THE IMPACT OF COUNTRY CHARACTERISTICS ON TRADE PERFORMANCE

(1)

THE IMPACT OF COUNTRY

CHARACTERISTICS ON TRADE

PERFORMANCE

A sector-level approach on the impact of levels of development

on Value Added in Exports

MASTER THESIS

RESIT

June 18th₂₀₁₉

Double Degree Master in Economic Development and Growth (MEDEG) Master in International Economics and Business

University of Groningen

Master in Economic Development and Growth University of Lund

Enzo Michel Boccara

(2)

Abstract

The increasing fragmentation of production across borders has presented developing countries with new opportunities regarding sources of income. Benefits stemming from export-oriented goods have long been linked to a country’s level of development, by stimulating increases in generation of value added. By taking a sector-level approach, the present paper attempts to estimate the relationship between human capital, institutional quality and transport infrastructure on developing countries’ value added in exports (VAX). We shall demonstrate that transport infrastructure has had a positive effect on VAX growth between 2005 and 2015, while no conclusions can be drawn as far as human capital’s and institutional quality’s effects are concerned. We also prove that human capital has had a higher impact on VAX growth of capital-intensive sectors. Eventually, we pinpoint the limits of this approach and advocate for further studies to be implemented through a more fine-grained analysis on educational attainment variables as well as regarding the assimilation of learning processes.

(3)

I. Introduction

The continuous decline in transport and communication costs has reformatted the way production processes are configured. It gradually became more profitable for firms operating in developed countries to outsource parts of the production stages of goods and services to developing countries. Under this new form of industrial organization, and the increased specialization in production of parts of a good, new forms of dependencies have emerged between trading economies. Instead of exporting goods produced entirely domestically, firms now rely on the import of (cheaper) inputs from other countries in order to produce their own part of a good. This new reality begs the following question: have developing countries been able to take advantage of this new form of production? In other words, have they been able to extract benefits from exporting domestically produced goods consumed in other countries? This thesis aims at providing an answer to these questions, by identifying the determinants tied to increased trade performance.

This increased fragmentation of production across borders has presented developing countries with new opportunities for enrichment from trade. It has allowed them to experience a different form of industrialization from the ones developed countries usually went through. By operating with lower trading costs, developing countries could specialize in a specific stage of a production process, sparing them from the considerable production investments required to manufacture a good in its totality. Because of this, developing nations can now diversify their exports, protecting them from sector-specific shocks that might affect their terms of trade and slow down subsequent growth. These positive implications have led policymakers to try and identify factors allowing them to increase trade with other nations, in order to extract higher levels of income from such activities.

This line of reasoning has also led scholars to ask themselves what forces were behind countries’ increased participation in trade1_{, and try to establish links between country}

characteristics (i.e. levels of development) and their capacity to increase trade with other countries. Improvements in institutional quality (Levchenko, 2007), transport infrastructure specialized for cross-country trade (Clark et al, 2004) and levels of educational attainment (Amin et al, 2008) have been identified as being relevant development indicators in explaining the extent to which countries are able to export and import goods. In reality, developing countries have been dubbed the “factory” economies of the world since they tend to generally simply provide labour and take part in the production stage of goods that generate relatively low income (Baldwin et al, 2015). In order to fully understand the benefits that countries are able to extract from participating in trade, an analysis with a higher level of detail is required. Intimately connected to what we just observed, it is worth mentioning that levels of development have been positively related with trade flows (i.e. positive effect on both exports and imports). This leads us to the following question: what is the net effect of levels of development on benefits from trade participation? To this issue, it is necessary to start by defining and circumscribing some key notions: value added, VAX and intermediate inputs. Value added represents the mobilization of a country’s factors required to realize an activity. As defined in standard National Accounting, the value added generated in a country is the sum

(5)

of the value of production of each sector minus the value of all inputs used in each sector. By doing so, this indicator is able to provide information on the net income countries generate when exporting goods produced domestically. The main motivation behind this thesis lies in presenting answers on the link between the value added a country creates when producing export-oriented goods, and its levels of development.

The indicator we have chosen as a variable to provide an answer to this question is called value added in exports (VAX). First presented by Johnson and Noguera (2012), VAX measures the value generated by a country, using its factors, in order to satisfy final demand abroad. In other words, VAX subtracts a specific set of imports from a country’s exports. This type of input is known as an intermediate input, and is defined as a good that is used for the production of other goods. By doing so, the authors are able to extract information on what a country precisely gains from their exports of goods and services. This indicator differs from value added in the sense that it measures the value generated in exported and final goods, while value added does not distinguish between types of goods. By exactly quantifying a country’s income stemming from exports, VAX is a powerful indicator for estimating economic gains from trade. As we will explain in this thesis, the extent to which increases in VAX can generate benefits for an economy depends (in part) on whether the factors of production (i.e. capital and employment) are being used optimally or not.

Overall, the literature showing how country characteristics end up affecting a country’s exports to –and imports from– other countries is vast. However, comparative studies looking into the factors explaining increases in VAX for developing countries are rare. This research gap underlines the importance of presenting a study looking into which (and to what extent) country characteristics play a role in increasing gains from trade. By taking into account the imports of goods used to produce other goods that are to be exported and consumed in foreign countries, VAX is an indicator that can relay precise information on what a country gains from trade. This leaves us with the following question: what is the net effect of country characteristics on VAX? Therefore the research question around which this paper will be articulated could be summarized as follows: did levels of development have positive implications on VAX growth between 2005 and 2015 for developing countries?

(6)

growth, at both the country and sector level. As far as the latter is concerned, we have decided to assess whether higher levels of human capital have a bigger impact on VAX growth of capital-intensive sectors than on those that are not. This choice is justified by the fact that capital-intensive sectors, characterized by a relatively low ratio of labour inputs, tend to require skilled labour capable of managing the machines. The second hypothesis therefore consists in testing if additional years of education have a bigger effect on a sector’s value added in exports if that sector is capital-intensive. Ultimately, we expect increases in a population’s educational attainment to have a higher impact in changes in VAX of capital-intensive sectors, than non-capital-intensive sectors.

In order to do this study, we need to look at VAX growth of similar developing countries. To calculate the VAX for different sectors and countries we will be using the OECD’s Inter-Country Input-Output (ICIO) Tables. The particularity of this type of dataset is that it recognizes the sizeable fragmentation of production across different countries, allowing us to observe which sectors are being mobilized for the production of goods. To bring about our analysis, we have used a panel dataset composed of 32 countries, each with 36 different sectors, between 2005 and 2015. Causality between country- and sector characteristics and VAX growth will be tested using regression analysis. For multiple reasons, an Ordinary Least Squares model estimation method does not fit the data. Hence, the option we made for using a method correcting these issues. The development indicators –included in this study as explanatory variables in our model– are the average years of schooling for the population over 25, an index of Regulatory Quality and Rule of Law, and a Port Quality index. These variables are used to proxy for the level of human capital, institutional quality, and transport infrastructure, respectively. The control variables included are year dummies and country population growth. In order to test our second hypothesis, we also choose to include an interaction variable linking the capital-intensity of a sector, and the level of educational attainment of the associated country.

