Globalization and Poverty in Bolivia

Globalization and Poverty in Bolivia

A case study

Nicole van der Pauw

International Economics & Business Masterthesis

Acknowledgements

This study contains an in-depth analysis of the relation between globalization and poverty in Bolivia. The situation and changes of the Bolivian economy are compared to other economies on the Latin-American continent for the period 1985-1999. This research is my Master Thesis for the International Economics & Business curriculum at the faculty of Economics of the University of Groningen, the Netherlands.

I acknowledge that without the support and feedback of my advisor dr. ir. Dirk Bezemer it would have been impossible for me to achieve this result. Therefore, I am particularly thankful to him. Furthermore I would like to thank my second supervisor Gaaitzen de Vries, Miss C. Castaldi and all other staff of the faculty of Economics, family and friends who supported and helped me during this rather intense learning experience.

Globalization and Poverty in Bolivia

... 1

... 2

... 5

... 5

... 6

... 6

PART 2 : BOLIVIA & DEFINITIONS OF TERMS

... 8

§2.1 Bolivia: current situation

... 8

§2.2 Definitions

... 8

2.2.1 Globalization ...8

2.2.2 Poverty ...9

PART 3: LITERATURE REVIEW

... 10

§ 3.1 Background

... 10

§ 3.2 Model about globalization leading to poverty reduction

... 11

3.2.1 International trade ...13

3.2.2 International Capital Flows...14

3.2.3 International labor flows ...14

3.2.4 Technological Change ...14

3.2.5 Complimentary policies ...15

§ 3.3 Insights about Latin America

... 15

§ 3.4 Insights in Bolivian policies

... 16

PART 4: METHODOLOGY & MODEL DESCRIPTION

... 17

§4.1 Hypotheses

... 17

§4.2 Model specifications

... 18

4.2.1 Relationship between globalization and economic growth ...18

4.2.2. Relationship between globalization and poverty ...19

4.2.3 Relationship between structural reforms and poverty ...19

§4.3 Data collection

... 20

§ 4.4 Dependent variables

... 21

§ 4.5 Globalization

... 22

4.5.1 Trade integration ...22

4.5.2 Financial integration ...22

4.5.3 International labor flows ...23

4.5.4 Technological change ...24

§ 4.6 Structural reform indices

... 24

§ 4.7 Control variables

... 25

... 28

§5.1 Analysis of relationship between globalization and economic growth

... 28

5.1.1 Unit Root Test ...28

5.1.2 Multicollinearity ...28

5.1.3 Panel data ...29

5.1.4 Autocorrelation ...30

5.1.5 Model estimations ...30

5.1.6 Regression interpretation...32

§ 5.2 Analysis of relationship between globalization and poverty

... 34

5.2.1 Unit Root test ...34

5.2.2 Multicollinearity ...34

§ 5.3 Analysis of relationship between structural reforms and poverty

... 39

5.3.1 Unit Root test ...39

5.3.2 Multicollinearity ...39

5.3.3 Autocorrelation ...40

5.3.4 Model estimation...40

5.3.5 Regression interpretation...42

PART 6: CONCLUSIONS & DISCUSSION

... 44

PART 1: INTRODUCTION

§1.1 Introduction

"Globalization is not a phenomenon. It is not just some passing trend. Today it is an overarching international system shaping the domestic politics and foreign relations of virtually every country, and we need to understand it as such."

Thomas Friedman

The quote above reflects the complexity of globalization and the impact of it on the entire world. These days, we are striving for a world in which global consensus is achieved. In the past decades, international organizations have been established to diminish conflicts and to reach agreements in the fields of politics, trade and the environment. Countries have become more and more integrated with each other, and companies do not longer operate locally, but in an international environment with customers spread all over the world.

This process of continuing interdependence of countries can be characterized as globalization. And although globalization is a central issue in international economics, the term has been in common usage only since the second half of the 1980s (Santarelli and Figini, 2002). Despite, the last two decades are known as the 'globalization years', globalization is not a new phenomenon; the first globalization wave took already place in the period prior to World War I. And though the world's markets for goods and services are more integrated now than ever before, the pattern is different; with a rise in intra-industry trade, compared with the predominance of inter-industry trade in the earlier period of globalization (Fischer, 2003).

International trade always played an important role in the policies of many developed and developing countries. In the early post-World-war II period, the theory of import-substituting-industrialization (ISI) was a in developing countries a popular tool to achieve economic prosperity, and its implementation for some time seemed to produce positive results. However, as time went by, it was observed that countries that had pursued export promotion strategies were more successful than those that had focused on keeping imports out and that the returns to ISI seemed to be diminishing (Fischer, 2003).

Proponents of this view accordingly advocate a shift of focus from relative to absolute measures of well-being (Goldberg and Pavcnik, 2005).

Despite the importance of the above argument, there has been little work on the effects of trade policy on absolute measures of well-being, such as poverty. The scarcity of studies on this topic is primarily due to the difficulties associated with the measurement of poverty on one hand, and the identification of the trade policy effects on the other (Goldberg and Pavcnik, 2005).

Nowadays, there is a large discussion about what the impact of globalization is on the poor population in the world. Main questions are “does globalization lead to increased economic growth” and “does this implies a positive change for the poor”?

§1.2 Purpose of paper

A case study about Bolivia is conducted to investigate the relationship between globalization and poverty. I chose Bolivia as the country to conduct a case study for, since this country in the Andean mountains was long one of the poorest and least developed Latin American countries, but made a lot of progress after heavy reforms in the 1980s. However, although there have been noticeable attainments in terms of poverty reduction, (as indicated by the improvement in various social indicators in recent years,) Bolivia's social indicators still remain weaker than the average for Latin America and are close to levels observed in Sub-Sahara Africa. Social conditions are especially acute in rural areas, where 90 percent of the population still lives in poverty (Jemio & del Carmen Choque, 2003).

To investigate the possible relationship between globalization and poverty, the Latin-American region is studied, since in this region the culture and institutions are similar, and therefore information about other countries can be used in order to build an economic model.

The main questions I will address in this paper are the following:

Is there a clear relation between globalization and level of GDP per capita in Latin America during the period 1985-1999? If this relationship holds, does a change in GDP per capita have an influence on the level of poverty?

Have the structural reforms during the past decades lead to a reduction in poverty? Which reforms are likely to be most beneficial for the poor?

§1.3 Outline

PART 2 : BOLIVIA & DEFINITIONS OF TERMS

§2.1 Bolivia: current situation

Historically, the country Bolivia, named after independence fighter Simon Bolivar, broke away from Spanish rule in 1825; much of its subsequent history has consisted of a series of nearly 200 coups and counter-coups. Comparatively democratic civilian rule was established in 1982, but leaders have faced difficult problems of deep-seated poverty, social unrest, and illegal drug production (CIA factbook, 10 January 2006).

