THE EFFECT OF
GLOBALIZATION ON
INCOME INEQUALITY
IN THE NETHERLANDS
Abstract
This paper investigates to which extent globalization affects income inequality in The Netherlands. As a measure for globalization and income inequality we use the KOF-index and the Gini coefficient, respectively. A time series analysis is conducted on annual observations, in the period 1989-2013. As a result of both beta regression and OLS regression, no significant relation between globalization and income inequality is found. Empirical analysis did identify two significant variables: unemployment and life expectancy. But both variables have a different relation with income inequality than is suggested by the literature. A possible reason for these odd results is the small number of observations.
Bsc Thesis Economics
Florian Korn
June 2016
10459944
1
Statement of Originality
This document is written by Student Florian Korn who declares to
take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is
original and that no sources other than those mentioned in the text
and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for
the supervision of completion of the work, not for the contents.
2
Index
1. Introduction ... 4
2. Literature Review ... 6
2.1 Different aspects of globalization ... 6
2.1.1 Economic globalization ... 6
2.1.2 Sociocultural globalization ... 7
2.1.3 Political globalization ... 8
2.1.4 Technology the accelerator for globalization ... 8
2.2 KOF-index of globalization ... 9
2.3 Developments in income inequality ... 10
2.3.1 Global inequality developments ... 10
2.3.2 Dutch inequality developments ... 10
2.4 Gini coefficient ... 11
2.5 Theoretical framework ... 12
2.5.1 Relationship globalization and income inequality ... 13
2.5.2 Other determinants of income inequality ... 14
2.5.3 Conclusion ... 14
2.6 Hypothesis ... 14
3 Data ... 16
3.1 Dependent variable ... 16
3.2 Independent variables ... 16
4. Estimation Method and methodology ... 18
4.1 Estimation method ... 18
4.2 Methodology ... 19
4.2.1 Beta regression ... 19
4.2.2 Ordinary least-squares regression ... 20
5. Results ... 22
5.1 Beta regression ... 22
5.2 OLS regression ... 24
6. Conclusions ... 25
3 Appendix ... 28
4
1.
Introduction
The effect of globalization on income inequality has been studied extensively. But there is no definitive conclusion on the effect of globalization on income inequality.
Over the last few decades globalization has increased worldwide. Figure 1 shows the development of globalization in The Netherlands. The graph shows that globalization increased steadily until the 2000 and since then it stabilized slightly above ninety in the KOF-index of globalization. Indicating that The Netherlands is a highly globalized country.
Figure 1: KOF-index of The Netherlands!
Source: KOF Swiss Economic Institute According to the United Nations, globalization is a multi-dimensional process
characterised by: the acceptance of a set of economic rules for the entire world designed to optimize productivity by universalising production and markets, and to obtain the support of the state with a view to making the national economy more competitive and productive (UNESCO, 2016).
Bechtel argues that globalization reduces income inequality and causes the economic anxiety to decrease thereby increasing the consumer demand, institutional trust and general well-being (2014).
Dreher and Gaston’s research on the other hand reveals a relationship between globalization and labour market participation (2008). Furthermore they argue that the increased degree of globalization worsened the income inequality. OECD countries in contrast to less-developed nations experience a much greater impact of globalization (Dreher & Gaston, 2008). Bhugwati argues that this effect is caused by the willingness
5 to accept the integration and globalization (1999). He has observed that less-developed nations oppose integration while OECD countries embrace globalization (Bhugwati & Srinivasan, 1999).
Jaumotte et al. (2013) report that globalization is made up of two components: trade and financial globalization. Jaumotte finds that there is an opposite effect in both parts of globalization on income inequality. An increase in trade openness reduces the inequality, while financial openness increases inequality (Jaumotte, 2013).
The lively discussion of the effect of globalization on income inequality is not at its end. Since most studies focus on panel-series data and thereby compare countries. We will focus on the effect of globalization on the income inequality in the Netherlands, thereby using a time-series approach. Researching this relation can lead to policy
recommendations for the Dutch government with respect to their attitude towards further globalization.
To measure the effect of globalization on income inequality, the KOF-index is regressed on the Gini coefficient with several explanatory variables. Two methods of regression are used, beta regression and OLS regression. The dataset combines a period from 1989 until 2013, so there are 25 observations.
As a result of the empirical analysis finds no significant relation between
globalization and income inequality. Which is an unsatisfying result. On the other hand, there is a significant relation between globalization and two explanatory variables: unemployment and life expectancy. But these relations are opposite from what literature suggests.
The paper is organized as follows. The next chapter covers the literature. Then the data is presented is the third chapter. The estimation method and methodology is described in chapter four. Chapter five presents the results of empirical analysis and chapter six concludes.