The thesis is structured as follows. Our Literature Review is presented in Section II. We first provide a brief overview of important economic concepts, followed by a variable discussion and justification behind our choice of VAX as the dependent variable in our model. We then look into the advantages developing countries can obtain from trading with other countries and how this might relate to VAX. Finally, we look at past studies that have identified development determinants that explain variations in countries’ trade flows. Particularly, we will be looking at human capital, transport infrastructure and institutional quality. In Section III we describe the methodology and estimation model, explain the sources of the data, chosen indicators and consequent variable construction. Section IV presents the results obtained in this work, in which we discuss the testing of both our hypotheses. Finally, Section V provides concluding remarks and the implications of our results, as well as the limitations of this thesis and possibilities of further studies. We present the list of References and Appendix attachments in

(7)

II. Literature Review

II.1. Trade and Value Added in Developing countries

II.1.1. Value Added in World Trade

Value added is a critical component of an economic entity, as it essentially represents the mobilization of factors in a country (or sector) required to realize an activity. The OECD explains that value added includes –among others- the compensation of employees, gross operating surplus, mixed income, and taxes on production less subsidies on production (OECD, 1992). This indicator of economic prosperity has been used to understand a country’s income stemming from exports (Johnson and Noguera, 2012), in which stage of a good’s production does the country mainly partake (Hummels et al, 2001) and to identify how increasingly significant the trade of intermediate goods is today (Koopman et al, 2012). In their work, Johnson and Noguera (2012) develop the value added in exports (VAX) indicator which measures the “value generated by a country, using its factors, in order to satisfy final demand abroad”. By doing so, they are able to extract information on what a country gains from producing an exported good or service.

When trying to measure both a country’s trade performance and value generated during the production of a good, scholars have developed numerous indicators. For instance, studies have looked at a country’s gross exports of goods and services. This is a solid indicator for measuring the scale of export-oriented production across countries and identifying each country’s central industries. As shown in Figure II.1.1. below, gross exports of goods and services is an essential part of a country’s income and heavily participates in increasing its GDP, with the world average standing at approximately 30% of GDP since 2000 (World Bank).

FIGURE II.1.1. – GROSS EXPORTS OF GOODS OF SERVICES AS A PERCENTAGE OF GDP

Source: Graph constructed using data from the World Bank.

Note: Represented in the graph are the world average and three world regions.

Due to the current scale of international production fragmentation, simply looking at gross exports when analyzing a country’s trade performance would provide incomplete and/or biased results (Timmer et al, 2013). As we shall explain further ahead, gross exports allows us to observe the exchange of final products between countries, but does not provide information regarding the net gains from exporting those products to other countries. Koopman, Wang and

0 10 20 30 40 50 60 70 2000 2003 2006 2009 2012 2015 G ro ss E xp o rts o f goo d s a n d s erv ice s (% o f GDP)

Gross Exports of goods and services (% of GDP)

Central Europe and the Baltics

East Asia & Pacific (excluding high income)

(8)

Wei (2012) stress this by mentioning that studies that do not take into account the imported content in the country’s exports are likely to exaggerate a country’s export sophistications; such is the case of Rodrik and Schott when they analyze China’s exports (see Rodrik, 2006; Schott, 2008). As we have stated, value added in exports quantifies the domestic resources a country has mobilized in order produce a good that is to be exported. By doing so, it represents the net income a country is able to extract from producing export-oriented goods, and appears as a more adequate indicator for this study. In the next section, we present the schematic outline of how firms are organized today, in order to later provide a discussion on how VAX appears advantageous as compared to other indicators, when looking at the benefits a country can obtain when trading with other countries.

II.1.2. A schematic outline of a Global Value Chain

The comparative advantage theory has played an essential role in explaining the international production fragmentation observed today in cross-border trade. This phenomenon has spurred the appearance of Global Supply Chains, a term pointing at the process of outsourcing parts of the production chain to cross-border low-cost suppliers (Gereffi et al, 2012). In other words, it became more profitable for firms to separate different stages of production of a single good between various regions rather than producing the entirety of it in one region. Developing countries, heavily endowed in low-cost labour, started being included in production processes of multinationals from developed countries, benefitting from new sources of income and employment possibilities. Such is the case of the emergence of the maquiladoras in the 1960’s which served as the first large-scale offshoring process of assembly of parts (Dicken, 2007). In order to represent the various stages in the conception of a good, scholars have come up with another term deemed more adequate in understanding contemporary trade: Global Value Chains, which represents the full range of activities required to get a good from the producer(s) to the consumer. By using the term full, we refer to activities that, apart from the production stage, include pre-production tasks (e.g. R&D, Design) and post-production tasks (e.g. advertisement, brand management, specialized logistics). Studies have shown that the physical assembly of a good per se tends to represent a lower part of the total share of value added embodied in a good, as compared to other stages such as Research and Development or Marketing of the product (Criscuolo et al, 2017). This can be seen in Figure II.1.2. below, which pits the different activities within an industry to their share of value added (i.e. the returns on unit of labor and capital), also known as the Smile Curve.

FIGURE II.1.2. – THE SMILE CURVE

(9)

Having depicted how the appearance of GVCs have changed the way benefits from trading with other countries is to be analysed, in the next sub-section we present a discussion on the various indicators derived from GVC literature and conclude on the indicator we consider best fitted to assess gains from trade of developing countries.

II.1.3. Why VAX? An illustrative example of a GVC

In order to engage a discussion on which indicator is best suited to deliver an answer to our hypotheses, let us present an illustrative example of a Global Value Chain (see Appendix C for graphic example). Suppose that a firm in country A manufactures processors. In order to produce it, country A will require inputs. Suppose for simplicity of explanation that the intermediate inputs required to manufacture this product are plastic and electronic components. The firm based in country A can either produce these inputs domestically or import them from other countries. Furthermore, these processors can be consumed as final products (e.g. to be used by consumers to customize their own electronic devices) or used as intermediate inputs in the production of another good (e.g. used to build GPS trackers). The processors produced could also be sold either in the domestic market, or exported to be consumed (or used as inputs) in other countries.