In the early eighties of the last century Bolivia experienced a disastrous economic crisis; the enormous hyperflation. Changes were needed and the country started to drastically reform its economy. Initially, these strong economic reforms spurred real GDP growth in the following decade, and poverty rates fell. However, economic growth stagnated in 1999 again due to the global slowdown, in combination with domestic factors such as political turbulence, civil unrest and soaring fiscal deficits, all of which hurt (foreign) investor confidence. In 2003, after a period of continuing violent protests, there came an end to the pro-foreign investment economic policies of President Sanchez de Lozada. His resignation meant the cancellation of plans to export Bolivia's newly discovered natural gas reserves to large northern hemisphere markets. Because of uncertainty, foreign investors hesitated to invest, so foreign direct investment dried up. During the period 2003-2005 real GDP growth was positive, but still below the levels experienced during the 1990s. The growth was mainly due to increased exports of natural gas to neighboring country Brazil. Still today, Bolivia remains dependent on foreign aid from multilateral lenders and foreign governments (CIA factbook, 10 January 2006).

In December 2005, Evo Morales, leader of the Movement Towards Socialism, was elected by the Bolivian population as president after he promised to change the country's traditional political class and empower the nation's poor majority. (CIA factbook, 10 January 2006).

§2.2 Definitions

In this section the definitions for globalization and poverty are reflected. These definitions will be used throughout the entire paper.

2.2.1 Globalization

The definition of globalization used in this paper is more detailed, and derived from Round and Whalley (2002). They define globalization as the increase in integration between and within countries, manifested through an increase in the movement of commodities, labor, capital (financial and physical capital), and technology.

2.2.2 Poverty

Poverty characterizes a section of the population which cannot satisfy its basic needs for food, clothing, housing, education, health, security, and citizen participation as a result of the lack of opportunities to obtain sufficient income, reduced access to public services, high vulnerability, and social exclusion. As a result, the causes of poverty are both economic and social (IMF, prsp 2000).

The definition of poverty is the following: A person is considered poor if his or her consumption or income level falls below the poverty line or a minimum level necessary to meet basic needs. However, poverty is an outcome of more than economic processes. Economic, social, and political processes interact with and reinforce each other in ways that can worsen or alleviate the deprivation faced by poor men and women (World Bank, development data 2002).

Poverty is computed on the basis of the number of individuals whose expenditure (or income) is below a conventional threshold, the poverty line. This threshold is defined relative if it is determined annually with respect to the population's average level of income, absolute if it is determined with respect to the monetary value of a bundle of necessary goods and services, updated every year to take account of the variation in prices and bundle composition (Figini and Santarelli, 2004).

PART 3: LITERATURE REVIEW

§ 3.1 Background

Nowadays, globalization and poverty reduction are considered important issues in development economics. Therefore, a lot of research has already been conducted in this field, and extensive literature about these particular topics exists. In the following passages the focus will be on the most relevant issues that are important for my research.

In the 1990s the term globalization has become an expression to describe the increasing integration of national economies through cross border transactions. This integration usually refers to the international trade of goods and services, to private cross-border capital flows, to international labor migration flows and to the worldwide exchange of knowledge. The increasing globalization and integration has initiated an intense debate on its impact on economic growth and poverty reduction in developing countries. Many economists have argued that globalization may either increase or reduce within-country poverty. Globalization is a multifaceted process - characterized by a wave of privatization in public utilities and other previously state-owned industries, reforms of both domestic financial markets and taxation systems, and liberalization of labor markets - which has produced unprecedented acceleration in the flows of both international trade and foreign direct investment (FDI) (Sala-i-Martin, 2002a and 2002b).

While several recent surveys review the evidence on the relationship between globalization and poverty, the authors of these surveys acknowledge that they can only review the indirect evidence regarding the linkages between globalization and poverty. There have been almost no studies which test for the direct linkages between the two (Harrison, 2005). Goldman and Pavcnik (2004) point out that "while the literature on trade and inequality is voluminous, there is virtually no work to date on the relationship between trade liberalization and poverty". Studies so far have either focused on the relationship between globalization (measured in terms of trade openness) and economic growth, or on the relationship between growth and poverty reduction.

Chart 1: link between globalization and poverty reduction

However, the extensive literature using cross-country comparisons has left ambiguous implications for the impact of trade on poverty within countries. Whether the growth effects

are strong enough to entail that poverty falls with trade openness also remains unclear (Ravallion, 2004).

In her paper, Ann Harrison (2005) summarizes several findings of economists about the linkages between globalization and poverty. She mentions: what can be concluded from the various studies that have been conducted so far, is that the assumption that 'poor countries with a comparative advantage in producing unskilled labor intensive goods will gain from trade reform' is much too simple (Harrison, 2005). There are more factors, like the mobility of labor, that determine whether a poor country can gain from a trade reform.

A second conclusion of Harrison (2005) is the finding that 'the poor are more likely to share in the gains from globalization when there are complementary policies in place.' This finding has major implications; the fact that other policies are needed to ensure that the benefits of trade are shared across the population suggests that relying on trade reforms alone to reduce poverty is likely to be disappointing.

She also finds that that evidence suggests that export growth and incoming foreign investment have contributed towards reducing poverty.

Furthermore she mentions that research finds that financial crises are very costly to the poor. Financial deregulation is likely to lead to higher consumption and output volatility in low-income countries. The need for complementary policies, such as the creation of reliable institutions and macroeconomic stabilization policies is reinforced. While financial crises resulting from unrestricted capital flows are associated with a higher likelihood of poverty, foreign direct investment inflows are associated with a reduction in poverty.

The final conclusion that Harrison (2005) reaches, based on her summary of several findings of economists, is that 'globalization produces winners and losers among the poor.' Globalization is not beneficial for everyone in the society, but as long as the people that gain from globalization is larger than the people that will loose because of globalization, a reduction in poverty can be achieved.

Santarelli and Figini (2002), find in their research that if growth is distribution neutral, as it is claimed, and trade enhances growth, then it can be argued that trade is beneficial for poverty. But the evidence, both theoretical and empirical, is much more complex than this.

§ 3.2 Model about globalization leading to poverty reduction

growth. At the second level, rapid economic growth, together with employment and industrial structure, will lead to a reduction of the poverty level observed. In the chart below a schematic representation between globalization and poverty reduction is reflected. It can be seen that the relationship between globalization and poverty reduction is very complex, including many factors that need to be controlled in order to achieve a reduction in poverty.

Chart 2: schematic representation between globalization and poverty reduction

Source: chart derived from Husain (2001)

Globalization International Labor Flows International Trade Technological Change Changing Public Sector Expenditure Rural Urban Migration Effects through External Sector Macro Economic Stability Removal of Price Controls on Agriculture

Industrial Structure GDP growth Employment

Observed Outcome at Household and

Individual Level International

During this study, the four main components that Husain attributes to globalization, are used as the factors that together form globalization.

These factors are: international trade, international capital flows, international labor flows and technological change (particularly in information technology (IT) and telecommunications)

All these factors are discussed in more detail below.

