Groningen, 18.06.2019
Master International Economics
Does Structural Change in Manufacturing Subsectors Entail Potential for Manufacturing Productivity Growth in Developing Economies?
Tim Morawietz (S3629570)
Abstract
This thesis addresses the research question, whether and to what extent developing econo- mies have been able to establish manufacturing subsectors as a way to boost productivity growth between the years 2000 and 2012. The thesis also looks specifically at structural change towards technologically sophisticated manufacturing subsectors and contributions from labor reallocation to productivity growth in these sectors. For this purpose, I constructed a data set indicating structural change in manufacturing subsectors on the 4-digit level for 34 economies. The thesis is motivated by the recent debate on Premature Deindustrialization and the potentials entailed in manufacturing as a way to lead to sustainable economic growth.
Increasing contributions from structural change (the “between effect”) to productivity growth in manufacturing subsectors will be interpreted as an indication for industrialization. The be- tween effect increases if either employment in manufacturing increases or productivity in- creases. Both of these variables have been taken as an indication for a country’s ability to industrialize by scholars in the past (Rodrik, 2015). This links the idea of Premature Deindus- trialization to the contributions from structural change to productivity growth.
The main findings are that the countries of the sample have not experienced large contribu- tions from structural change to productivity growth by reallocating labor between manufac- turing subsectors. Countries at the lower end in terms of GDP per capita have experienced even fewer contributions from structural change to productivity growth, supporting the idea of Premature Deindustrialization. The thesis also finds that increasing levels of GDP per capita, educational levels, FDI inflows or an increasing Trade Openness have promoted the role of structural change for productivity growth in manufacturing subsectors.
The thesis concludes that labor reallocation between manufacturing subsectors in general
seems to entail limited potential for boosting productivity growth rates in manufacturing sub-
sectors. This is in line with the conclusions from Premature Deindustrialization. The thesis fur-
ther implies that research on agro-industrial sectors should be done to elaborate on develop-
ing economies’ future perspectives in more detail.
Content
1. Introduction
2. Literature Review and Hypothesis
a. Literature Review Structural Change
b. Literature Review Expected Coefficients c. Hypothesis
3. Data and Methodology
a. Decomposition Method
b. Reflection on Data Quality and Alternatives
c. Descriptive Statistics
4. Econometric Analysis
a. Relative Structural Change
b. Technology Intensive Structural Change
5. Limitations and Conclusion
6. References
7. Appendix
8. Endnotes
1
1. Introduction
Have developing economies been able to shift employment to manufacturing subsectors to increase domestic productivity growth in manufacturing subsectors? In other words: have de- veloping economies experienced productivity enhancing labor reallocation in manufacturing subsectors, which has been defined as industrialization (Rodrik, 2015)? Industrialization has frequently been referred to as the Engine of Growth and an important feature for a developing economy to obtain middle-income status. This idea is based on observations throughout his- tory. As shown by Maddison, countries did not differ substantially in terms of GDP per capita in the 1500s, but Western economies experienced significant increases in terms of GDP per capita between 1500 and 1820, thus during the European industrial revolution (Maddison, 2005). In contrast to Western experiences, Asian, Latin American and African economies stag- nated in their GDP per capita levels in the same time frame, resulting in a diverging global economic development across countries and regions. Some South-East Asian economies did manage to industrialize beginning in the 1950s, triggering substantial economic growth rates.
All in all, empiric observations underline the correlation between industrialization and eco- nomic growth.
Industrialization, and in particular manufacturing, theoretically entails a wide set of ad- vantages for developing economies, such as:
• Manufacturing’s ability to absorb both skilled and unskilled labor (Page, 2012)
• Manufacturing’s high degree of interconnectedness within the domestic economy (Page, 2018)
• Manufacturing providing potential for technological spillovers between producers and knowledge spillovers and absorption from interactions with foreign producers and suppliers (Rodrik, 2015)
However, low income economies nowadays have not been able to benefit from industrializa- tion to the same extent as South-East Asian economies did 50 years ago. Dani Rodrik analyzed this development in 2015, referring to the reduced potential from industrialization for eco- nomic development with the term “Premature Deindustrialization”. His main finding is that developing economies in the post-1990 era have deindustrialized at an earlier stage as meas- ured in employment in manufacturing and in real value added of manufacturing as a share of domestic GDP, relative to the USA. An additional finding by Rodrik is that the decrease in em- ployment in manufacturing is particularly present in low-skill-intensive manufacturing.
Rodrik, as well as other scholars, names reasons based on a country’s domestic environment and economic structure for Premature Deindustrialization. Examples for these internal dy- namics include:
• Weak institutions and developing economies’ inability to exert political power in order
to build up traditional manufacturing-based industries (Frankema, 2018; Rodrik, 2015)
2
• A shifting domestic demand for services rather than manufactured goods beyond a certain point in a country’s level of income, potentially disincentivizing a shift towards industrialization and manufacturing (Rodrik, 2015)
• The process of substituting low-skilled labor with technology-based capital-intensive production schemes, stripping low income countries of their comparative advantage (Rodrik, 2015)
As well as a country’s external environment and its situation vis-à-vis trading partners, such as:
• A persisting low wage level in South-East Asian economies maintaining South-East Asian predominance in manufacturing sectors due to a long-lasting comparative ad- vantage, based on low labor costs (Frankema, 2018)
• A re-domestication of production to industrialized Western economies due to an in- creasingly capital and technology intensive production (McMillan, 2017)
Contrary to this view of limited potential entailed in manufacturing and industrialization for developing economies, authors such as Monga (2010), or Hallward-Driemeier (2018) stress the fact that (some) manufacturing subsectors still entail the potential for developing econo- mies to increase productivity growth and promote economic growth.
Scholars usually address the question of whether economies have been industrializing by con- ducting studies at a high level of aggregation, without differentiating between technology in- tensive and less technology intensive industrialization. As one main goal in scientific research is to give policy advice, scholars frequently conduct analysis on an economy-wide level to take interlinkages between different sectors into account. Researchers do not typically analyze de- velopments and structural change in only one particular sector. Thus, there is a knowledge gap in the existing literature on structural change and the establishment of specific manufac- turing subsectors or technologically sophisticated manufacturing subsectors.