As a first example, and to get familiarized with the illustration, we first compare value added in Exports with Gross Exports. There is an increasing disconnect between gross exports and income generation (Timmer et al, 2013) due to the fact that a country’s VAX is not completely reflected by that country’s exports. Let’s pretend that in order to produce its processors2_,

country A uses the electronic components also produced domestically, but imports the plastic from country C (which can produce them at a lower cost). Once the processor has been produced, it is exported to country B to be consumed. If we were to look at country A’s gross exports of processors, all would be attributed to country A, since it is the country that has exported that good. However, a part of the value added created during the production of the processors occurred in country C. Therefore, Gross Exports will overestimate country A’s gains from exporting the processors, leading to an erroneous conclusion. It is precisely at this stage that an indicator such as VAX shows its usefulness, since it correctly distinguishes between intermediate inputs used domestically and those that are imported.

At this point of the discussion, lets briefly take a look at another indicator: GVC Income. Presented by a team of researchers at the University of Groningen (Timmer et al, 2013) its construction is similar to that of VAX. They have developed an indicator that measures changes in competitiveness3_{. Focusing on the European manufacturing sectors, they calculate the “(…)}

value added (…) that [is] directly and indirectly related to the production of final manufacturing goods”. They look at the creation of income related to activities involved in the production of final manufacturing goods, devising the term GVC income. The authors use this indicator to look at foreign and domestic demand, as they wish to observe the totality of the benefits of participating in GVC’s. Relating this to our illustration, suppose that the processors produced by country A are consumed as final products domestically and in country B. Unfortunately for

2_{Electrical component vital for the functioning of many electrical devices.}

3_{Timmer et al (2014) define competitiveness as “the ability to perform activities that meet the test of international}

(10)

country A, a competitor (country D) also produces processors and is able to sell them at a lower price. As a consequence, let’s say that country B now imports (and consumes) half of its processors from country A and the other half from country D. Suppose that, at the same time, there is a surge in domestic demand of country A for processors that exactly compensates for the loss incurred by country A due to the introduction of country D4. In this particular case, GVC income stemming from processor sales of country A will stay identical. However, because VAX only considers foreign final demand, it should decrease as a result of loss in sales of its processors to country B. Consequently, GVC income appears as a better indicator than VAX when wishing to study the total value added generated by a country in the production of final goods, as it also considers domestic demand. However, our study consists in looking at the potential impact levels of development have on the benefits stemming from trading with other countries. Therefore, by not being interested in domestic final demand, VAX appears as a more fitted indicator to provide an answer to our research question.

In a seminal paper, Koopman, Wang and Wei elaborate indicators to measure how much value added by a country is included in its gross exports. Looking at VAX and domestic content in exports (DCE), they explain that while the former captures the value generated by the factors employed in the country for the export of goods that will be absorbed by another country5_{, the}

latter does not attribute importance to where the export value is used6 (Koopman et al, 2012). This distinction provides the DCE with an advantage: by also looking at exported intermediate goods, it is capable of presenting more elaborate information on a country’s participation in international chains than VAX7. It is by looking at the final demand of goods that VAX becomes a more advantageous indicator for estimating the impact development indicators have on trade performance and value added generation. This is because it does not look at the export of intermediate goods, but at foreign final demand. The question we face is therefore why do we choose to look at final demand instead of final demand and intermediate goods? Let’s resort to our previous example to explain the problems that might arise if we were to look at the DCE to answer our research question. Suppose that country C first exports the plastic it produced to country A, so that the latter may produce its processors. Additionally, country C is involved in a GVC that produces GPS trackers, and is responsible for the assembly of the parts. To be able to assemble that good however, it needs the processors that country A produces, so it decides to import some. Once it assembles the GPS trackers, it exports them to country D to be consumed as a final product. The crucial point here is that in order to assemble the GPS trackers country C had to import an intermediate input from country A that had already imported an intermediate good from country C! The DCE indicator does not distinguish between exports of intermediate goods and exports of final goods. Therefore, the DCE will take into account both country C’s export of plastic and GPS trackers. This leads to a double counting problem, since country C’s plastic exports will be counted twice: once as an intermediate input exported to

4_{Although this scenario is not probable, this hypothetical situation is presented to understand the differences that}

might arise between GVC Income and VAX.

5_{We explain the calculation method in Section III.2.}

6_{It does not matter whether the exported good is intermediate or final.}

7_{Countries that participate in these global chains do not only export final goods. In a world of production}

(11)

country C, and another time in the export of final good to country D. The VAX indicator does not suffer from this disadvantage; by observing the foreign final demand alone, we avoid such double counting problem.

Value added in exports seems nevertheless to suffers some limits. Any change in its value observed throughout time could be due to a variety of reasons that would be hard to disentangle. For instance, by not breaking down value added into its various components, one is unable to provide answers to which ones are mostly contributing to changes in VAX. Furthermore, keeping imports constant, it could be due to a sudden surge in a country’s exports. Another indicator developed by Johnson and Noguera (2012) called the VAX ratio8 solves this issue, by netting out the effect of gross exports on the VAX. Since it represents a measure of the intensity of production sharing, the VAX ratio is therefore best fitted when looking at intermediate input contributions between sectors rather than providing total value added in foreign final demand and capturing shifts in world trade9.

Via this study, we wish to test how a country’s level of development affects its trade performance and generation of value added. In order to do so, our indicator should include the imports of intermediate inputs from other countries in the domestic production of goods. Additionally, it should not count more than once the goods that are exported to other countries, so that we avoid overestimating a country’s gains from trade. What is more, it should only consider foreign final demand, since domestic demand is not of interest when analyzing benefits stemming from trade. Value added in exports therefore appears as a measure that allows us to do this, by fulfilling all three of our requisites. One begs the following questions: To what extent do increases in VAX benefit developing countries? Is “maximum” VAX something to be pursued? In the next sub-section, we shall tackle these issues.

II.2. Benefits of increased trade and VAX

Before looking at the reasons why countries should strive to increase their VAX, and thinking on whether increasing it is always beneficial, some brief words must be said on how developing countries can benefit from increasing the intensity of trade with other countries. What opportunities has trade offered developing nations? For one, the fragmentation of production across borders allows developing countries to experience a different industrialization from the ones developed countries usually went through. If the latter had to reach a certain scale and range of competencies to pave way for the two major sector shifts (agriculture to industry to services for instance), the drastic reduction in both transport and communication costs now allows for the possibility of specializing in a specific stage of the production process. This is a very promising sign for developing countries, particularly since it seems less and less probable for a country to grow via the traditional process of resource reallocation from industries producing low value added goods (agriculture) to the ones producing high value added goods (manufactures and services). This reality is due in part to pre-mature industrialization (i.e. a

8_{It is calculated by dividing the VAX by gross exports.}

9_{For instance, Johnson and Noguera (2012) write that “Across sectors, the VAX ratio for Manufactures is low}

(12)

situation in which a developing nation becomes a service economy without having had a proper industrialization experience), a phenomenon that can be observed in developing countries. Another reason explaining why increasing trade with other countries might be beneficial for developing countries, is the increased possibility of diversifying their exports. A diversified export base can help protect countries from sector-specific shocks and their negative effects on export revenue, income, and growth. Countries that expand their exports beyond a limited number of products also lower their risks of worsening their terms of trade (Hummels et al, 2005). Hausman et al (2007) have also found that the export basket of a country –that is, the products that they export– has important implications for developing countries’ long-run economic growth. Countries that export goods with high value added appear to have a higher economic growth than those that produce goods with low value added. Scholars have also established a relation between export variety and growth emanating from learning by exporting or structural change (Lederman et al, 2003).