3.2.1 International trade

Many economists have used the Heckscher-Ohlin framework in international trade to argue that trade reforms in developing countries should be inherently pro-poor, since poor countries have a comparative advantage in producing goods that use unskilled labor (Harrison, 2005). From this perspective, the conclusions are often that expanding trade opportunities for poor countries should cut poverty and reduce inequality within these countries, since the capital abundant rich nation will export capital-intensive goods, while the labor-abundant poor nation will export labor-intensive goods. This implies for developing countries that unskilled labor would benefit most from globalization, so the poor in particular should benefit from globalization (Topalova, 2005).

However, since the real world more complex than models, different studies have reached different conclusions. Ravallion (2004) argues that based on the data available from cross-country and over time comparisons, it is hard to maintain the view that trade openness is, in general, a powerful force for poverty reduction in developing countries, since the extensive literature so far has led to ambiguous conclusions.

So although theory predicts that international trade is directly beneficial for the poor, evidence suggests that the poor are more likely to share in the gains from globalization when there are complementary policies in place (like investments in human capital and infrastructure, as well as policies to promote credit and technical assistance to farmers, and policies to promote macro-economic stability) (Harrison, 2005). Another statement is put forward by Bhagwati and Srinivasan (2002), who point out that if a country wants to maintain an export-led development strategy, that is, if a country wants to rely on free trade, it must maintain a framework of macroeconomic stability. Because stability implies low inflation, it is another channel through which trade affects the poor positively, since the poor tend to be hardest hit by high inflation.

3.2.2 International Capital Flows

In theory, openness to financial capital flows could alleviate poverty through several channels. Prasad et al. (2004) point out that if greater financial integration contributes to higher growth by expanding access to capital, expanding access to new technology, stimulating domestic financial sector development, reducing the cost of capital and alleviating domestic credit constraints, then such growth should reduce poverty. However, since the authors do not find clear linkages between financial integration and growth in the aggregate cross-country evidence, they mention that a strong and causal relationship between financial integration and poverty are also likely to be difficult to find.

Prasad et al. (2004) conclude their study by suggesting that if financial globalization is approached with the right set of complementary policies, then it is likely to be growth-promoting. These policies include the use of flexible exchange rates, macroeconomic stabilization policies, and the development of strong institutions.

3.2.3 International labor flows

Husain (2001) mentions that in both the long and short-run, international migration generally helps the poor. If labor can move without restrictions, the demand and supply of labor can be more balanced, and migration can take place to areas that have shortages of labor. Since unemployment will be lower, free international labor flows can be considered beneficial for the poor.

3.2.4 Technological Change

3.2.5 Complimentary policies

In the model displayed above, Husain (2001) also mentions that globalization and economic growth does not automatically result in poverty reduction. He stresses that the strength and speed of transmission from globalization to poverty reduction, are determined by complimentary domestic policies, good governance and institutions delivering public service. The policies that facilitate unhindered flows of international trade, capital, and participation in labor flows are summarized as follows:

- reduced tariff and removal of non-tariff barriers

- removal of price distortions

- flexible regulations and legislation of labor

- healthy and sound financial sector and capital markets

- investment in skill development and technological assimilation

- macroeconomic stability

§ 3.3 Insights about Latin America

Because of the energy crisis in the 1970s, globalization and the international debt crisis, Latin American countries were during the 1980’s pushed into market liberalizing policy reforms that were supposed to bring about greater economic efficiency and productivity and allow the countries of this region to participate in, and benefit from, the open, market-based economies. (Teichman, 2001). At first, these reforms seemed to be quite successful, since, with a resurgence of economic growth in the first eight years of the 1990s per capita incomes rose and poverty rates fell. However, with stagnating growth rates in the later 1990s, poverty reduction also stopped (Gindling, 2005). Gindling (2005) also mentions that accompanying these changes in economic outcomes in the 1980s and 1990s in Latin America, there were significant changes in the economic environment. The most important was that the regulation regarding the level of protection afforded by governments to the jobs and earnings of workers changed. Broad structural adjustment policies were designed, to lead to a restructuring of production and employment. Import-substitution protections against foreign competition were reduced, as well as reduced subsidies for agriculture, reduced public sector employment, and generally increased reliance on private investment and market forces.

Gindling (2005) puts forward in his paper that policies to address the poverty problem in Latin America need to go beyond promoting economic growth and providing safety nets targeted to the currently poor. These policies should focus not only on aggregate economic growth, temporary income subsidies and the targeted provision of services to the currently poor, but also on improving the income-earning assets of those nearly poor and middle-income families who are likely to be vulnerable to macroeconomic shocks that could push them into poverty.

§ 3.4 Insights in Bolivian policies

In 1985 in Bolivia, a new government took office and implemented a wide ranged set of reforms aimed at stabilizing the economy and restoring economic growth after the hyperflation of the early 1980’s. The stabilization program was designed to let the country move from a static and closed economy to a market-based, open economy (Sachs, 2005), The main elements of the IMF style structural reforms were: trade liberalization, a massive devaluation and unification of the exchange rate, increases in public sector prices (in particular those of domestic petroleum products), and reductions in government expenditures to levels that could be financed by available funds (Jemio & del Carmen Choque, 2003). Andersen and Nina (2002) find that these reforms successfully reduced inflation and fiscal deficits. Furthermore, it is concluded that the country has achieved economic stability and moderate but stable economic growth during most of the 1990's. Jemio and del Carmen Choque (2003) mention that the reform process however has produced far less favorable results in terms of employment generation and poverty reduction, and Bolivia's social indicators remain weaker than the average for Latin America.

PART 4: METHODOLOGY & MODEL DESCRIPTION

In this section of the paper the data used for the empirical analysis is discussed. The reasons for choosing certain variables are clarified, as well as the way in which they will be used in the analysis.

As mentioned before, the key topic of this paper is to measure whether globalization (= increased openness) has a positive and significant impact on GDP per capita, and if this relationship exists, whether changes in the level of GDP per capita have led to a change in the poverty situation encountered in Bolivia.

A second major topic that is addressed is whether structural policy reforms have a direct influence on poverty. Statistical tests will show if, and which of the structural reforms in the field of trade policy, financial policy, tax policy, labor legislation and privatization have a positive and significant impact on poverty.

§4.1 Hypotheses

Most economists hold the opinion that reducing barriers to economic integration have a positive effect on economic growth and poverty reduction. According to Kappel (2003), this view is often referred to as the “openness hypothesis”, which interprets liberal regimes of international transactions as a major cause of higher rates of economic growth and poverty reduction. Proponents of the “institutional hypothesis” argue that well-defined and enforced institutions are the major determinants of growth and poverty reduction, which subsequently leads to liberalization of international transactions.