This thesis addresses the research question, whether and to what extent developing econo-
mies have been able to establish manufacturing subsectors as a way to boost productivity
growth between the years 2000 and 2012. The thesis also examines structural change towards
technology intensive sectors in particular, due to Rodrik’s observation that employment de-
creases occur mainly in low-skill intensive manufacturing. Therefore, the thesis also addresses
the question as to which factors and country endowments have shown to be beneficial for
structural change towards manufacturing subsectors and technologically sophisticated man-
ufacturing subsectors. Based on employment and value added data provided in the UNIDO
Indstat 4 Revision 3 dataset, I calculate structural change within the manufacturing sector at
the 4-digit level. This average annual structural change is taken, relative to total productivity
increases. Therefore, the dependent variable Relative Structural Change indicates, whether
3
labor reallocation between manufacturing subsectors has been driving a country´s productiv- ity growth in manufacturing. In order to assess potentials entailed in technologically sophisti- cated subsectors, the thesis calculates the portion of structural change towards technology intensive subsectors as a second dependent variable in a separate regression.
The thesis adds to the existing literature by looking specifically at structural change in manu- facturing subsectors. The thesis’ main contribution to the existing literature is the construction and analysis of a data set at the 4-digit level. Some of the main results include that:
• a combination of dynamics based on a country’s internal environment and its ties to the global economy is decisive for explaining structural change between manufactur- ing subsectors, and structural change towards technologically sophisticated sectors in particular. Only considering one dimension, either the internal or the external environ- ment in isolation, does not sufficiently explain observable patterns of structural change. A country’s Trade Openness Growth, FDI inflows, educational system and level of development show to be the most stable predictors for the degree to which struc- tural change between manufacturing subsectors contributes to productivity growth.
• Structural Change does not trigger productivity growth in manufacturing subsectors in my sample. The contributions from structural change to productivity growth are spread around the value of zero.
• A pattern of reduced structural change beyond a certain point in terms of GDP per capita, in accordance with the findings of Premature Deindustrialization, can be ob- served in my data set on manufacturing subsectors at the 4-digit level. In conclusion, labor reallocation in manufacturing subsectors does not seem to contribute to produc- tivity growth. This questions the potential entailed in manufacturing for economic de- velopment in low income economies.
The results for the role of structural change for explaining productivity increases in manufac- turing subsectors are in line with results obtained in previous studies for structural change towards the industrial sector in general. However, further research on this topic needs to be undertaken with a larger data set to confirm my results and conclude on structural change enhancing factors towards manufacturing subsectors specifically.
The thesis is organized as follows. After the introduction, I begin with a literature review,
systematically analyzing the problem of Premature Deindustrialization by examining both re-
gion-specific experiences with structural change and factors, promoting or hindering industri-
alization in two separate parts. Ultimately, the literature review results in the thesis’ hypoth-
esis. Section 3, the data and methodology section, introduces the structural change decom-
position technique applied to construct the thesis’ data set. The section further introduces
main decomposition results as well as a descriptive analysis of the data set. The subsequent
4
econometric analysis in section 4 regresses several independent variables to test the hypoth- esis. Section 5 concludes on the findings and the research question posed in the beginning.
2. Literature Review and Hypothesis
In the first stage, the following literature review briefly introduces region specific experiences with structural change. Revising this literature underlines the limited role of structural change for explaining increasing productivity growth in manufacturing. Therefore, the first part of the literature review serves as a problem analysis. In the second stage, the literature review ex- amines factors that have shown to determine structural change towards the industrial sector.
a. Literature Review Structural Change
The comparison of different studies and region-specific experiences with structural change brings together a set of analysis on varying levels of aggregation and geographic regions. The studies also differ in the way they calculate the between effect in a productivity growth de- composition. Regardless of the respective way the between effect is calculated, they all define the between effect as labor, reallocating between different sectors, thereby contributing to productivity growth in an economy. Note that these studies do not focus on productivity growth in the manufacturing sector but are conducted on an economy-wide level. De Vries et.
al (2012) is an exception, where the within effect specifically indicates the within effect of the industrial sector. Thus, this comparison should not show equal quantifications of structural change between different studies. Instead, it underlines that a limited role from structural change for productivity growth can be observed, regardless of the level of aggregation or de- composition method used.
Table 1: Past Experiences with Structural Change towards Industrialization Author Level of Aggre-
gation
Time Region Within Effect Between Ef- fect
De Vries et.
al (2012)
i3 sector level 1995 - 2008 Brazil 0.2% 0.6%
4 sector level 1995 - 2008 Russia 1.2% 0.9%
3 sector level 1991 - 2008 India 0.9% 0.9%
3 sector level 1997 - 2008 China 4.4% 1.2%
McMillan et.
al (2013)
10 sector level 1990 - 2005 Asia 3.31% 0.57%
10 sector level 1990 - 2005 LA 2.24% - 0.88%
10 sector level 1990 - 2005 SSA 2.13% - 1.27%
Timmer et. al (2014)
10 sector level 1990 - 2010 SSA 1.7% 0.1%
10 sector level 1990 - 2010 Asia 3.1% 0.5%
10 sector level 1990 - 2010 LA 1.0% - 0.1%
Yilmaz (2016) 10 sector level 1950 – 2005 MIT 1.45% 0.48%
10 sector level 1950 - 2005 NMIT 3.7% 0.67%
7 Sector Level 1999 - 2008 Asia 3.3% 2.0%
7 Sector Level 1999 – 2008 LA 0.6% 0.2%
5
Roncolato
and Kucera (2014)
7 Sector Level 1999 – 2008 MENA 1.8% - 0.6%
7 Sector Level 1999 - 2008 SSA 1.9% 0.5%
LA = Latin America, SSA = Sub Sahara Africa, MIT = Middle Income Trap Countries; NMIT = Non-Middle-Income Trap Coun- tries; MENA = Middle East Northern Africa
This limited role from structural change for productivity growth is in line with the assumed and observed pattern of Premature Deindustrialization. The trend across different studies also indicates that Asian economies seem to have been experiencing the highest productivity growth rates with an above-average between effect in Asian economies. Contrary to these experiences, the role of structural change for explaining productivity growth has been small in Latin American economies. Some other key take-aways from these and other studies in the literature include the following points, supporting the claim of an ongoing assumed dynamic of Premature Deindustrialization:
• Premature Deindustrialization cannot be seen in all countries unambiguously. Some countries seem to have been able to industrialize, above all China. Contrary to the Chi- nese experience, Brazil seems to have been struggling with industrialization (De Vries, 2012). This observation is in line with McMillan et. al (2017) who refer to Brazil as a
“postindustrial country”, claiming that Brazil is unable to reindustrialize as wages have risen too high in order for the economy to reestablish a strong industrialized sector, built on low labor costs. De Vries et. al’s large between effect stems from labor reallo- cation from agriculture to services and is not due to industrialization.