(13)

both exports and imports, and argue why linking these development levels to VAX growth is crucial for better understanding gains from trade.

II.3. Development indicators and Value Added in Exports

II.3.1. Country Characteristics

After having explained the advantages of focusing on VAX in order to study the potential link between a country’s level of development and benefits from trading with other countries, and shown what those benefits can be and under what circumstances increasing VAX would favour these developing countries, let’s now identify the main development indicators responsible for increases in VAX. Since the literature does not provide an answer to this question, we first start by looking at which of the development indicators are responsible for increasing trade intensity. We conclude this section by presenting the research question that we will later try to provide an answer to, and the hypotheses that will guide us in this ambition.

Which country characteristics can be deemed impactful in explaining the trade intensity of countries? The bulk of the literature seems to suggest that one of the most important factors behind a country’s ability to increase its exports is the characteristics of its population. Although factors such as culture and religion have been included in past quantitative studies in trying to explain trade volumes between countries, the following characteristic has been pinpointed as being crucial: human capital. Human capital refers to the abilities and qualities of people that make them productive (Becker, 1975). These qualities range from knowledge and habits to creativity and social attributes, and are central for the production of economic value. For instance, in a study focusing on growth disparities and patterns between developing countries, McMillan et al. (2017) argue that future development will not stem from inter-sectoral changes but from the steady accumulation of skills and human capital. How has human capital been responsible for trade intensity? Amin et al (2008) identify human capital as being a crucial factor in enabling a country to export services in GVCs. This attests to the fact that, at least for certain sectors, human capital would seem to aid in increasing exports. In a world context in which GVC’s exist, imports of intermediate products are also essential in order to produce these exported goods. Albeit the positive impact human capital accumulation may have on a country’s growth (Barro, 2001), what is the net effect on benefits stemming particularly from exporting domestically produced goods to other countries? Studies on this are lacking, emphasizing the importance of a study analyzing this particular issue.

(14)

otherwise unattainable (Micco et al, 2006). The vital advantage originating from increasing the quality of transport infrastructure would therefore seem to lie in both reducing transport costs and the impact that distance between two countries might have on trade volumes. As with human capital, the literature on the impact transport infrastructure has on exports and imports is vast, but studies providing an answer to its impact on VAX growth is lacking.

Having discussed population characteristics and transport characteristics, what can be said about the mechanisms that stipulate how business is to be handled? The third and final development indicator we have identified in the literature as being relevant in explaining trade performance is the institutional quality of a country. Definitions of what is to be considered as an institution varies between scholars. Douglass North (1990) famously defines Institutions as “the rules of the game in a society or, more formally, (…) the humanly devised constraints that shape human interaction”. In other words, institutions must be “humanly devised”, must set “rules of the game” that constrain human activities, and should have an effect on society via incentives (Acemoglu et al, 2010). How can this relate to trade? Jones and Kierzkowski (2001) demonstrate that higher costs for legal procedures in case of contact breach make firms less likely to implant themselves there. Levchenko (2007) also looks at institutional differences by focusing on the number of incomplete contracts, and finds that they have a negative impact on trade flows. Looking at a particular case regarding a Brazilian aircraft producer, Joppert (2012) shows that uncertainty in contracts makes firms more reluctant in creating partnerships with local suppliers. All of these studies attest to the fact that institutional quality plays a big role in explaining trade flows between countries. Once more we have to ask ourselves: what is the net impact of institutional differences on VAX?

We have seen that many factors come into play in explaining a country’s trade flows, but one begs the question: what is the net effect of increases in development levels on a country’s income stemming from participating in world trade? This leads us to presenting the research question of this thesis: did levels of development have positive implications on VAX growth between 2005 and 2015 for developing countries? By subtracting the imported intermediate goods used for the domestic production of exported final goods, VAX appears as a valid indicator capable of providing an answer to this question. As we have seen, factors such as human capital, transport infrastructure and institutional quality appear to be critical in explaining increases in exports and imports. However, we suppose that, for developing countries, higher levels of development will be associated to increased VAX, rather than the other way around. Which is why, with the objective of providing an answer to our research question, our first hypothesis will be that Increases in Human Capital, Transport Infrastructure and Institutional Quality will have a positive and statistically significant impact in explaining value added in exports growth.

II.3.2. Sector Characteristics

(15)

others, on those characterized by low technological content. This presence of heterogeneity at the sector-level, due to comparative advantage and specialization, shows us that differences in VAX should not merely be looked at from a country-level perspective. In order to provide a more complete answer to our research question, we decide to extend our analysis at a higher level of disaggregation of economic activity. Particularly, we are interested to see if higher levels of human capital have a bigger effect on a sector’s value added in exports if that sector is capital-intensive10. We choose to test the link between levels of human capital and the capital-intensity of sectors for the following reason: capital-intensive sectors have relatively low ratio of labour inputs and generally requires skilled labour. This is because the workers managing the machines require sufficient skills and training to be able to operate them. For this reason, we predict the following in our second hypothesis: Increases in levels of human capital will have a higher impact in VAX growth of capital-intensive sectors, than non-capital-intensive sectors. In the following section we relate what has been found in the literature to our model.

III. Methods and Data

III.1. Methodology

III.1.1. A simple representation of an econometric model adapted to our study

The main objective of this work is to test the eventual causal relationship between levels of development and changes in value added in exports. Based on what was discussed in Section II, we expect this variable to be positively affected by higher levels of development. In other words, we will be looking at causality from the various country characteristics on VAX growth. We will be testing this potential causal relationship between 2005 and 2015, at both the country (hypothesis 1) and sector (hypothesis 2) level.