In a developing country, globalization is expected to cause substantial changes in the structure of the national economy, from reliance onto production of primary commodities to more labor intensive manufacturing, primarily meant for export markets. Globalization is also expected to result in easier and sounder access to international resources and macroeconomic stability. This should result in improved economic performance, with accelerated growth in output and employment and, as a consequence, a reduction of absolute poverty (Athukorala, 1998). Since "in general, (trade) liberalization is an ally in the fight against poverty" (Mc Culloch et al. 2002), a causal relationship between globalization, economic growth and poverty is expected.

Throughout this paper the influence of structural reforms (that took place in many countries in Latin America during the past decades) on poverty will be investigated. Since structural reforms can be directly beneficial for the population, it is expected that structural reforms have a direct influence on poverty.

It is expected that:

Hypothesis 3: Structural reforms have a direct positive influence on poverty reduction.

§4.2 Model specifications

In the following paragraphs the models chosen for analysis of the relationships between globalization, economic growth, structural reforms and poverty are clarified.

4.2.1 Relationship between globalization and economic growth

To examine the relationship between globalization and economic growth, a linear regression function is estimated, whereby globalization is measured by several variables of the globalization model of Husain (2001) and a number of controls.

For the first regressions, a dlog- dlog model is chosen. This implies that all variables are in dlog levels, to interpret the parameters in regression results as elasticities. The impact of a one percentage point change of the independent variables causing a βx percentage point change of the dependent variable is revealed.

The function looks as follows:

[1] d log yit = β0 + β1 Cit + β2 d log TRADEit + β3 d log FDIit + β4 log d INTTELit + β5 d log

JOURNALSit + ε

4.2.2. Relationship between globalization and poverty

If the above described relationship is existing and labeled significant, the relationship between growth and poverty can be examined. A logit-model is chosen, since the dependent variable is a percentage that can take values only between 0 and 1. If an ordinary regression model would be chosen, there is no assurance that the predicted value will lie between 0 and 1. To make sure that such a situation does not arise, the following functional form is commonly adopted:

Ln (P /(1-P)) = β0 + β1 X + ε

Where P is the value of the dependent variable between 0 and 1. This model is known as the logit-model (Ramanathan, 1995). For the regression this means that this function will look as follows:

[2] ln (HCI$2/(1-HCI$2))it = β0 + β1 Cit + β2 d log GDPpercapitait + ε

where ln (HCI$2/(1-HCI$2))it is the logit estimation for the HCI of poverty at the $ 2 per day level, Cit are the control variables, and d log GDPpercapitait is the first difference of the natural logarithm of GDP per capita (ppp) in country i in year t and ε is the error term.

To solve the above stated relationship for HCI$2, both sides are exponentiated. The following equation is obtained:

HCI$2it = 1/(1+e-(β0+β1 Cit+β2 d log GDPpercapitait+ ε))

From the above equation it can be derived that if β < 0, then HCI$2 approaches the value 0 when d log GDPpercapita increases. If β > 0, then HCI$2 approaches the value 1 when d log GDPpercapita increases.

Since it is hypothesized that economic growth (that is a positive d log GDPpercapita) has a positive influence on poverty reduction, (since it is expected that economic growth leads to poverty reduction), it is expected that the coefficient of β2 carries a negative sign.

4.2.3 Relationship between structural reforms and poverty

Again a logit-model is chosen, since the dependent variable is a value between 0 and 1. (A more detailed description of the logit-model is given in paragraph 4.2.2.)

The function looks as follows:

[3] ln (HCI$2/(1-HCI$2))it = β0 + β1 Cit + β2 TRADEINDEXit +β3 FINANCIALINDEXit +β4

LABORINDEXit + β5 TAXINDEXit + β6 PRIVATINDEXit + ε

where ln (HCI$2/(1-HCI$2))it is the logit estimation of the HCI of poverty, Cit are the control variables, TRADEINDEXit is the reformindex of trade policy, FINANCIALINDEXit is the reformindex of financial policy, LABORINDEXit is the reformindex of labor legislation, TAXINDEXit is the reformindex of tax policy, PRIVATINDEXit is the reformindex of privatizations of country i in year t and ε is the error term.

To solve the above stated relationship for HCI$2, both sides are exponentiated again. The following equation is obtained:

HCI$2it = 1/(1+e-( β

0+ β1 Cit+ β2 TRADEINDEXit + β3 FINANCIALINDEXit + β4 LABORINDEXit + β5 TAXINDEXit + β6 PRIVATINDEXit + ε )

)

From the above equation it can be derived that if β < 0, then HCI$2 approaches the value 0 when the indices increase. If β > 0, then HCI$2 approaches the value 1 when the indices increase.

According to hypothesis 3, the variables measuring the degree of efficiency of policy reforms are hypothesized to achieve a reduction in the level of poverty. Hence the coefficients of β2,3,4,5,6 are expected to have a negative sign.

§4.3 Data collection

Data about structural reforms in Latin America were derived from Lora (2001). This author measured for 19 countries in Latin America the advance of structural reforms and calculated the status of progress in policies in the trade, financial, tax, privatization, and labor areas. These 19 countries, listed in Appendix II, formed the basis during selection of countries suitable for analysis. The data derived from Lora cover the period 1985-1999.

Ideally, the following analyses thus include 285 observations per variable, although occasionally the number of observations is lower due to missing data.

The next subsections will highlight a more detailed discussion of the dependent variable, the input variables and the control variables.

§ 4.4 Dependent variables GDP per capita

During the first regression, d log GDP per capita is the dependent variable. It is probably the most straight forward variable leading to poverty reduction, since GDP per capita (ppp) is the sum of: total private consumption, investment, government expenditure and net export (WDI, 2005). The values for GDP per capita with respect to ppp are directly derived from the WDI. It is implicated that an increase in consumption leads to a higher GDP, ceteris paribus, and vis-à-vis. Therefore, a GDP growthrate should imply lower poverty rates, so that everyone can benefit from economic growth. This relation has been investigated and found significant by many economists, like Dollar and Kraay (2001) and Roemer and Gugerty (1997). Therefore, this variable is used to analyze whether globalization as an impact on economic growth, and secondly to determine whether this growth has led to a poverty reduction in Latin America.

Poverty

During the second and third regressions, the dependent variable is the level of poverty.

Poverty consists of several dimensions, namely: nutrition, health, consumption,

powerlessness, and income levels. Since income is easier to consider and to measure than most of the other dimensions, it is usually taken as a reliable proxy for determining the "adequate level of consumption" and therefore poverty, particularly for the purposes of international comparisons (Santarelli and Figini, 2002).

During this study, the measure for poverty is an absolute measure, namely the Headcount poverty indices (HCI). They measure, as in equation [4], the percentage of the population falling below the poverty line.

where P is the number of units with income (or consumption) below an arbitrary poverty line z, and N is the total number of individuals.

This paper includes the Poverty Headcount ratio at $2 a day (PPP), measuring the percentage of people in a certain country i that live below the $ 2 per day poverty line, with respect to Purchasing Power Parity (PPP). This indicator provides an indication of the magnitude and directions of change regarding poverty.