• Middle Income Trap Countries (MIT), seem to experience smaller productivity growth gains as opposed to Non-Middle Income Trap Countries (NMIT) (Yilmaz, 2016). Accord- ing to Yilmaz, this productivity growth gap is primarily driven by the manufacturing sector and can be traced back to both the within and the between effect. Thus, the inability to establish a strong manufacturing sector prevents countries to advance to- wards higher income status.
• The service sector has been working as an engine for economic growth, contributing to increasing productivity growth in a large share of developing and middle-income countries. This further promotes the idea of limited potential entailed in the manufac- turing sector (Roncolato and Kucera, 2014).
• In a study on long run developments between 1960 and 2010, Szirmai and Verspagen find that employment in African agriculture decreased by 16.2%, 15.2% of which trans- ferred into the services sector and only 1% transferring to the manufacturing sector (Szirmai and Verspagen, 2011)
• Verspagen shows that countries at lower levels of development in terms of GDP per
capita have experienced modest structural change towards manufacturing in general,
and even less structural change towards technologically sophisticated products in par-
ticular (Verspagen, 2014).
6
However, the mere observation of developing economies’ inability to industrialize and the limited contributions from structural change to productivity growth alone does not address the question, which factors potentially explain structural change in some regions as opposed to lacking structural change in other regions. Thus, for a complete problem analysis, leading to a sound hypothesis and a research question, the factors triggering premature deindustrial- ization across different regions need to be analyzed. This aspect will be addressed in the fol- lowing section to give a complete picture of the ongoing dynamics in the process of deindus- trialization.
b. Literature Review Expected Coefficients
As indicated in the introduction, an analysis of factors, contributing to structural change differs between dynamics, based on a country´s internal environment, as well as a country´s ties to the global economy, thus the external environment.
The Internal Environment and Structural Change
The economic theory for a correlation between structural change to manufacturing and a country’s level of development goes back to Kaldor’s Growth Laws (Kaldor, 1966). In the En- gine of Growth hypothesis, Kaldor states that manufacturing sectors, characterized by econo- mies of scale and highly interlinked with remaining parts of the domestic economy, generate additional demand for non-manufacturing sectors and thus serve as an engine of growth.
Thus, as productivity in manufacturing increases and labor reallocates towards the manufac- turing sector, GDP per Capita is expected to increase.
Empirically, a number of different scholars have investigated the role of GDP per capita for structural change and industrialization. Some of the most important findings are listed below:
• Alexiou (2010) provides evidence for the dynamic of Kaldor’s laws, showing that man- ufacturing has worked as the engine of growth, increasing GDP levels in five Mediter- ranean economies between 1975 and 2006.
• There seems to be an inverse U-shaped relationship between a country’s level of de- velopment, measured in GDP per capita, and its structural change towards manufac- turing (Herrendorf et. al, 2013; Rodrik, 2015). Thus, as a country develops in terms of GDP per capita, structural change towards manufacturing increases. From a certain (breaking) point onwards in a country’s level of development however, a country be- gins to deindustrialize.
These findings suggest that increasing GDP per capita is expected to increase structural change
towards manufacturing, at least until reaching a certain breaking point. I therefore expect a
positive coefficient from increasing GDP per capita levels on structural change. This will be
further elaborated in section 3 and in figure 3 at a later stage, when analyzing my data set.
7
Another decisive factor expected to increase structural change both in general, and towards manufacturing subsectors in particular, is labor market flexibility. Labor reallocation across different sectors and towards manufacturing subsectors should benefit from a less restrictive labor market. Additionally, a more restrictive labor market may incentivize a more capital- intensive production or the evolvement of a larger informal sector, as labor may become a less attractive input good for employers and producers. Hence, I expect increasing flexibility to have a positive effect on a country’s structural change. The role of labor market flexibility for structural change has also been addressed in a number of empiric studies:
• Buadi-Mensah et. al show that a less restrictive labor market fuels structural change in African economies (Buadi-Mensah et. al, 2018).
• The Asian Development Bank further supports the assumption that more flexible labor markets tend to promote structural change. The authors refer to the limited contribu- tions from structural change to productivity growth in India, supporting this line of ar- gumentation, as India, as opposed to other developing economies such as Korea, has a more restrictive labor market, measured by labor market regulations (McGregor, 2016).
Additional factors such as tenure security for land are also being used to underline the gov- ernment’s role in interfering with the market and thereby incentivizing or disincentivizing structural change in different studies. As tenure security may however work both ways, in favor or against structural change, the effect from labor market flexibility seems less ambigu- ous as compared to tenure security.
When revising suggested reasons for Premature Deindustrialization in the introduction, polit- ical inability to exert power and enforce a coherent national strategy is expected to hinder structural change towards industries and manufacturing (Frankema, 2018). Motivated by this statement, I include a variable indicating a country’s degree of Fundamentals. Including a di- mension for fundamentals into the analysis is coherent with research on structural change:
• McMillan et. al come to the conclusion that while institutions are deemed decisive for a country’s ability to pursue structural change in general, it remains difficult to deter- mine which institutions exactly enable structural change. Instead, they find that the decisive institutions depend on the individual country (McMillan et. al 2017).
• Nicet-Chenaf et. al (2018), shows that institutions have an impact on the ability to per- form structural change in particular at lower levels of country development in terms of GDP. A comparison of structural change experienced in Asia and Africa and institu- tions present in those regions underlines their argument.
Furthermore, better institutions typically correlate with a higher quality level of education and
levels of GDP, thereby supporting the movement of workers out of agriculture in developing
economies.
8
A certain degree of education among the workforce should be necessary in order to establish manufacturing subsectors. The role of education and the specific education proxy for explain- ing structural change may vary, depending on whether structural change in general, or struc- tural change towards technology intensive sectors is analyzed. Empirics support considering a proxy for education:
• In a cross-section analysis as well as in a longitudinal analysis, Martins shows the level of education across the workforce to have a positive significant effect on a country’s productivity enhancing structural change (Martins, 2018). Similar observations were made by Anand (2012) and McMillan (2014).
• In a country comparison between China and India and structural change experienced in both of these counties, Bosworth and Collins (2008) found China to have successfully established manufacturing in the domestic economy, while India was primarily shifting labor to services but also maintaining a large share of the workforce in the agricultural sector. The authors argue that the level of education across the Indian workforce in general was lower as compared to China, which may have been a contributing factor.
• Dabla-Norris et. al note the possibility that education may increase in importance for structural change towards manufacturing, as a country increases its share of manufac- turing in domestic GDP and enters high technology manufacturing (Dabla-Norris et. al, 2017).