The country characteristics explained in the previous section –human capital, transport infrastructure and institutional quality– will serve as our explanatory variables. In Equation (1) below, we present a basic representation of a linear econometric model, in order to understand how causality between these three development indicators and VAX growth will be tested. On the “left hand side” of our model, is VAX growth (𝑉𝐴𝑋_𝑖𝑠𝑡), which will serve as our dependent variable, i.e. the variable whose variation in outcome is being studied. The subscripts 𝑖, 𝑠 and 𝑡 identify the country, sector and period of the observations respectively. The “right hand side” of the regression model includes the independent variables, i.e. the variables that (we suppose) might explain variations in the dependent variable. In Equation (1) 𝐻𝐶𝑖𝑡 stands for human

capital, 𝐼𝑄_𝑖𝑡 represents institutional quality and 𝑇𝐼_𝑖𝑡 denotes transport infrastructure quality, with the three variables showing variation at the country-level and between years, but not at the sector-level. The 𝐶𝐼_𝑠 variable represents the capital-intensity of sector 𝑠 and does not vary neither per country nor per year. The particularity of this variable is that it is a dummy variable, i.e. a variable that is equal to one in the presence of some categorical effect, and zero if not. Finally, in our study we will also include control variables, which are a specific type of independent variable: although they are not variables of interest for our research question, we

10_{Capital-intensive sectors are those whose productive processes require a high percentage of investment in fixed}

(16)

also expect them to explain changes in the dependent variable. Therefore, their inclusion is mainly to control for the other regressors. The 𝑢_𝑖𝑠𝑡 term is known as the error term, and its value displays the margin of error within a statistical model (i.e. all the variation in VAX growth that is not explained by the explanatory variables).

𝑉𝐴𝑋𝑖𝑠𝑡 = 𝐻𝐶𝑖𝑡× 𝛽1+ 𝐼𝑄𝑖𝑡× 𝛽2+ 𝑇𝐼𝑖𝑡× 𝛽3+ 𝐶𝐼𝑠× 𝛽4+ 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 × 𝛿 + 𝑢𝑖𝑠𝑡 (1)

The scale of the impact the three independent variables have on VAX growth is represented by 𝛽. For example, in the case of a linear Ordinary Least Squares regression 𝛽 is the degree of change in the outcome variable for every 1-unit change in the independent variable. It is these estimated values for 𝛽1, 𝛽2 and 𝛽3 that will allow us to test our first hypothesis. Recalling the

hypothesis, we predicted that all three development indicators would have a positive and statistically significant impact on VAX growth. Ergo, our hypothesis would be confirmed for values of 𝛽 larger than zero. If the 𝛽 values were to be quasi null or negative, our hypothesis would be rejected since the indicators would not be positively associated with VAX growth. Even in the case of positive 𝛽 values, if the causal relationship (given our data) were to not be strong enough –i.e. not statistically significant– our hypothesis would also be rejected. Regarding the control variables, as we will see in the subsection below in which we present our model choice, we include population growth and time dummy variables11. However, the estimated values of 𝛿 are not of direct interest for providing an answer to our hypotheses, with the main interest of their inclusion being to increase the explanatory power of our model. Finally, 𝛽₄ is the estimated coefficient of our dummy variable indicating whether the VAX value on the left hand side of the equation is of a sector that is capital-intensive or not. The estimated coefficient of this type of variable is to be interpreted in the following way: if 𝛽₄ were to be positive and statistically significant, it would mean that VAX growth is higher for capital-intensive sectors than the reference group (i.e. non capital-capital-intensive sectors). However, this coefficient will not be useful in providing an answer to either hypothesis structuring this study. How can we estimate the effect higher levels of human capital have on VAX growth, subject to the capital-intensity of a sector? Equation (2) displays the additional term that will allow us to test for this:

𝑉𝐴𝑋𝑖𝑠𝑡 = 𝜌𝑉𝐴𝑋𝑖𝑠(𝑡−1)+ 𝐻𝐶𝑖𝑡 × 𝛽1+ 𝐼𝑄𝑖𝑡× 𝛽2+ 𝑇𝐼𝑖𝑡× 𝛽3+ 𝐶𝐼𝑠× 𝛽4+ (𝐶𝐼𝑠× 𝐻𝐶𝑖𝑡) × 𝛽5+

𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 × 𝛿 + 𝑢𝑖𝑠𝑡 (2)

By including an interaction term between human capital values and whether a sector is capital intensive or not, our model is now able to provide an answer to our second hypothesis. As a reminder, we had predicted higher levels of human capital to have a positive and stronger impact on VAX growth of sectors with high- or medium high capital-intensity, as compared to those that are not. Therefore, our second hypothesis would be confirmed for a positive and statistically significant value of 𝛽₅, while null or negative values would lead us to reject the hypothesis. In sum, for our first hypothesis to be accepted 𝛽1, 𝛽2 and 𝛽3, should be positive,

and for our second hypothesis, 𝛽5 should be positive. Based on these results, we will be able to

assess whether, at the country- and sector-level, levels of development had positive implications on VAX growth for developing countries between 2005 and 2015.

(17)

III.1.2. Discussion on variable choice

Since levels of educational attainment are the most used proxy for human capital, authors typically rely on enrolment rates (see Blyde, 2014) or number of graduates. However, enrolment rates have a series of shortcomings that might render the study less precise. Enrolment rates are an indicator of input in the educational system, since it is obtained by dividing the number of students of a particular age group enrolled in all levels of education by the size of the population of that age group12. A more fitted indicator of human capital would be one providing information on output; that is, on the level of education already attained by a country’s population. An indicator capable of providing this sort of information is the average years of schooling of a population.

Another indicator of human capital well suited for our study is the percentage of a population with a certain level of educational attainment (e.g. completed secondary school, completed tertiary education). By distinguishing between different levels of educational attainment, the usefulness of this indicator lies in providing various types of skill. For example, an individual having completed secondary school could be classified as being low-skilled labor, while completed tertiary education would classify that person as medium- or high-skilled labor. Since our study consists in looking at developing countries however, at higher levels of educational attainment, there is very little (or no) variation in these percentages in public datasets.

An option that does not suffer from this issue is the average years of schooling of a population over 25. At this age, we could suppose that people having pursued tertiary education to have completed it (or be close to concluding it). The advantage of this is that we do not need to account for lagging effects of educational attainment since we are looking at a population that is (supposedly) already in the labor market. A drawback of this indicator is that we are unable to distinguish between levels of human capital, be it low-, medium- or high-skilled. Another potential drawback is the low variation in annual values, since only the population reaching 25 years of age in a year would change values in the indicator. Due to the lack of sufficient data for the percentage of population over 25 at higher levels of educational attainment for developing countries, we choose to look at average years of schooling of a population over 25 years of age.