§ 4.5 Globalization

Throughout this paper globalization is measured in 4 different dimensions, namely: trade integration, financial integration, international labor flows and technological change. These dimensions are based on the globalization model of Husain (2001), as discussed in paragraph 3.2.

4.5.1 Trade integration

The first dimension, trade integration, can be described as the free movement across international borders of goods and services. Trade integration is interpreted as total exports and imports of country i, measured in current US $ on the Balance of Payments.

Exports of goods and services comprise all transactions between residents of a country and the rest of the world involving a change of ownership from residents to nonresidents of general merchandise, goods sent for processing and repairs, nonmonetary gold, and services. Imports of goods and services comprise all transactions between residents of a country and the rest of the world involving a change of ownership from nonresidents to residents of general merchandise, goods sent for processing and repairs, nonmonetary gold, and services (WDI, 2005).

In order to measure whether a one percentage change in trade causes a βx change in yit, the natural logarithm of the total sum of exports and imports is taken. This number will be used as an initial proxy for the trade openness of an economy.

[5] trade integration = log [Exports + Imports]

4.5.2 Financial integration

Secondly, globalization is measured as financial integration: the free movement of financial flows across international borders. Financial integration refers to an individual country's linkages to international capital markets (Pradad et al ,2004). The authors mention financial integration as an aggregate concept that refers to increasing global linkages created through cross-border financial flows.

Gross foreign direct investment is the sum of the absolute values of inflows and outflows of foreign direct investment recorded in the balance of payments financial account. It includes equity capital, reinvestment of earnings, other long-term capital, and short-term capital. This indicator differs from the standard measure of foreign direct investment, which captures only inward investment. The indicator is calculated as a ratio to GDP in U.S. dollars (WDI, 2005). This indicator is multiplied with GDP, in order to obtain the absolute values of gross foreign direct investment. Of this absolute values the log is taken to reveal the impact of an one percentage change in gross FDI on GDP. [6]

There is much debate on whether FDI can be used to interpret financial openness. Figini and Santarelli (2004) argue that FDI does not fully account for both sides of financial openness. Skeptical people argue that financial liberalization feeds all kind of negative aspects like unemployment, financial speculation and poverty. According to this point-of-view, FDI is strongly correlated with short-term speculative capital movements. However, FDI measures the ability of a country to attract foreign investment in the medium to long-term. Therefore, FDI might capture the negative effect of portfolio movements in the short-term (Figini and Santarelli, 2004).

[6] Financial openness = log [ratio gross FDI * GDP]

Prasad et al. (2003) state that “While financial globalization can, in theory, help to promote economic growth through various channels, there is as yet no robust empirical evidence that this causal relationship is quantitatively very important. This points to an interesting contrast between financial openness and trade openness, since an overwhelming majority of research papers have found a positive effect of the latter on economic growth.”

4.5.3 International labor flows

Another measure of globalization are international labor flows. According to the World Bank, migration is closely linked to the labor markets. International net migration as a percentage of total population is a measure to capture the degree in which international labor flows take place. However, since the data about international migration are only available every 5 years, they are not suitable for the analysis conducted, and thus cannot be included in the analysis. Inclusion of data about international migration would lead to problems with data insufficiency for regression analysis.

International communication

International telecommunications outgoing traffic refers to the telephone traffic, measured in minutes per subscriber, that originated in the country with a destination outside the country (WDI, 2005). Since it can be assumed that friends and families that still live in the home country will keep in contact with the emigrated people, the change in outgoing international phone calls can be considered an indirect measure of International labor flows.

Another reason to include international telecommunications in the analysis is the reason that the more international communication that takes place, the more open an economy can be considered, and thus the more globalized the country is.

[7] International communication = log [International telecom] 4.5.4 Technological change

Technological change will also be taken into account when measuring globalization. It is assumed that the more globalized a country is, the larger the increase in technological change will be. To measure technological change, the percentage change in the number of publications in scientific and technical journals is taken as a proxy. Scientific and technical journal articles refer to the number of scientific and engineering articles published in the following fields: physics, biology, chemistry, mathematics, clinical medicine, biomedical research, engineering and technology, and earth and space sciences (WDI, 2005). It is assumed that the larger the increase in the number of published articles is, the higher the technological progress is.

[8] Technological change = log [number of publications in scientific and technical journals]

§ 4.6 Structural reform indices

Lora (2001) created a Structural Reform Index, based on five sub-indices in the areas of trade, finance, taxation, privatization and labor for 19 countries of Latin America and the Caribbean.

The indices make it possible to compare the status of different policy areas within the same country or of each policy between countries.

The different policy reform indices are used to compare the degree of effectiveness of reforms in Latin America. The closer the number is to 1 the more reforms have taken place to improve the efficiency. The indices enable to make a cross-country comparison about the impact of structural reforms in different fields.

[9] trade reforms = [tradeindex] [10] tax reforms = [taxindex]

[11] privatization reforms = [privatization index] [12] labor legislation = [laborindex]

[12] financial reforms = [financialindex]

§ 4.7 Control variables

By introducing control variables the variance of the dependent variable explained by the models described above, increases. Therefore, a number of control variables are added to elaborate relations between globalization, structural reforms and poverty reduction.

Literacy rate

Recent studies have shown that there exists evidence that human capital has a significant and positive effect on economic growth. The report of the U.N. Economic Commission for Latin America and the Caribbean (ECLAC, 2000) notes that the accumulation of human capital (especially health and education) is not only important for the well-being of children in the short run, but also for reducing their vulnerability to poverty when they become adults. A very basic measure of human capital is the literacy rate in a country. Although the literacy does not make a skill distinction among literate people, it provides a general indication of the prospects of a country.

In a study conducted by Gregorio (1991) it was found that in the Southern Latin American countries the literacy rate had an important positive effect on growth whereas the relatively low literacy rate in Bolivia, Brazil and Guatemala had a negative impact on growth.

Since many authors find a positive and significant relationship between literacy and growth, the literacy rate is used as a proxy for human capital, and thus is added as a control variable.

[13] Literacy = [percentage of population above 15 that is literate]

Urban population

transformed in tandem for industrial development to begin, in order to achieve economic growth.

Furthermore, studies have shown that the people that live in urban regions, are on average less poor than people that live in rural areas. Increased urbanization can therefore be explained in terms of income differentials received by workers in urban vis-à-vis rural occupations. (Jemio and del Carmen Choque, 2003).

In general, cities provide many opportunities for employment of workers, and create a possibility of achieving economic growth. For the population it is therefore more attractive to live in the city than in rural areas.

Hence, the share of the population living in urban areas is used as a control variable, whereby is assumed that the more urbanized a country is, the lesser the degree of poverty is in that specific country, since urbanization can have a positive effect on poverty reduction. Furthermore, the change in the population living in urbanized regions is assumed to have a positive effect on economic growth.