The way the level of education across the workforce is being considered differs and there are a wide variety of possible proxies for measuring education in the literature. Table A2 in the appendix gives an overview of all education proxies tested.
Baumol’s hypothesis suggests that employment levels in the industrial sector decrease as productivity levels increase (Baumol, 1967). Furthermore, Autor shows that productivity in- creases in manufacturing sectors marginally reduce employment in those sectors but raise economy-wide employment levels, due to the high degree of interlinkages of manufacturing sectors in the domestic economy (Autor, 2017). However, scholars also track a different effect from increasing value added on employment, depending on technology intensity:
• As shown by Bessen, employment in technology intensive manufacturing sectors in- creased simultaneously to productivity increases, as long as there was sufficient de- mand for the manufactured products in European economies (Bessen, 2017).
This dynamic and the dependence on a large market promotes the idea to use manu- facturing value added in combination with a proxy indicating global demand effects to predict rising employment and structural change in technology intensive subsectors.
• Hausmann argues that a country’s existing export basket influences its subsequent di-
versification (Hausmann, 2014). With a larger share of manufacturing in domestic GDP,
a country is therefore more able to enter high-technology manufacturing sectors.
9
• Furthermore, structural change is found to be more likely to occur towards sectors that are similar in nature (Radebach et. al, 2016). This implies that structural change to- wards technologically sophisticated sectors may simultaneously fuel structural change to other technologically sophisticated sectors, as technologically sophisticated subsec- tors are assumed to be similar in nature.
These findings suggest the cumulative nature of establishing manufacturing in an economy and the cumulative nature of productivity growth in manufacturing, especially for technolog- ically sophisticated subsectors. Further reasons for this cumulative nature may be related to some key features of manufacturing. Manufacturing is highly interlinked in the domestic econ- omy and provides potential for knowledge spillovers (Page, 2018; Rodrik, 2015). To account for this, the dependent variable Manufacturing as a share of domestic GDP is expected to partly capture dynamics entailed in the cumulative process of establishing manufacturing sec- tors in the domestic economy. Furthermore, this also suggests that structural change as a source for productivity growth in (high technology) manufacturing subsectors may depend even more on the access to foreign knowledge and information spillovers. Those are pro- moted by increasing interactions with the global economy that yield in information spillover and learning effects.
As some literature suggests that increasing productivity has not decreased employment per se, but the share of labor in value added in developed OECD economies (Autor, 2017), I will include a model not including manufacturing value added as a precaution.
The External Environment and Structural Change
One suspected reason for Premature Deindustrialization in developing economies was the predominance of South-East Asian economies in manufacturing (Rodrik, 2015; Frankema, 2018). Based on this assumed factor, it makes sense to consider a country’s interactions with the global economy, measured by its Trade Openness Growth. The idea behind the sum of imports and exports, relative to GDP, having an effect on structural change is that it indicates access to knowledge and information, and the ability to produce at a larger scale. All of these factors are expected to increase productivity and incentivize structural change. Taking a coun- try’s interactions with the global economy into account is supported by additional empirics:
• Trade Openness Growth tends to increase a country’s level of productivity. As a result, at lower levels of development, increasing productivity levels in manufacturing should yield in structural change towards these productivity gaining sectors (Duarte and Restuccia, 2009; Dabla Norris et. al, 2013; Fiorini et. al, 2013)
ii.
A country’s Exchange Rate value will be considered as another key variable, as a depreciating
domestic currency significantly increases domestic producers’ ability to sell their goods on the
10
international market, thus increasing demand for domestically produced products. Further- more, as currencies depreciate, importing substitute goods gets more expensive, reducing the overall level of competition on the domestic market. I therefore expect depreciating curren- cies to be fueling structural change, suspecting that a reduced number of imports from abroad protects domestic manufacturing from foreign competition while also enabling to sell prod- ucts cheaper on international markets.
Theoretically, a depreciating currency could also mean that intermediate input goods sourced from abroad become more expensive, making structural change towards manufacturing more difficult altogether. Restricted access to intermediate inputs would make production in man- ufacturing more expensive, thereby disincentivizing structural change. However, this logic only holds if intermediate inputs are indeed sourced from abroad, not locally. Additionally, empirics support the assumed positive coefficient from a depreciating currency on structural change:
• Several scholars state that depreciating exchange rates may promote the development of tradeable sectors both in general and in particular for modern manufacturing. This implies that a depreciating exchange rate is likely to fuel structural change while an overvalued exchange rate is found to impede structural change (McMillan et. al, 2011;
McMillan et. al, 2014; Martins, 2018).
• Dabla-Norris et. al also find that the effect from a depreciating exchange rate is partic- ularly strong at low value-added manufacturing sectors (Dabla-Norris et. al, 2013)
In order to capture another dimension of a country’s interconnectedness to the global econ- omy as well as its access to knowledge, I include FDI inflows as an additional variable. I expect FDI inflows into a country to have a positive effect on structural change towards manufactur- ing subsectors. Similar to the dynamics, explained for Trade Openness Growth, FDI inflows should lead to knowledge and information spillovers, promoting structural change towards manufacturing subsectors. There is additional ample empiric evidence for considering FDI in- flows and expecting a positive effect on structural change towards manufacturing:
• Structural change towards both sophisticated and less sophisticated manufacturing subsectors was accompanied by large amounts of FDI floating into countries. The BRICS managed to establish strong manufacturing sectors, specializing on different subsec- tors while all being integrated into the global trade system and reporting very high export figures (Naudé, 2015; Martins, 2018)
Technically, it is also important to consider FDI as an explanatory variable, as it is expected to
correlate with several other variables such as the level of fundamentals (Walsh, 2010). Not
including FDI would endanger biasing the obtained remaining coefficients by capturing effects,
actually traceable to FDI but affecting included variables.
11
Table 2 Independent Variables and Expected Coefficients
Variable Expected Coefficient Source
GDP per capita + World Bank Statistics
Exchange Rate + Bruegel Institute
Trade Openness Growth + World Bank Statistics
FDI + World Bank Statistics
Labor Flexibility + Fraser Index Statistics
Fundamentals + World Bank Statistics
Education + World Bank Statistics
c. Hypothesis
To form a hypothesis on which variables are expected to have an especially large effect on structural change, one needs to take into account that most independent variables are ex- pected to be highly correlated with one another. While this will be tested for in the correlation analysis at a later stage, the hypothesis does not include an excessive number of independent variables for explaining structural change, as this may lead to multicorrelation. Furthermore, the limited sample size limits the number of independent variables for the econometric anal- ysis. Therefore, the hypothesis based on the literature review identifies the key variables for structural change in the following way:
• H
1: productivity enhancing structural change between manufacturing subsectors is contingent on GDP per capita, the domestic educational level, Trade Openness Growth and FDI inflows.