The second explanatory variable included in our model is Institutional Quality. The difficulty in selecting variables for this sort of indicator is twofold. Firstly, in the way institutions is to be defined. As explained in Section II, we consider institutions to be “humanly devised”, as setting the “rules of the game” that shapes human activity, and having an effect on society via incentives. Therefore, our chosen variables should comply with these characteristics. The second difficulty lies in quantifying something that could be considered as intangible and therefore hard to measure. Prominent economists that have included indicators of institutional quality in econometric analysis are Acemoglu et al (2010), which distinguish between two types of institutions: contracting institutions and property rights institutions. The former looks at institutions that ensure the full realization of contracts allowing to decrease the risk of contract breaking between two or more parts, while the latter refers to the people’s right to private capital

(18)

and protection against expropriation. When wanting to include indicators that proxy for institutional quality, we choose to follow their strategy and definitions. This is why we include an index of Rule of Law as a proxy for the quality of property rights institutions, and an index of Regulatory Quality as a surrogate for the quality of contractual institutions; both are provided by the World Bank Database. Seeing as our chosen indicators fulfill the three characteristics of institutions as defined by North, we consider them fit as proxies for institutions.

Regarding Institutional Quality, a remark on the heterogeneity of our country sample is necessary. Our panel dataset consists of 32 countries distributed across all continents, and classified as lower income and upper middle income by the OECD. Therefore, it is reasonable to expect differences in institutional quality between countries in Latin America or Southern Asian, and those in the European continent for example. When doing this sort of study on developing countries however, problems like these are to be expected.

When wishing to estimate the effect that a country’s quality of transport infrastructure has on VAX growth, which transport methods should be considered? The ideal model would be to include proxies for all methods of cross-country trade: land, sea and air. Starting with the latter, information on the number of specialized airports for trade would have been a useful indicator to include. However, due to the scale of investment needed to construct such an airport, this sort of indicator would probably show very little variation in time. Therefore, it would be better adapted to a cross-section study rather than a time-series one. Since we will be working with panel data, this variable would lose its explanatory power. Furthermore, public data on these airports specialized for international trade is unavailable for our chosen sample of developing countries. Regarding quality of transport infrastructure by land, we face similar issues. One could include a variable measuring the investments in highways or the amount of bridges connecting countries, with the intention of reducing the travel time and transport costs between neighboring trading countries. Once more, this level of detail for data on developing countries is not easy to come by. Furthermore, contrary to air and sea transport methods for cross-country trading which require specific investments, it is difficult to separate land transport infrastructure specialized in trading from population use.

This leaves us with transport infrastructure for trade via sea. An indicator that has been used to proxy for measuring infrastructure quality and transportation efficiency is an index of port quality (Blyde, 2014). Not only can this variable directly be associated to trade, data is also available for many developing countries. To represent the countries’ performance in this, we chose to include the following variable from the World Bank Database: Quality of Port infrastructure. This variable will serve as proxy for the level of development of trade transport infrastructure. By looking at one of the main sources of goods transportation across borders, we will be able to capture at least a part of the effect of better infrastructure and efficient transport methods on VAX growth.

(19)

Since we wish study the impact of development on VAX at a country level, but also at a sector-level, it will be in our interest to classify sectors in the model regarding the level of capital-intensity used. To do this, we look at the International Standard Industrial Classification of all economic activities (UNIDO, 2010) which classifies industries by capital-intensity. This will enrich our study by enabling our model to distinguish between sectors, testing the hypothesis of human capital levels on VAX growth, contingent on sector particularities. Having described the relationship we expect country and sector characteristics to have with VAX growth, and discussed the choice of variables that will enable us to test our two hypotheses and shed some light on our research question, we now explain our model choice.

III.1.3. Model Description

Having specified the dependent, independent and control variables that constitute our model, as well as the values of the estimated coefficients that would confirm (or conversely, reject) our hypotheses, we now need to determine which model will be best fitted for our data. A Pooled Ordinary Least Squares (POLS) estimation model would not allow us to account for individual heterogeneity. This is because in the POLS method, information that observations come from the same country (or sector) or the same year is neglected. Additionally, since we are studying a multitude of countries and several sectors over a decade, it is unfeasible to account for all the explanatory variables. We might also encounter problems of omitted variable bias due to measurement difficulty regarding certain variables (e.g. cultural aspects of a population that might make them more productive). In order to be able to account for unobservable differences between these countries over time, a model better suited for our data than POLS should be used. Models that account for unobserved time-invariant individual characteristics, such as a Fixed effects or Random effects model, account for such issues. Country and Sector fixed effects will allow us to account for unobservable idiosyncratic time-invariant characteristics, increasing the explanatory power of the model. Additionally, when working with time-series, stationary tests must be conducted in order to ensure the avoidance of spurious relationships. However, these models are suitable for static panel data models, in which it is assumed that past values of the dependent variable do not affect current values. This might appear to be a problem for our study, since it is probable that past values of VAX, in part, determine a country’s current VAX. In order to confirm this, a test is run to observe any eventual serial correlation. There is evidence at a 0.01 level of significance that our data has first order serial correlation, confirming our suspicions of dynamic data. Thus, we dismiss both the Fixed and Random effects models in order to study causation between our variables and VAX.

(20)

growth is taken into account. How does this estimation method function? On the right hand sign of Equation (1) displayed below we can see this parameter which is displayed as 𝜀_𝑖 (country) and 𝜖_𝑠 (sector). The X vector containing all explanatory and control variables detailed in previous paragraphs is multiplied by the coefficients vector 𝛽; there is also an additional error term 𝑢𝑖𝑠𝑡 which is time-variant. What the estimator does is to take the difference between

observations throughout the period, as shown by Equation (4), in order to eliminate the 𝜀_𝑖 and 𝜖_𝑠 terms. Is this model appropriate for our study? Are the underlying assumptions that attribute validity to the use of the model respected by our data? This is what we will be looking into in the following paragraphs.

