370 Liquid Pumps # Other centrifugal pumps 0.155%

47 851220 Electrical Lighting and Signalling Equipment # Other l ighting or visual signalling equipment 0.155%

48 841590 Air Conditioners # Parts 0.145%

49 848310 Transmissions # Transmission shafts (including cam shafts and crank shafts) and cranks 0.141%

50 392190 Other Plastic Sheetings # Other 0.133%

51 940320 Other Furniture # Other metal furniture 0.130%

52 940510 Light Fixtures # Chandeliers and other electric ceil ing or wall l ighting fittings, excluding those

of a kind used for l ighting public open spaces or thoroughfares

0.123%

53 854110 Semiconductor Devices # Diodes, other than photosensitive or l ight emitting diodes 0.123%

54 850490 Electrical Transformers # Parts 0.122%

55 401693 Other Rubber Products # Gaskets, washers and other seals 0.120%

56 842199 Centrifuges # Other 0.118%

57 841459 Air Pumps # Other 0.118%

58 391990 Self-adhesive Plastics # Other 0.118%

59 940540 Light Fixtures # Other electric lamps and lighting fittings 0.116%

60 230990 Animal Food # Other 0.116%

61 401699 Other Rubber Products # Other 0.113%

62 850450 Electrical Transformers # Other inductors 0.109%

63 901819 Medical Instruments # Other 0.109%

64 392310 Plastic Lids # Boxes, cases, crates and similar articles 0.108%

65 848390 Transmissions # Toothed wheels, chain sprockets and other transmission elements presented

separately; parts

0.105%

66 842240 Washing and Bottling Machines # Other packing or wrapping machinery (including heatshrink

wrapping machinery)

0.104%

67 847190 Computers # Other 0.100%

68 902790 Chemical Analysis Instruments # Microtomes; parts and accessories 0.100%

69 392321 Plastic Lids # Of polymers of ethylene 0.100%

70 852910 Broadcasting Accessories # Aerials and aerial reflectors of all kinds; parts suitable for use

therewith

0.094%

71 850131 Electric Motors # Of an output not exceeding 750 W 0.090%

72 842129 Centrifuges # Other 0.089%

73 850152 Electric Motors # Of an output exceeding 750 W but not exceeding 75 kW 0.087%

74 871200 Bicycles # Bicycles and other cycles (including delivery tricycles), not motorised. 0.087%

75 701090 Glass Bottles # Other 0.086%

76 870893 Vehicle Parts # Clutches and parts thereof 0.085%

77 871690 Trailers # Parts 0.085%

78 392330 Plastic Lids # Carboys, bottles, flasks and similar articles 0.083%

79 841360 Liquid Pumps # Other rotary positive displacement pumps 0.081%

80 842121 Centrifuges # For fi ltering or purifying water 0.078%

31

81 392350 Plastic Lids # Stoppers, l ids, caps and other closures 0.074%

82 841381 Liquid Pumps # Pumps 0.073%

83 392390 Plastic Lids # Other 0.073%

84 850431 Electrical Transformers # Having a power handling capacity not exceeding 1 kVA 0.071%

85 320890 Nonaqueous Paints # Other 0.070%

86 853620 Low-voltage Protection Equipment # Automatic circuit breakers 0.070%

87 850710 Electric Batteries # Leadacid, of a kind used for starting piston engines 0.069%

88 491110 Other Printed Material # Trade advertising material, commercial catalogues and the like 0.069%

89 321519 Ink # Other 0.068%

90 871419 Bi-Wheel Vehicle Parts # Other 0.068%

91 481910 Paper Containers # Cartons, boxes and cases, of corrugated paper or paperboard 0.067%

92 847490 Stone Processing Machines # Parts 0.067%

93 848350 Transmissions # Flywheels and pulleys, including pulley blocks 0.066%

94 854420 Insulated Wire # Coaxial cable and other coaxial electric conductors 0.066%

95 842123 Centrifuges # Oil or petrolfi lters for internal combustion engines 0.066%

96 842290 Washing and Bottling Machines # Parts 0.065%

97 841899 Refrigerators # Other 0.065%

98 731210 Stranded Iron Wire # Stranded wire, ropes and cables 0.063%

99 870810 Vehicle Parts # Bumpers and parts thereof 0.063%

100 391910 Self-adhesive Plastics # In rolls of a width not exceeding 20 cm 0.063%