By focusing on these four independent variables in the hypothesis, proxies for both internal factors within a country and proxies for a country’s external environment are being taken into account. The hypothesis explicitly refers to structural change between manufacturing subsec- tors, as the dependent variable will be defined as the relative contribution from structural change to productivity growth in manufacturing subsectors, as explained in section 3.
Additional variables will replenish the analysis at a later stage. However, the four variables included in the hypothesis reoccur consistently in prior research and therefore I expect to obtain the most appropriate model with these variables.
When assessing structural change towards technology intensive manufacturing subsectors, I
expect high technology subsectors to experience productivity enhancing structural change as
a country interacts with the global economy. Interactions trigger knowledge and technology
spillover effects. Therefore, I suspect variables tied to the external environment to be of par-
ticular importance for structural change towards technologically sophisticated manufacturing
subsectors. The suspected self-reinforcing nature of structural change towards technologically
sophisticated manufacturing subsectors should further strengthen the dynamic of structural
change as a contributor to productivity growth. Furthermore, the literature showed increasing
12
shares of manufacturing in domestic value added and increasing interactions with the global economy to trigger employment increases as well as productivity increases in technologically intensive sectors. Therefore, I expect a joint effect from productivity in manufacturing, paired with increasing interactions with the global environment, to have a positive effect on employ- ment and productivity in technologically intensive manufacturing subsectors, hence a positive effect on structural change.
• H
2: structural change towards technologically intensive subsectors increases due to variables capturing interactions with the global environment and the joint effect of increasing manufacturing productivity levels and interactions with the global environ- ment.
Note that for hypothesis H
2, I no longer consider the relative contribution from structural change but analyze structural change towards technologically sophisticated subsectors. By ex- ploring the hypothesis H
1and H
2, the thesis ultimately addresses the research question whether and to what extent developing economies
iiihave been able to establish manufactur- ing subsectors as a way to boost productivity growth between the years 2000 and 2012. An additional aspect is whether establishing technologically sophisticated manufacturing subsec- tors requires different dynamics. The research question addresses a gap in the literature, as most studies have analyzed structural change within economies, but did not focus uniquely on the manufacturing sector. Furthermore, a limitation on technology sophisticated subsec- tors is an interesting complement to the recent discussion.
3. Data and Methodology
Independent variables have been identified in the literature review section and do not need to be calculated but merely retrieved from adequate sources. Appendix A2 gives an overview over the precise definition and source per variable. The following section will introduce a de- composition method for productivity growth, in order to isolate structural change in manu- facturing subsectors from overall productivity increases.
a. Decomposition Method
For calculating structural change towards manufacturing subsectors, the thesis uses the
UNIDO INDSTAT 4 dataset, Revision 3. This data set provides data for a total of 34 developing
economies worldwide for country-specific time frames between the years 2000 to 2012. The
selection criteria for developing economies was based on Official Development Aid (ODA) eli-
gibility in the year 2001. The following different, not mutually exclusive classification schemes
are reported in table 3.
13
Table 3: Overview Regions and Categories
Category or Region Countries
Former USSR (FUSSR) Azerbaijan, Georgia, Kyrgyzstan, Latvia, Lithuania, Russia
South East Asia (SEA) China, India, Indonesia, Republic Korea, Sri Lanka, Vietnam, Malaysia
Middle East Northern Africa (MENA) Iran, Iraq, Jordan, Kuwait, Morocco, Palestine Latin America (LA) Brazil, Colombia, Ecuador, Uruguay
Brazil, Russia, India, China (BRICS) China, India, Brazil, Russia
Middle Income Trap Countries (MIT) Brazil, Colombia, Ecuador, Iran, Jordan, Malaysia, Romania, Turkey
Low Income Economies (LDC) Azerbaijan, Brazil, Kyrgyzstan, Malawi, Vietnam The LDC classification scheme identifies countries in the sample at lowest levels in terms of GDP per capita. The MIT classification scheme is based on Yilmaz (2016) and identifies middle income countries, that have been unable to advance towards a high-income status and have been stagnating in their economic development.
When quantifying structural change per country, the thesis follows the approach explained by de Vries et. al in 2014, decomposing productivity growth into a total of three components.
While classic structural change decompositions, as practiced by McMillan (2011), split produc- tivity into two components, a within component and a between component, differentiating between three components bears the advantage of distinguishing between the effects from a high productivity level in a sector (static effect, second term in equation (1)), and a high productivity growth rate in a sector (dynamic effect, third term in equation (1)). The dynamic effect takes labor reallocation ∆𝜃 due to changed productivity ∆𝑝 into account. My decompo- sition of productivity growth per sector yields the following equation.
𝑃
𝑡= ∑ 𝜃
𝑖,𝑡∆𝑝
𝑖,𝑡+ ∑ ∆𝜃
𝑖,𝑡𝑝
𝑖,𝑡+ ∑ ∆𝑝
𝑖,𝑡∆𝜃
𝑖,𝑡(1)
Where P indicates productivity growth in sector i at time t, and 𝜃 represents the employment share sector i at time t. Respectively, the variables are calculated in the following way
∆𝑝
𝑖,𝑡= 𝑝
𝑖,𝑡− 𝑝
𝑖,𝑡−1(2)
𝜃
𝑖,𝑡=
𝐸𝑖,𝑡𝐸𝑡
and ∆𝜃
𝑖,𝑡= 𝜃
𝑖,𝑡− 𝜃
𝑖,𝑡−1(3)
In order to obtain the percentage increase, relative to t
0of each of the decomposed elements from equation (1), they are individually divided by 𝑌
𝑡, yielding in
𝑃
𝑡= ∑
∆𝑝𝑖,𝑡𝜃𝑖,𝑡𝑦 𝑡
+ ∑
∆𝜃𝑖,𝑡𝑝𝑖,𝑡𝑦 𝑡
+ ∑
∆𝜃𝑖∆𝑝𝑖𝑦 𝑡
(4)
14
Where Y is the economy specific manufacturing total productivity, measured in value added.
Value added is denominated in current US Dollars, using the average period exchange rate as given in the IMF Financial Statistics for the years 2000 – 2012 for a currency conversion be- tween local currency and US Dollars. One caveat of this analysis is that employment is meas- ured by people employed, not hours worked, which would be the more accurate data for cal- culating productivity but is difficult to obtain.