𝑉𝐴𝑋_𝑖𝑠𝑡 = X𝛽 + 𝜌𝑉𝐴𝑋_{𝑖𝑠𝑡−1}+ 𝜀_𝑖 + 𝜖_𝑠+ 𝑢_𝑖𝑠𝑡 (3)

∆𝑉𝐴𝑋𝑖𝑠𝑡 = 𝑉𝐴𝑋𝑖𝑠𝑡− 𝑉𝐴𝑋𝑖𝑠𝑡−1 = ∆𝑋𝑖𝑠𝑡𝛽 + 𝜌∆𝑉𝐴𝑋𝑖𝑠𝑡−1+ ∆𝑢𝑖𝑠𝑡 (4)

Our first differences model will, by definition, calculate the difference between an observed value and its average, then calculate the difference between the values observed at the previous period with the average, and finally report the difference of these two measures. What does this mean regarding the interpretation of the estimates our model will provide us with, and how will this estimation model be advantageous for the study we wish to carry out? Controlling by the average VAX, our estimator takes out the trend our variable shows throughout the years in our panel, leaving us only with the shock from one year to the other. This is particularly useful for example, when handling potential spurious relationships between variables (see section III.2). In our model, time fixed effects will also be included as control variables, in order to isolate any effect that major macroeconomic events might have caused on all (or most) countries included. This last type of variable is particularly important for the time interval chosen for this study, since the economic crisis of 2008/2009 is included. The models used to provide an answer to our research question are the following13:

∆ln (𝑉𝐴𝑋_𝑖𝑠𝑡) = 𝜌∆ln (𝑉𝐴𝑋_{𝑖𝑠𝑡−1}) + ∆𝐸𝑑𝑢𝑐_𝑖𝑡𝛽₁+ ∆𝑅𝑒𝑔𝑅𝑜𝐿_𝑖𝑡𝛽₂+ ∆𝑃𝑜𝑟𝑡_𝑖𝑡𝛽₃+ ∆𝑃𝑜𝑝_𝑖𝑡𝛽₄+ ∆ 𝑇𝑒𝑐ℎ_𝑠𝛽₉+ ∆ ∑10_𝑗=1𝛿_𝑗𝑌_𝑗+ ∆𝑢_𝑖𝑠𝑡 (5) ∆ln (𝑉𝐴𝑋_𝑖𝑠𝑡) = 𝜌∆ln (𝑉𝐴𝑋_{𝑖𝑠𝑡−1}) + ∆𝐸𝑑𝑢𝑐_𝑖𝑡𝛽₁+ ∆𝑅𝑒𝑔𝑅𝑜𝐿_𝑖𝑡𝛽₂+ ∆𝑃𝑜𝑟𝑡_𝑖𝑡𝛽₃+

∆𝑃𝑜𝑝_𝑖𝑡𝛽₄+ ∆ (𝑇𝑒𝑐ℎ_𝑠× 𝐸𝑑𝑢𝑐_𝑖𝑡)𝛽5+ ∆ ∑10𝑗=1𝛿𝑗𝑌𝑗+ ∆𝑢𝑖𝑠𝑡 (6)

The results that are presented in Section IV stem from using equations (5) and (6) in order to estimate the 𝛽, 𝛿 and 𝜌 coefficients. The first model excludes the interaction terms, and tests only the independent effect of the regressors on the dependent variable. Since we are using a first difference estimator, our variables should all be interpreted as changes (∆). Therefore, the variables in equation (5) that are time-invariant (i.e. no subscript t) are not included in equation (6). The component on the left hand side ln (𝑉𝐴𝑋𝑖𝑠𝑡) is the dependent variable of our model and

represents VAX growth. As we can see, the variable is logged; this is to decrease the biasing effect outliers might have our estimates. These values differ per time period t, per country i and sector s. The variable 𝐸𝑑𝑢𝑐_𝑖𝑡 is our proxy for human capital and only changes per period and country. Our transport infrastructure indicator is 𝑃𝑜𝑟𝑡𝑖𝑡, which represents the Quality of Port

Infrastructure. Regarding our third type of variable –institutional quality–, we choose to include

(21)

Regulatory Quality and Rule of Law. However, due to the high correlation of the values taken by these two variables across our dataset, problems of multicollinearity might arise. It is with the intention of solving this issue that we choose to create a unique indicator calculated as the average of these two per country per year, represented above as 𝑅𝑒𝑔𝑅𝑜𝐿_𝑖𝑡14_{. In order to account}

for sector characteristics we include the dummy variable 𝑇𝑒𝑐ℎ_𝑠, which is equal to 1 if a sector is capital-intensive, and equal to zero if it is not. In order to test our second hypothesis, we will need to include an interaction term that can capture the potential different impact education levels have on VAX growth, depending on the country i and sector s. Finally, we include year dummy variables in order to capture time-specific effects on VAX growth by controlling for exogenous general macroeconomic shocks.

The estimation model that is used to analyze causation in our data is the two-step first difference Generalized Methods of Moments estimator, developed by Arellano and Bond in 199115_{. Since}

we will be using a one period lag of the dependent variable as an independent variable, we will be running a first order autoregression. A consequence of the First Difference estimation model is the loss of one degree of freedom for every cross-sectional observation due to the time-demeaning (Wooldridge, 2008). It is worth mentioning that, of the totality of VAX observations between 2005 and 2015, seven appear to have negative value added in the OECD’s ICIO16_.

Since no explanation is provided by the OECD regarding the reason behind these negative values, we choose to remove these observations from our dataset. We also choose to exclude the Private Households with employed persons sector, due to its values being seldom different from zero in the ICIO tables. Along with the introduction of instruments, our panel is reduced to 10860 observations, spanning 10 years (2006-2015).

III.2. Data

III.2.1. Setup of an Input-Output Table Framework

Before computing the VAX for each country and sector in the dataset, we explain the second part of the framework. Input Output tables (IOT) have been used for countless studies, such as analyzing trade between countries (Leontief, 1953), or looking at the extent to which countries are involved in GVCs (Timmer et al, 2013). How are these IOTs structured, and why do we choose to use them in our study? An IOT, as defined by the OECD, “describes the sale and purchase relationships between producers and consumers within an economy”. In other words, it is a quantitative representation of a production structure. A national IOT presents the value of the flow of goods for that economy during a particular year. Both domestically produced and imported goods can be used for the production of another good, consumed domestically or exported and consumed in another country. Since we focus our study on various countries, we use International Input Output tables; specifically, the Inter-Country Input-Output Tables (ICIO’s) provided by the OECD. These International IOTs are created by combining National IOTs with bilateral international trade flows, and provide information on the origin and destination of both imports and exports. In the OECD database there is data for 64 countries,

14_{See subsection III.4. for a more detailed explanation of this.}

15_{This option is available in the STATA program with the xtabond2 command, which is used for our model.} 16_{Chile, sector D27 (2009); Bulgaria, sector D09 (2010 & 2011); Brazil, sector D19 (2011,2012,2013) and Latvia,}

(22)

divided into 36 different sectors17_{plus one labeled “Rest of the World”, which covers the}

remaining part of the world economy. The outline of an ICIO is as illustrated in Figure III.2.1 below. There are 36 rows for every country, each one indicating the use of products from that particular sector and country. The columns indicate the values of all inputs used in the production activities of a sector, also divided into 36 sectors for every country. Said differently, each element in this “square” represents the money value of the delivery of products from a specific country and industry, to another country and industry. Each good stemming from every country and sector can also be consumed as final products domestically or in other countries, and is reflected in the Figure below in the “Final Demand” column18_{. The second section}

figuring below the production matrix “intermediates input” of the IOT framework includes total value added by labor and capital for each country and sector. The final column displays the total use of a country and sector’s production, while the final row shows the gross output of each sector. These must be equivalent to each other in each sector’s sum.