[14] urban population = [percentage of population living in urban regions]

Unemployment

A higher employment rate leads to lower poverty levels, or at least less deep seated poverty. Unemployment therefore has a negative impact on poverty. Furthermore it is assumed that unemployment has a negative effect on economic growth. The derivative of the percentage of the population that is unemployed is taken as a proxy to measure the change in unemployment in percentage points.

[15] Unemployment = [percentage of population that is unemployed]

Liberalized

Countries that were liberalized in a certain year were given a value of 1 and the ones that were still closed were given a value of 0 (dummy variable). It is assumed that liberalized countries on average achieve higher growth rates than not liberalized countries. This is the main reason why liberalization is added as control variable.

Landlocked

"Bolivia is a landlocked country, up in the Andean mountains, facing incredibly high transport costs. The only products that Bolivia has ever been able to export are commodities with a very high value per unit weight because only those commodities can successfully overcome the high transportation costs." (David Morawetz, World Bank consultant. In: Jeffrey Sachs, 2005). Research shows that countries that have a coastline have an advantage in (international) transportation compared to the landlocked countries in the area of international trade. In the period 1960 – 1992, landlocked developing countries on average faced a 1.5 per cent lower economic growth per year compared to developing countries that are not landlocked (MacKellar et al, 2000).

PART 5: FINDINGS

§5.1 Analysis of relationship between globalization and economic growth

Before the actual regressions are run, several tests are performed in order to check whether the regression outcomes can be interpreted without biases. These tests include a Unit Root test, correlation coefficients of independent variables, auxiliary regressions, the bounds test and residual plots.

5.1.1 Unit Root Test

Since time-series are involved in the analysis, it is important to test whether these time series are stationary, meaning that the series’ mean and variance are constant over time, and the covariance between two values from the series depends only on the length of time separating the two values, and not the actual times at which the variables are observed (Hill e.a., 2001). When using non-stationary series in regression analysis, spurious regressions might occur, indicating a significant relationship when there is none.

In Appendix III, Augmented Dickey-Fuller (ADF) Unit Root Tests are performed to test whether the economic variables used do not have a trend and thus influence the interpretation of the regression. Furthermore, the possible trends in the economic variables are displayed graphically. The control variables unemployment, urbanpop and the literacy are rather constant over time, and do not show a trend, so these variables can be used without biases in the analysis. This is also the case for the dummy variables landlocked and liberalized.

By using the results of the ADF test the null-hypothesis of non-stationarity cannot be rejected, for the all variables measuring globalization: logtrade, loggrossfdi, loginttel and logjournalarticles. Furthermore non-stationarity cannot be rejected for the dependent variable, which is log GDPpercapita. Hence it can be concluded that of all these variables, the first-difference is taken in order to fulfil the requirement of stationarity.

5.1.2 Multicollinearity

All independent variables are tested for multicollinearity, using a common rule of thumb. This rule states that if the correlation coefficient exceeds the absolute critical value of 0.8, there is evidence of a strong linear association and a potential harmful collinear relationship between two independent variables, and thus a problem with multicollinearity might occur (Hill et al., 2001).

From the table below, it can be derived that for none of the variables this critical value of 0.8 is exceeded.

Furthermore, auxiliary regressions are performed to test whether variation in one of the explanatory variables is not explained by more than two other explanatory variables. If one of

the regressions exceeds a value of the R2 of 0.7 or 0.8, multicollinearity in the data occurs. In

the table below the auxiliary regressions for the independent variables are displayed.

Auxiliary regressions R2

d log TRADE = β0 + β1 d log GROSSFDI + β2 d log INTTEL + β3 d log JOURNALARTICLES + β4 LITERACY + β5

UNEMPLOYMENT + β6 URBANPOP + β7 LANDLOCKED + β8 LIBERALIZED

.0528

d log GROSSFDI = β0 + β1 d log TRADE + β2 d log INTTEL + β3 d log JOURNALARTICLES + β4 LITERACY + β5

UNEMPLOYMENT + β6 URBANPOP + β7 LANDLOCKED + β8 LIBERALIZED

.0610

d log INTTEL = β0 + β1 d log GROSSFDI + β2 d log TRADE + β3 d log JOURNALARTICLES + β4 LITERACY + β5

UNEMPLOYMENT + β6 URBANPOP + β7 LANDLOCKED + β8 LIBERALIZED

.0901

d log JOURNALARTICLES = β0 + β1 d log GROSSFDI + β2 d log INTTEL + β3 log TRADE + β4 LITERACY + β5

UNEMPLOYMENT + β6 URBANPOP + β7 LANDLOCKED + β8 LIBERALIZED

.0331

LITERACY = β0 + β1 d log GROSSFDI + β2 d log INTTEL + β3 d log JOURNALARTICLES + β4 d log TRADE + β5

UNEMPLOYMENT + β6 URBANPOP + β7 LANDLOCKED + β8 LIBERALIZED

.4879

UNEMPLOYMENT = β0 + β1 d log GROSSFDI + β2 d log INTTEL + β3 d log JOURNALARTICLES + β4 LITERACY + β5 d

log TRADE + β6 URBANPOP + β7 LANDLOCKED + β8 LIBERALIZED

.1144

URBANPOP = β0 + β1 d log GROSSFDI + β2 d log INTTEL + β3 d log JOURNALARTICLES + β4 LITERACY + β5 d log

TRADE + β6 UNEMPLOYMENT + β7 LANDLOCKED + β8 LIBERALIZED

.5488

From the table above it can be derived that for the data used in this analysis there are no problems with multicollinearity. (There are no auxiliary regressions for the variables landlocked and liberalized, since both these variables are dummy variables.)

5.1.3 Panel data

Panel data usually use fixed effects as an assumption for the intercept of the regression, a common intercept for the regression is unusual (Westbrook, 2001). In a fixed effects model, the regression coefficients are assumed to be common across cross-sectional units, but the regression intercepts are taken to be distinct across cross-sectional units, in this case the 19 Latin American countries. However, since the data analyzed during this study only have a relative short time-series, the fixed effect model is not applicable, due to the lack of data.

DLOGTRADE DLOGGROSSFDI DLOGINTTEL

DLOGJOURNALA

Instead, it is assumed that the regression has one common intercept across all cross-sectional units.

The assumptions for the Partial Regression Coefficients are also common. This implies that at least some of the partial regression coefficients are common across cross-sectional units. Ordinary Least Squares are used as a method to estimate the regression, taking into account White-heteroskedasticity consistent standard errors of the variables. No weighting is used, since no equation-specific heteroskedasticity is assumed.

5.1.4 Autocorrelation

A bounds-test is performed to test for first-order autocorrelation. This test uses the Durbin-Watson test statistic with an upper and lower bound for the 5 percent confidence interval. These values are derived from a statistical table and depend on the number of explanatory variables including the constant term (k) and the number of observations (T). The number of observations will be 15, since this is the time span that is investigated.

For model 1, 2, 3 and 4 (described in the next section, section 5.1.5), the according critical values for the Watson test are displayed below. Furthermore the observed Durbin-Watson values with OLS-regressions are given:

Table 2: critical and observed Durbin-Watson values with OLS

d*L d*U D*obs

Model 1 .814 1.750 1.4386

Model 2 .447 2.472 1.4425.