The first term quantifies productivity increases in sector i, attributable to changed productivity within the sector (within effect), while the second and third term are the aforementioned static and dynamic effects. Both the static and the dynamic between effect are important in my analysis and need to be considered in combination to obtain the total between effect.
Appendix A1 gives additional information on the data set INDSTAT 4 as well as alternative decomposition methodologies tested.
As a dependent variable for the subsequent econometric analysis, the thesis takes the total between effect, thus the sum from the dynamic effect and static effect, relative to total productivity growth per country. Therefore, the ultimate equation applied to calculate the dependent variable, named relative structural change STR2 is calculated as follows
𝑆𝑇𝑅2 =
∑ ∆𝜃𝑖,𝑡𝑝𝑖,𝑡∑ 𝑃+∑ ∆𝑝𝑖∆𝜃𝑖𝑡
(5)
For testing hypothesis H
2, I limit the dependent variable to structural change towards sectors, classified as high technology intensive and medium-high technology intensive. For an identifi- cation of these technology intensive sectors, I rely on information provided by UNIDO, that categorizes ISIC sectors 353, 2423, 30, 32 and 33 as high technology sectors, as well as sectors 32, 34, 24 (excluding 2423), 352, 359 and 29 as medium-high technology sectors. This leads to the second dependent variable HTMT1, that will be analyzed in a second econometric model at a later stage.
Returning to the research question, as the contributions from structural change to productiv- ity growth increase, I take this as an indication for the ability to establish manufacturing sub- sectors in an economy. Rodrik (2015) identified Premature Deindustrialization with the two indicators manufacturing employment and productivity in the industrial sector. As observable in equation (4), both of these variables ∆𝜃 and ∆𝑃 are included in the between effect. Regard- less of what actually triggers an increasing between effect, the consequence will be referred to as industrialization.
b. Reflection on Data Quality and Alternatives
One key advantage from the decomposition technique applied here is the introduction of the
dynamic term. As Rodrik identified both decreasing employment shares in industries as well
15
as decreasing productivity of industries, it makes sense to rely on a productivity decomposi- tion technique, in which both of these elements are considered in the between effect. This is done in the decomposition method chosen, but not in alternatives as shown in Appendix A1.
Authors such as Timmer and van Ark (2000) extend the methodology from equation (4) further by taking the effects from diminishing marginal productivity into account. As labor, and in particular surplus labor, reallocates from a labor abundant agricultural sector to other sectors, credit for increased productivity should be given to the labor absorbing sectors, not the sector getting rid of surplus labor. Thus, Timmer and van Ark extend the methodology to take the marginal effects from increasing labor inflows into account. However, as I am exclusively fo- cusing on a single sector in my dataset, I restrict the analysis of structural change to the de- composition method shown in equation (4).
I decided to take the contribution from structural change relative to total productivity in- creases as explained in equation (5) as opposed to the simple absolute structural change to- wards manufacturing subsectors. One main advantage of this approach is that although coun- tries may report similar absolute structural change values, with an overall lower productivity growth, structural change should stand out as a more important driving force for productivity growth in relative terms. This is not being accounted for when only considering the absolute contribution of structural change to productivity growth as seen in equation (4). This differ- ence between the relative structural change and the absolute structural change will become evident when looking at the Regional Analysis in part 3b, figures 2 and A6, and in the econo- metric analysis.
Although the data quantity is limited to 34 economies, expanding the data set by including additional countries would then include middle- and high-income countries. By applying the ODA eligibility scheme, I include a maximum of countries that can still be classified as devel- oping or less developed countries. Sticking to this classification scheme is necessary to address the research question.
The data source is the United Nations Industrial Development Organization (UNIDO) database, which can be considered reliable. Yet, especially low-income countries provide annual data that may fluctuate. To avoid these fluctuations from having an impact on the average annual structural change per country, I calculate the country specific annual average growth rate by taking t
0and t
k, where k is the country specific latest point in time, when data is available.
Dividing the growth rate between t
0and t
kthrough k yields in an annual average growth rate, that is less affected by potential annual fluctuations.
As an alternative to the cross-sectional study conducted with period averages, I could have
treated each annual productivity growth observation per country as a separate data point,
thereby conducting a longitudinal analysis with k observations per country. When doing so, I
obtain a highly unbalanced panel data set, as data availability per country in the time frame
16
2000 – 2012 varies. While a more advanced econometric analysis with a panel data set would be possible theoretically, in practice the high degree of imbalance prevents a subsequent econometric analysis. Therefore, I decided to treat the data as a cross-sectional data set.
c. Descriptive Statistics
Table 4 Overview Productivity Decomposition, Annual Means
Region Total Within Static Dynamic Between Relative
All 0.84% 0.64% 0.1% 0.11% 0.21% 24.3%
SEA 0.35% 0.28% 0.08% - 0.01% 0.07% 20.51%
LA 0.25% 0.18% 0.02% 0.05% 0.07% 26.61%
MENA 1.06% 0.78% 0.14% 0.14% 0.28% 26.52%
FUSSR 1.44% 1.12% 0.15% 0.16% 0.32% 21.98%
BRIC 0.48% 0.45% 0.01% 0.02% 0.03% 6.29%
LDC 1.12% 1.02% 0.03% 0.07% 0.1% 8.68%
MIT 0.41% 0.36% 0.01% 0.05% 0.05% 13.1%
Total = total labor productivity growth as an average, Within, Static and Dynamic refer to the decomposable effects as ex- plained in equation (4), Between refers to the sum of the dynamic and static effect, Relative refers to the between effect as explained in equation (5)
These decomposition results are obtained when looking at the annual average productivity growth rate per country and applying a regionally unweighted classification scheme as ex- plained in table 3. A more detailed view on country specific data, rather than region specific data can be found in Appendix A4.
Table 5 Descriptive Statistics, Entire Sample
Total Within Static Dynamic Between Relative
Median 0.6% 0.43% 0.06% 0.06% 0.13% 22.84%
Variance 0.01% 0.01% 0.00% 0.00% 0.00% 14.10%
Std. Dev. 0.96% 0.70% 0.19% 0.16% 0.34% 37.00%
Based on my decomposition of structural change in 4-digit manufacturing subsectors, I find that the mean manufacturing compound productivity growth for all 34 countries in my sam- ple, based on country-specific time frames between 2001 – 2012 is 0.84%. As opposed to the average value, the median of the entire sample is at 0.6%, indicating that the distribution for both total productivity growth and relative contribution of structural change has a lower me- dian than mean. The distribution is therefore right-skewed. As indicated by the large standard deviation compared to the mean, the data is very heterogenous and unevenly spread with country specific total productivity increases being diverse. Furthermore, the contribution of relative structural change to productivity growth is very diverse across the country sample.