The usefulness of these tables for the calculation of the VAX indicator, and providing an answer to our hypotheses, is threefold. Firstly, by looking at cross-sector supply (and use) of intermediate inputs, the IOTs appear as a source of data adapted to a world with GVCs: we are able to take into account the import of intermediate inputs from other countries, used in the production of a country’s exported good. Secondly, these tables show the destination of the exported goods. This is a crucial aspect, as where the exported good is consumed has important implications on a country’s VAX. Thirdly, the value added is shown for every one of the 36 sectors, allowing us to look into sector characteristics such as degree of capital-intensity and test our second hypothesis. In the following sub-section, we look into how these inter-country IOTs can be used to compute the VAX indicator.

FIGURE III.2.1. - OECD INTER COUNTRY INPUT OUTPUT TABLE

17_{See Appendix A for list of sectors.}

18_{For simplicity, in the graph we show total final demand. However, the OECD distinguishes between six types}

of Final Demand: Household final consumption expenditure (HFCE), Non-Profit institutions serving households (NPISH), General Government final consumption (GGFC), Gross fixed capital formation (GFCF), Changes in inventories and valuables (INVNT) and direct purchases by non-residents (P33).

19_{Current prices do not account for inflation.}

Layout of an OECD ICIO table

(In millions of current US Dollars)19 Intermediates use

Final Demand (by countries) Total Use Country 1 … Country 65 Country 1 … Country 65 Sector 1 … Sector 36 Sector 1 … Sector 36 Supply from country-sectors Country 1 Sector 1

MATRIX INDICATING THE MONEY VALUE OF THE DELIVERY OF INPUTS FROM COUNTRY i AND INDUSTRY j, TO COUNTRY i’ AND INDUSTRY j’

MATRIX INDICATING THE MONEY VALUE OF THE DELIVERY OF PRODUCTS

FROM COUNTRY i AND INDUSTRY j, TO FINAL DEMAND OF COUNTRY i’

OUTPUT AT BASIC PRICES … Sector 36 … Country 65 Sector 1 … Sector 36

Value Added + Taxes – Subsidies on

Intermediate Products VALUE ADDED AT BASIC PRICES Gross Output

(23)

III.2.2. Dependent Variable: Value Added in exports

Now that we have presented the schematic outline of an international IOT, how can it be

used to compute VAX? The methodology followed in this work is inspired by a

decomposition technique introduced by Leontief (Leontief, 1936) in which the economy is divided into specific industries. In order to build our VAX indicator, we require matrix calculations using elements from the ICIO tables. Before presenting the equation of how VAX is calculated however, it is necessary to explain how the elements in the ICIO tables are put under matrix form. Assume that we dispose of 𝑁 countries and 𝑀 sectors. Denote Z the matrix of intermediate deliveries between sectors, F the matrix of final demand and 𝐱 as the output vector. What are the dimensions of said matrices? Since we dispose of 𝑁 countries and 𝑀 sectors, the dimension of intermediate deliveries Z will be a square matrix of 𝑁𝑀 × 𝑁𝑀. The

F matrix shows each country’s final demand of goods stemming from every country and sector

in the world economy. Therefore, it will have a dimension of 𝑁𝑀 × 𝑁. Finally, the output vector 𝐱 will be of dimension 𝑁𝑀× 1, with each element representing the total use of inputs from all sectors, in a sector’s production of goods.

𝐙 = [ 𝐳11 _⋯ ⋮ ⋱ 𝐳1𝑐 ⋮ ⋯ 𝐳1𝑛 ⋰ ⋮ 𝐳𝑐1 _⋯ _𝐳𝑐𝑐 _{⋯ 𝐳}𝑐𝑛 ⋮ ⋰ 𝐳𝑛1 _⋯ _𝐳⋮𝑛𝑐 _⋯⋱ _𝐳𝑛𝑛⋮ _] , 𝐅 = [ 𝐟11 _⋯ ⋮ ⋱ 𝐟1𝑐 ⋮ ⋯ 𝐟1𝑛 ⋰ ⋮ 𝐟𝑐1 _⋯ _𝐟𝑐𝑐 _{⋯ 𝐟}𝑐𝑛 ⋮ ⋰ 𝐟𝑛1 _⋯ _𝐟𝑛𝑐⋮ _{⋯ 𝐟}⋱ 𝑛𝑛⋮ _] , 𝐱 = ( 𝐱1 ⋮ 𝐱𝑐 ⋮ 𝐱𝑛 )

Taking the case of a particular country 𝑐, element 𝑧_𝑗𝑘𝑐𝑛 of the Z matrix provides the monetary value of intermediate inputs from sector 𝑗 in country 𝑐 to sector 𝑘 in country 𝑛 (in millions of US dollars). Element 𝐟_𝑗𝑐𝑛of matrix F yields the deliveries from sector j in country 𝑐 for final consumption in country 𝑛. The element 𝑥_𝑗𝐶 of vector 𝐱 yields the output of sector j in country 𝑐. The market clearing condition, i.e. that the supply of goods be equal to the consumption of those goods, demands that 𝐱 = 𝐙𝐩_𝑁𝑀+ 𝐅𝐩_𝑁, where 𝐩_𝑁𝑀 is the summation vector consisting of ones, and of dimension 𝑁𝑀 × 1. The input coefficients of the matrix with dimension 𝑁𝑀 × 𝑁𝑀 are presented under the form of 𝐀 = 𝐙𝐱̂−1, meaning that 𝑎_𝑗𝑘𝑐𝑛 = 𝑧_𝑗𝑘𝑐𝑛/𝑥_𝑘𝑛. These elements yield the intermediate deliveries per unit of the receiving industry’s output. Subsequently, we can substitute 𝐙𝐩𝑁𝑀 = 𝐀𝐱 in the market clearing equation so that 𝐱 = 𝐀𝐱 + 𝐅𝐩𝑁, which can

also be expressed as 𝐱 = (𝐈 − 𝐀)−1_𝐅𝐩

𝑁 = 𝐋𝐅𝐩𝑁 (7)

, with the 𝑁𝑀 × 𝑁𝑀 matrix 𝐋 ≡ (𝐈 − 𝐀)−1 known as the Leontief inverse. Each of its elements represents the amount of input of a certain sector required to satisfy one unit of Final Consumption (in US dollars).

THE IMPACT OF COUNTRY CHARACTERISTICS ON TRADE PERFORMANCE