101 848590 Boat Propellers # Other 0.062%

102 442190 Other Wood Articles # Other 0.061%

103 321590 Ink # Other 0.060%

104 901580 Surveying Equipment # Other instruments and appliances 0.059%

105 843120 Excavation Machinery # Of machinery of heading 84.27 0.058%

106 850140 Electric Motors # Other AC motors, singlephase 0.058%

107 851290 Electrical Lighting and Signalling Equipment # Parts 0.057%

108 830230 Metal Mountings # Other mountings, fittings and similar articles suitable for motor vehicles 0.055%

109 940330 Other Furniture # Wooden furniture of a kind used in offices 0.054%

110 851150 Electrical Ignitions # Other generators 0.054%

111 851140 Electrical Ignitions # Starter motors and dual purpose startergenerators 0.054%

112 340290 Cleaning Products # Other 0.053%

113 853931 Electric Filament # Fluorescent, hot cathode 0.052%

114 731816 Iron Fasteners # Nuts 0.052%

115 853649 Low-voltage Protection Equipment # Other 0.052%

116 391740 Plastic Pipes # Fittings 0.051%

117 730799 Iron Pipe Fittings # Other 0.049%

118 853190 Audio Alarms # Parts 0.047%

119 853641 Low-voltage Protection Equipment # For a voltage not exceeding 60 V 0.046%

120 820750 Interchangeable Tool Parts # Tools for dril l ing, other than for rock dril l ing 0.045%

32

Notes: Products are considered as risky if they constantly are classified in Group 7 during 2003-

2014. Products are sorted by their relative importance in the world trade between 2003-2014.

121 842131 Centrifuges # Intake air fi lters for internal combustion engines 0.044%

122 392329 Plastic Lids # Of other plastics 0.044%

123 491199 Other Printed Material # Other 0.044%

124 482110 Paper Labels # Printed 0.041%

125 820790 Interchangeable Tool Parts # Other interchangeable tools 0.041%

126 848360 Transmissions # Clutches and shaft couplings (including universal joints) 0.038%

127 340399 Lubricating Products # Other 0.037%

128 853180 Audio Alarms # Other apparatus 0.035%

129 820559 Other Hand Tools # Other 0.035%

130 820719 Interchangeable Tool Parts # Other, including parts 0.035%

131 830140 Padlocks # Other locks 0.035%

132 851310 Portable Lighting # Lamps 0.033%

133 960390 Brooms # - Other 0.032%

134 700711 Safety Glass # Of size and shape suitable for incorporation in vehicles, aircraft, spacecraft or

vessels

0.031%

135 731829 Iron Fasteners # Other 0.031%

136 392610 Other Plastic Products # Office or school supplies 0.031%

137 940290 Medical Furniture # Other 0.030%

138 830249 Metal Mountings # Other 0.030%

139 871680 Trailers # Other vehicles 0.030%

140 732020 Iron Springs # Helical springs 0.029%

141 732620 Other Iron Products # Articles of iron or steel wire 0.028%

142 848280 Ball Bearings # Other, including combined ball/roller bearings 0.028%

143 630900 Used Clothing # Worn clothing and other worn articles. 0.026%

144 482010 Paper Notebooks # Registers, account books, note books, order books, receipt books, letter

pads, memorandum pads, diaries and similar articles

0.025%

145 680422 Milling Stones # Of other agglomerated abrasives or of ceramics 0.024%

146 848420 Gaskets # Mechanical seals 0.023%

147 731029 Small Iron Containers # Other 0.022%

148 731822 Iron Fasteners # Other washers 0.022%

149 853929 Electric Filament # Other 0.022%

150 350699 Glues # Other 0.019%

151 848490 Gaskets # Other 0.019%

152 401039 Rubber Belting # Other 0.018%

153 400942 Rubber Pipes # With fittings 0.017%

154 820890 Cutting Blades # Other 0.015%

155 400911 Rubber Pipes # Without fittings 0.014%

156 820320 Hand Tools # Pliers (including cutting pliers), pincers, tweezers and similar tools 0.013%

157 901780 Drafting Tools # Other instruments 0.013%

158 820411 Wrenches # Nonadjustable 0.013%

33

Annex III. Fragility maps over time

1. Importers of fragile products

2003

2009

2013

Notes: Overtime, the share of imports of risky products have increased in Latin America, Russia and

Australia, while it decreased in Europe, and East Asia.

34

2. Exporters of fragile products

2003

2009

2013

Notes: Regions exporting risky goods are stable overtime, and even more concentrated in the last

years than at the beginning of the 2000s.