This is indicated by a very large standard deviation for the relative contribution of structural
change to productivity growth as compared to the mean relative contribution of 24.3%. These
findings underline the heterogeneity in the data set and different country experiences with
17
structural change. This heterogeneity is also expressed by figure 1, visualizing the distribution of relative structural change across my country sample. Based on the literature review which shows regionally different experiences with structural change, the large standard deviation for structural change makes sense.
Figure 1 Histogram Relative Structural Change
Previous studies report that average annual structural change contributed marginally in most regions. Lacking contributions from structural change is particularly evident in Brazil and the LA region, which is observable both in de Vries et. al and in my data set. One major difference between my data and previous research is structural change in China. My findings report lack- ing contributions from structural change to productivity growth in China. This is not in line with previous research, which indicates China experiencing average annual contributions from structural change to productivity growth by 1.2% (de Vries et. al, 2012). This larger effect from China is also assumed to drive overall developments in the SEA region. Note however, that de Vries’ reports refer to economy wide structural change whereas my analysis is limited to man- ufacturing subsectors. Subsequently, my findings for SEA as a regional unweighted average also differ from previous studies. An additional reason for differences between my data and previous research is data on China only being available between the years 2003 – 2007 in the UNIDO data set.
Regional Analysis
When analyzing the unweighted region-specific averages from table 4, some key observations need to be addressed further:
• My data confirms Latin American countries to have experienced the least amounts of structural change. However, when looking at the relative contributions from structural change to manufacturing productivity growth, regional experiences in Latin American countries did not differ substantially from other regional experiences.
• MIT countries and LDC countries experienced very modest contributions from struc-
tural change to productivity growth. This becomes clear when looking at the relative
18
contributions from structural change. This finding is interesting as it indicates that countries at lower stages of development (LDC) have greater difficulties to shift labor towards manufacturing subsectors, supporting the assumption of Premature Deindus- trialization. It also supports the idea of the industrial sector as an engine for economic growth, as both groups are stagnating in their economic development and report stag- nating contributions from structural change to productivity growth.
• The large productivity growth in FUSSR countries is driven by the within effect. Appen- dix A3 gives the decomposition results on a country level and shows that Azerbaijan and Georgia have been largely driving these high growth rates. The sector refined pe- troleum partly drives these results, contributing with a within effect of 2.27% in Azer- baijan. More information on a country-specific level is given in appendix A3.
• All in all, the regional analysis does support the claim of developing economies’ diffi- culties to industrialize.
Sectoral Analysis
For a more precise understanding of the difficulties related to industrialization for developing economies, this section analyzes sectoral patterns of productivity growth, based on technol- ogy intensity. To do so, I calculate average annual productivity growth per sector per country.
This country and sector specific annual data is then averaged over the entire sample for each sector, yielding in the average annual sector specific productivity growth rate. Based on tech- nology intensity, these sectors can then be added up into three different categories: high tech- nology, medium technology and low technology intensity (table 6). Note that the sum of all columns in table 6 do not add up to structural change as indicated in table 4, as table 6 measures the contributions for each sector separately for low-, medium-, and high technology intensive sectors. Table Appendix A4 shows sectoral average annual growth over all 34 coun- tries. Some key takeaways from this analysis include:
• The sum of annual mean structural change across all countries towards high technol- ogy intensive sectors was low. Sectors, classified as high technology sectors report productivity growth of 0.04%, with the between component only accounting for 0.01%. Structural change has been higher for medium technology intensive or low technology intensive sectors.
• Productivity growth for LA countries on average shows to be low in general and close to 0 in high technology sectors. The modest increase in productivity growth from me- dium technology sectors may also be due to missing data in LA countries for the me- dium technology sectors 2330, 2732, 2923 and 3313.
• The large productivity growth reported in MENA countries on average is primarily
driven by the within effect in medium technology and high technology sectors. When
identifying the sectors driving this large within effect, it shows that those are primarily
19
sectors that process natural resources, such as refined petroleum products (sector 2320), bakery products (sector 1541) and wearing apparel (sector 1810), all of which requiring inputs into domestic production, based on resource extraction.
• Sector 2320 also drives the large within effect in FUSSR countries, contributing by 0.59% in this country group. A further analysis of the FUSSR group shows that the within effect in medium technology sectors has been driving productivity growth in this country group. Structural change in general played a very limited role and high technology subsectors played a marginal role for explaining productivity growth too.
• Similar to previous observations, the average BRIC country also performs better in me- dium technology sectors, with close to no growth in high technology sector productiv- ity. Additionally, productivity growth in manufacturing in BRICS was driven by within effects, rather than the reallocation of labor.
• A comparison between LDC countries and MIT countries shows MIT countries to expe- rience stagnating productivity growth while this does not seem to be the case for LDC countries. While LDC countries primarily benefit from within contributions, but still ex- perience some benefits from labor reallocation in low technology and medium tech- nology sectors, MIT countries exclusively rely on marginal within effects to trigger mar- ginal productivity growth. The within effects do not show to have been able to com- pensate for the negative between effect on average in MIT countries.
• The data shows LDC countries being completely unable to enter high technology man- ufacturing sectors altogether.
• Within productivity increases are large in sectors processing natural resources. This indicates that natural resource processing sectors and agro-industrial sectors may en- tail potentials for productivity enhancing structural change. While this idea is also elab- orated in the literature on Premature Deindustrialization, it will not be analyzed fur- ther in this thesis as it is not part of the research question. Yet, the tables indicate that this provides room for further research in Premature Deindustrialization.