Model 3 .251 2.979 1.4416

Model 4 .251 2.979 1.4097

Since during the OLS-regressions all observed Durbin-Watson statistics were within the critical boundaries of the Durbin-Watson statistic, the test is inconclusive about whether or not there is autocorrelation. But since the hypothesis of autocorrelation cannot be accepted, no autocorrelation will be assumed, so all regression estimates will be while using OLS.

5.1.5 Model estimations

the independent variables is included. In this model it is assumed that the immediate effect of changes in trade and international telecom is negligible, but the main effect of these changes may be felt during a later period, namely several years later.

A time-lag of 1,2,3,4 and 5 years was investigated. It turned out that a time lag of 1 year yielded the best results with respect to significance and variations in the standard errors of the variables, as well as fit of the model, therefore this time-lag was chosen.

During all regressions, White-heteroskedasticity consistent standard errors are used. In the table below these regressions are visualized:

Table 3

Ordinary Least Squares (OLS): Results with log GDP per capita (ppp) as the dependent variable N=227

Number of cross sections: 19

Model 1 2 3 4 Independent variables Intercept .0226*** (.0037) -.0242 (.0233) -.0256 (.0240) -.0148 (.0252) Predictors D Log Trade .1347*** (.0275) .1280*** (.0308) .1259*** (.0291) .0598 (.0426)

D log gross FDI -.0057*

Adjusted R–squared .1777 .1912 .1875 .0483 *** p < .01, ** p < .05, * p < 0.1 , standard errors are given in parentheses.

Model 1 tests the significance of the predictors d log trade, d loggrossfdi, dloginternational telecom and dlog journalarticles. From the table above it can be derived of the predictors only dlog journalarticles does not turn out to be significant.

In model 2 the controls unemployment, urban population and the literacy rate are added to test hypothesis 1. This hypothesis predicted a positive relationship between globalization and economic growth, thus a positive relation between the variables. Again d log trade and d log international telecom are both highly significant, and carry the right sign. D log Gross FDI and dlog journalarticles both carry a negative sign, while a positive relationship was expected. Nevertheless, both variables are not significant.

Unemployment has a negative influence on GDPpercapita, just as predicted. However, the literacy rate and urbanpopulation also feature a negative sign, implying that these rates have a negative impact on GDPpercapita. And in this case, this variable is not labelled significant. Model 3 extends the previous model by also including the dummy variables landlocked and liberalized. This is done in order to test whether these variables improve the fit of the estimated model. The F-test shows that the regression slope coefficients are significantly different from zero, which is also the case for the t-tests for the beta coefficients for the independent variables logtrade, loginttel included and unemployment in the model. Most signs turn out to be in line with the predictions; only the literacy rate and urbanpopulation are negative whilst a positive sign was expected.

It can be concluded that the fourth model, only unemployment turns out to be highly significant, while all other variables can not be labelled significant anymore. In only the first model the intercept turns out to be significant.

5.1.6 Regression interpretation

Since a double d log model is used for the regression analysis, β can be interpreted as the percentage point increase in the dependent variable associated with a 1 percentage point change in the independent variable. Relating the above information to the initial equation, the estimated relationship between globalization and growth can be displayed in the following way, keeping in mind the limitations mentioned above:

2) d Log GDPpercapita = -.0242 + .1280 d log TRADEit - .0058 d log GROSSFDIit + .0511 d log INTTELit - .0022 d log JOURNALARTICLESit – .0700 LITERACY - .1170 UNEMPLOYMENT - .0052 URBANPOP

3) d Log GDPpercapita = -.0256 + .1259 d log TRADEit - .0055 d log GROSSFDIit + .0521 d log INTTELit - .0018 d log JOURNALARTICLESit - .0793 LITERACY – .1309 UNEMPLOYMENT - .0149 URBANPOP - .0049 LANDLOCKED + .0026 LIBERALIZED

§ 5.2 Analysis of relationship between globalization and poverty

Again several tests are conducted before the actual regressions are run, in order to avoid biases in the regression outcomes.

5.2.1 Unit Root test

The dependent variable, which now is the logit estimation of the HCI$2, log(HCI$2/(1-HCI$2)), is tested for stationarity, in order to be sure that the time-series data do not have a linear or exponential time trend (Ramanathan, 1995). Once more, an Augmented Dickey-Fuller (ADF) Unit Root Test is performed to test the variable for possible trends that might influence the interpretation of the regression. In Appendix V the outcomes of the test are displayed, as well as a graph of the mean of the logit estimation of the HCI$2. According to the results of the ADF test, the null-hypothesis of non-stationarity should be rejected. Therefore, it can be concluded that the logit estimation of the HCI$2 is stationary and thus can be used during the regressions.

The assumption of stationarity also holds for the control variables, as well as for the first difference of GDPpercapita (see paragraph 5.1.1 for a detailed explanation).

5.2.2 Multicollinearity

Again a correlogram is created to test whether problems with multicollinearity might occur during the regressions. This table is displayed below. Since none of the correlation coefficients is equal to or exceeding the critical value of 0.8, multicollinearity is not present during these regressions.

Table 4: Correlogram

DLOGPPPGDPPERCAPITA UNEMPLOYMENT URBANPOP LITERACYPER LANDLOCKED LIBERALIZED DLOGPPPGDPPERCAPITA 1.000000 UNEMPLOYMENT -0.113590 1.000000 URBANPOP 0.068940 0.078046 1.000000 LITERACYPER 0.118188 0.082109 0.669943 1.000000 LANDLOCKED -0.030608 -0.232565 -0.276736 0.026112 1.000000 LIBERALIZED 0.127422 -0.048444 0.020438 0.113398 0.073713 1.000000

Auxiliary regressions R2 D log GDPpercapita = β0 + β1 UNEMPLOYMENT + β2 URBANPOP + β3 LITERACY + β4 LANDLOCKED + β5

LIBERALIZED

.0470

LITERACY = β0 + β1 d log GDPpercapita + β2 UNEMPLOYMENT + β3 URBANPOP + β4 LANDLOCKED + β5

LIBERALIZED

.5166

UNEMPLOYMENT = β0 + β1 d log GDPpercapita + β2 LITERACY + β3 URBANPOP + β4 LANDLOCKED + β5

LIBERALIZED

.0853

URBANPOP = β0 + β1 d log GDPpercapita + β2 LITERACY + β3 UNEMPLOYMENT + β4 LANDLOCKED + β5

LIBERALIZED

.5396

It can be concluded that since none of the auxiliary regressions exceed the absolute value of 0.8, there are no problems with multicollinearity.

5.2.3 Autocorrelation

In the table below the critical Durbin-Watson upper and lower bound are displayed, plus the observed Durbin-Watson from the OLS-regressions conducted with respect to the relationship between d log GDPpercapita and poverty. (See table 5.)