Table 6 Descriptive Statistics, Sectoral Analysis
Region Taxonomy Total Within Static Dy-
namic
Be- tween All
Low Tech 0.44% 0.35% 0.04% 0.05% 0.09%
Medium Tech 0.54% 0.43% 0.08% 0.07% 0.15%
High-Tech 0.04% 0.03% 0.01% 0.01% 0.01%
SEA
Low Tech 0.05% 0.05% 0.03% -0.03% 0.01%
Medium Tech 0.20% 0.19% 0.01% 0.01% 0.01%
High Tech 0.08% 0.06% 0.00% 0.01% 0.01%
LA
Low Tech 0.22% 0.13% 0.03% 0.06% 0.09%
Medium Tech 0.12% 0.10% 0.00% 0.02% 0.02%
High Tech 0.01% 0.01% 0.00% 0.00% 0.00%
Low Tech 0.62% 0.38% 0.12% 0.12% 0.24%
20
MENA Medium Tech 0.51% 0.45% 0.04% 0.03% 0.06%
High Tech 0.03% 0.02% 0.00% 0.00% 0.01%
FUSSR
Low Tech 0.53% 0.48% 0.01% 0.03% 0.05%
Medium Tech 1.13% 0.93% 0.18% 0.16% 0.33%
High Tech 0.03% 0.01% 0.01% 0.01% 0.01%
BRIC
Low Tech 0.14% 0.17% -0.02% 0.00% -0.03%
Medium Tech 0.35% 0.34% 0.00% 0.01% 0.01%
High Tech 0.08% 0.05% 0.02% 0.01% 0.03%
MIT
Low Tech 0.27% 0.22% 0.01% 0.04% 0.05%
Medium Tech 0.18% 0.16% 0.02% 0.02% 0.03%
High Tech -0.01% 0.02% -0.03% 0.00% -0.03%
LDC
Low Tech 0.68% 0.57% 0.05% 0.07% 0.11%
Medium Tech 1.23% 1.02% 0.11% 0.11% 0.22%
High Tech 0.00% 0.00% 0.00% 0.00% 0.00%
All Food Processing 0.23% 0.19% 0.01% 0.03% 0.04%
All Resource Processing 0.49% 0.44% 0.04% 0,04% 0.08%
I identified the ISIC sectors 15-16 as typical food processing sectors and ISIC sectors 17-21, 2320, 26-27 as typical resource processing sec- tors
Figure 2 Country Specific Absolute HTMT Structural Change
The mean between effect over all countries in my sample for structural change towards high- technology and medium technology intensive sectors is 0.16%. This small average between effect is visualized in figure 2, indicating that developing economies have not been very suc- cessful in establishing technology intensive manufacturing subsectors in the past.
Additional information and an additional visualization on structural change towards techno- logically sophisticated sectors can be found in Appendix A6. Figure A6 and figure 2 also visu- alize the difference between relative structural change and absolute structural change, ap- plied to technology intensive sectors HTMT as a way to clarify the difference further.
Decomposition Results and Independent Variables
One of the core lines of argumentation in Rodrik’s claim of Premature Deindustrialization is that countries do industrialize until reaching a certain point, beyond which countries begin to deindustrialize. This finding has been replicated in different studies (Herrendorf et. al 2013).
These studies trace a reverse U-shape relationship between absolute structural change in
21
countries and increasing levels of development, indicated by increasing levels of GDP per cap- ita. Economic theory reasons this observation by linking productivity growth to demand ef- fects as explained by Engel’s Law. Based on the assumption of an elasticity of substitution for manufactured goods smaller than 1, this implies that as income increases, consumers spend a smaller share of their budget on manufactured products and demand shifts to services, ulti- mately triggering deindustrialization beyond a certain point in GDP per Capita (Engel, 1857).
Looking at a scatterplot for my data, I find the relationship between GDP per capita and a country’s structural change to support previous findings of an inverted U-shape. Furthermore, the majority of countries being centered around a value of zero for structural change is in line with previous findings of a limited role of structural change for explaining productivity growth.
Figure 3 Scatterplot lnGDP p. Capita, Structural Change, Quadratic Fit
When looking at GDP per capita levels in my sample, I find that the mean lies at 8,091.12 USD while the median lies at 5,349.17 USD. This indicates that the distribution pattern of countries and their GDP per capita levels is skewed towards less developed economies in terms of GDP per capita. They are more strongly represented in my sample, as compared to more developed high GDP per capita economies. Therefore, I expect GDP p. capita to be having a positive effect on a country’s structural change towards manufacturing subsectors for my country sample. I suspect my countries to be typically at a stage, where increasing GDP per capita still increases structural change towards manufacturing, as countries have not yet reached the critical break- ing point, beyond which a further increase in GDP per capita is associated with deindustriali- zation. In line with this scatterplot, I expect the squared value of GDP per capita to have a negative coefficient due to marginally decreasing effects from GDP per capita on structural change.
In a closer analysis on summary statistics for my data set, it shows that countries perform
relatively similarly in terms of absolute structural change, GDP per capita, Trade Openness
Growth, exchange rates and labor market flexibility. The independent variable FDI stands out
22
as there is more heterogeneity across the countries in my sample. This is indicated by a com- paratively large mean standard deviation of 6.89 as compared to the mean value of 7. A closer analysis on the large standard deviation will be done in section 4a. Appendix table A8 gives further summary statistics for my independent variables.
4. Econometric Analysis
The following model tests the thesis’ first hypothesis
• H
1: productivity enhancing structural change between manufacturing subsectors is contingent on GDP per capita, the domestic educational level, Trade Openness Growth and FDI inflows.
Figure 5: Pearson Correlation Coefficient
Rel. Str.
Change
High- Tech.
GDP p.
Capita Ex- change Rate
Trade Open- ness Growth
FDI
Labor Flexibil- ity
Funda- mentals
Rel. Str. Change 1
High-Tech. 0.403 1
GDP p. Capita -0.166 -0.112 1
Exchange Rate -0.067 0.059 -0.544 1 Trade Openness
Growth 0.346 0.465 0.067 -0.025 1
FDI 0.276 -0.237 -0.422 0.135 -0.005 1
Labor Flexibility -0.10 0.027 -0.326 -0.357 0.289 0.392 1
Fundamentals 0.333 0.117 0.730 -0.503 0.318 -0.3 -0.297 1
Primary Enroll-
ment -0.19 -0.341 -0.145 -0.474 -0.257 0.263 0.532 -0.152
Schooling
(Mean) 0.295 0.091 0.316 -0.23 0.154 -0.485 -0.346 0.646
As a first step, a correlation analysis is used to get an idea for possible strong significant coef-
ficients and potential multicorrelation problems in a regression. The last two columns from
the correlation matrix have been omitted to maintain a good overview in the matrix as these
correlation coefficients did not entail additional important information for the correlation pat-
tern. For completeness, the full correlation matrix can be retrieved in appendix A9. Note that
the Spearman rank correlation analysis yields in similar results, thus I restrict my analysis to
the Pearson Correlation Coefficients. One major drawback when only relying on a correlation
analysis to determine explanatory variables for a model is the assumed linear relationship be-
tween variables in the correlation analysis, that may not necessarily apply to all variables
equally. One example for this is the unexpected negative coefficient for GDP p. Capita in the
23