For model 1, 2, 3 and 4 the according critical values for the Durbin-Watson test are displayed below. Furthermore the observed Durbin-Watson values with OLS-regressions are given:

Table 5: critical and observed Durbin-Watson values with OLS

d*L d*U d*obs

Model 1 .946 1.543 .6626

Model 2 .562 2.220 .8567

Model 3 .343 2.727 .9186

Model 4 .343 2.272 1.1123

Since only for the first model the calculated Durbin Watson value is smaller than d*L (.6626 < .946), it can be concluded that during that regression autocorrelation is present. For the other models there is no evidence for autocorrelation, since the observed Durbin-Watson value lies between the upper- and lower-bound value. In this case, the test is inconclusive. Therefore, only the first model will be re-estimated. The most common model that can be used to represent correlated errors is the first-order autoregressive model AR(1) model. This model implies that the largest correlation between errors is for those that are one period apart; for those errors further apart, the correlations are smaller and smaller. The characteristic of an AR(1) error model seems reasonable for many economic phenomena (Hill et al., 2001). In this model the error εt depends on its lagged value εt-1 plus another random component that is uncorrelated over time and has zero mean and a constant variance (Hill et al., 2001). In the table in paragraph 5.1.4, table 3, these AR(1) model regressions are visualized.

In table 5 in paragraph 5.2.3 the regressions are all displayed.

5.2.4 Model estimation

For the regressions that investigate the relationship between d log GDP per capita and HCI$2, a logit transformation of HCI$2 is used (see paragraph 4.2.2 for information about the transformation of HCI$2), since this variable is a percentage and thus cannot be treated as an absolute variable. However, since the values for HCI$2 always lie between 0 and 1, the estimation method is simply to transform this variable and obtain Y= ln (HCI$2/(1-HCI$2)). Y is then regressed against a constant and all explanatory variables (Ramanathan, 1995).

Again all analyses are conducted using a panel data set. In paragraph 5.1.2 a detailed description of the assumptions of the panel data is given.

The model is estimated by Least Squares (OLS). The first regression includes the independent variable d log GDPpercapita as predictor of the dependent variable, without controls. . This regression includes the AR(1) variable, since this variable corrects for first-order autocorrelation in the error term. In an AR(1) model the estimated coefficient, the coefficient standard errors, and t-statistics may be interpreted in the usual manner. The coefficient of the AR(1) model is the serial correlation coefficient of unconditional residuals (Eviews Users Guide, 2004).

The second regression is estimated by OLS. The independent variable d log GDPpercapita is regressed again against the logit estimation of HCI$2, while several controls are added, namely unemployment, urban population and the literacy rate. During the third regression, also landlocked and liberalized are added as control variables.

In the last model, again a time-lag is included, since it is assumed that the effect of a change in d log GDPpercapita is not directly affecting poverty in the same period, but during a later period. Time-lags for 1,2,3,4 and 5 years were tested. It turned out that a time-lag of 5 years seemed to provide the best test-results with respect to significance and variations in the standard errors of the variables, as well as fit of the model.

During all regressions, White-heteroskedasticity consistent standard errors are used. Below the regressions are displayed in table XX:

Table 6

Intercept -.7610*** (.2106) 5.6826*** (.5757) 5.7592*** (.5711) 6.2210*** (.7137) Predictors d Log GDPpercapita (ppp) -1.7982 (1.4188) -3.9375 (2.7886) -3.3829 (2.5278) -4.6007 (2.7967) Control variables Unemployment -2.2811** (.9857) -1.8218* (1.0841) -2.0216** (.9835) Urban population -.0830 (.5928) .5619 (.5296) .8743 (.6582) Literacy rate -7.5129*** (.9895) -8.6038*** (1.0337) -8.5136*** (1.3840) Landlocked .5296*** (.1477) .6501*** (.1629) Liberalized .3977*** (.1512) -.1996 (.0046) AR(1) .7966*** (.2676) F-statistic 27.13611 (.0000)*** 24.35309 (.0000)*** 18.18175 (.0000)*** 12.33140 (.0000)*** Durbin Watson .3605 .8567 .9186 1.1123 Adjusted R–squared .5990 .4932 .5178 .4963

*** p < .01, ** p < .05, * p < 0.1 , standard errors are given in parentheses.

Model 1 tests the significance of the predictor log GDPpercapita and AR(1). From the table above it can be derived that the predictor is not significant. However, the AR(1) and the intercept turn out to be very significant.

In model 2 the controls unemployment, urban population and the literacy rate are added to test hypothesis 2. A negative relationship between d log GDPpercapita and poverty is expected, since poverty is expected to decline parallel with an increase in d log GDPpercapita. During the second regression GDPpercapita is, as expected, negatively correlated with ln (HCI$2/(1-HCI$2)), although is not significant. The literacy- and unemployment rate as well as the intercept are significant in the second model. Although the unemployment is significant it is calculated by the model that it has a negative impact on poverty whereas a positive relation is expected. The urban population rate is not significant nor has it the right sign. From the table above it can be derived that there exists a negative and positive relation between literacy rate and poverty, unfortunately a positive relation is expected.

different from zero, which is also the case for the t-tests for the beta coefficients for the independent variables unemployment, literacy rate, landlocked and liberalized, all included in the model.

It is expected that liberalization has a negative impact on ln (HCI$2/(1-HCI$2)), the regression outcome proves this expectation. The other added variable, landlocked, is expected to have a positive influence on poverty. (Implying that being landlocked results in more poverty, ceteris paribus). Unfortunately, the regressions shows the contrary. The unemployment and the literacy variables are both significant, but only the unemployment rate carries the right sign.

The fourth and last model, where a time-lag is build in, produces the same result as the previous model except that now liberalization has a negative effect on poverty, as is assumed, but is no longer significant. The time-lag does not significantly improve the significance of the variables used and therefore there is no support to use a time-lag in the model. Unfortunately,

the only predictor in all the models, log GDPpercapita, is in non of the models valid.

5.2.5 Regression interpretation

Since a d log model is used for the regression analysis, β can be interpreted as the percentage point increase in the dependent variable (which is ln (HCI$2/(1-HCI$2)) associated with a 1 percentage point change in the independent variable. Relating the above information to the initial equation, the estimated relationship between globalization and growth can be displayed in the following way, keeping in mind the limitations mentioned above:

1. ln (HCI$2/(1-HCI$2))it = -.7610 – 1.7982 d log GDPpercapitait

2. ln (HCI$2/(1-HCI$2))it = 5.6826 – 3.9375 d log GDPpercapitait – 7.5129 LITERACY – 2.2811 UNEMPLOYMENT -.0830 URBANPOP

3. ln (HCI$2/(1-HCI$2))it = 5.7592 –3.3829 d log GDPpercapitait – 8.6038 LITERACY – 1.8218 UNEMPLOYMENT + .5619 URBANPOP + .5296 LANDLOCKED + .3977 LIBERALIZED