6
2. Literature Review
This chapter will discuss the definitions and the different aspects of globalization and income inequality. This is done by looking at historical developments globally and for The Netherlands in specific. The goal of this chapter to try to understand the underlying relation between them. First, we will discuss the different aspects of globalization, such as the definition and its origin. Secondly, the KOF-index (acronym for the German word "Konjunkturforschungsstelle", which means business cycle research institute) of globalization is explained. Section 2.3 will discuss the developments of income inequality. Then the Gini coefficient, named after Italian statistician Corrado Gini, is introduced. After that a theoretical framework will describe several studies and give an overview of the existing literature, including their methods and important variables for income inequality and globalization. Finally 2.6 will state our hypothesis.
2.1 Different aspects of globalization
Globalization in its general sense is the process of international integration arising from the interchange of world views, products, ideas and other aspects of culture. The process of globalization contains an economical, a cultural and a political load (Chomsky, 2010). But according to Chomsky there is also a more technical definition of globalization. Globalization in its technical sense, refers to a set of liberal and protectionist measures designed to serve the interests of financial institutions and investors (Chomsky, 2010). Globalization itself is a broad term which will include many variables. Below we will discuss different parts of globalization, i.e. economic, sociocultural, political and technological globalization more extensively.
2.1.1 Economic globalization
Economic globalization can be separated into two categories: financial and trade globalization (Jaumotte et al., 2013). An increase in economic globalization is caused by a decline of import tariffs and creation of free trade zones. Due to these changes companies are able to distribute their products all over the world. Companies are also able to decrease production costs by hiring cheap labour and by making use of special taxation regulations (Jaumotte, 2013).
The financial sector, the international capital market and the commodity market are a driving factor for economic globalization. Due to the increased use of the internet and the development that less cash is being used, the international capital market has become more globalised (Jaumotte, 2013).
But for some historical perspective, even in the late nineteenth century capital export was enormous, due to the possession of colonies (O'rouke, 2001). Which collapsed in 1920s, as a result of ‘the great depression.’ Since the 1970s the financial
7 sector has become increasingly more important. Currently the international capital market is embedded in western nations. The main reason for this is the presence of free traffic of capital between nations and continents and the existence of the internet. Another aspect of a more interconnected international capital market is the sharp
increase of financial assets traded at the end of the twentieth century, compared with the beginning of the century. The international commodity market developed in a similar way. It mostly increased due to the development that import tax and import quotes declined since 1950 and a lower cost price caused by a decrease in transportation costs (O'Rourke, 2001).
The financial and global economic crisis of 2008 is a prime example of the global economic integration (Taylor, 2009). Where a subprime mortgage problem in the US causes banks to fail, eventually plummeting most of the Western world into a recession.
2.1.2 Sociocultural globalization
The cultural aspect of globalization is mostly caused by the increased connectivity of the world, expansion of multinational companies and migration of people across national borders. A famous example of cultural globalization is the Americanization of the world. In which the presence of multinationals such as McDonalds and Starbucks can be a measure of the degree of cultural globalization (Beck et al., 2003).
International migration is a factor which has a large influence on social globalization. According to O'Rourke, migration was easier at the beginning of the twentieth century then it is today, while there were less barriers for migration (O'Rourke, 2001). Between 1965 and 1990 the share of migrants in the European population increased on average from 3.6% to 6.1% (O'rouke, 2001).
Since the Second World War there were many demographical changes in the Netherlands. During the 50s the Netherlands was poor and recuperating from the war which caused and enormous outflow of Dutch citizens towards North-America, South-Africa and Australia. However due to an unexpected fast reconstruction and increased industrialization the Netherlands was short in labour force, directly after the second world war lasting until the beginning of the 60s (De Lange, 2007). This shortage was resolved by attracting low schooled labour, mainly from Turkey and Morocco. During the 1970s the Dutch government allowed for family reunification, causing another increase in Turks and Moroccans. A more recent example of migration is the arrival of Syrians and North-African refugees. Due to local wars and crises back home. The Netherlands is willing to shelter and eventually integrate these people in the Dutch society, which leads to a more culturally enriched society and social globalization (Vluchtelingewerk, 2016). Finally, the Netherlands signed a treaty, together with all other European Union (EU) countries, for free movement of labour within the EU, leading towards more international migration within the EU (De Lange, 2007).
8
2.1.3 Political globalization
The third aspect of globalization is political. Political globalization is best described as a transition of political power from nations to supranational institutions. Such as the European Union, G8 and the international criminal court.
The political aspect is closely related to the economic aspect while political willingness can lead to free trade zones, a decline in import tariffs, removal of import quotes and creation of a monetary union (Brawley, 2008).
The Netherlands is one of the founders of the European Union, which was founded in 1958. The main reasons for creating this supranational entity was increasing trade between EU countries, increasing welfare for participating countries, forming a block against the Soviet Union and setting European standards (Europa, 2015). In 2015 the EU counts 28 members and has partly turned into a monetary union, where the Euro is used as a currency. To make this change happen all involved countries had to hand over the power of their national central bank to the European Central Bank. This transition from national institute to a supranational institute is an example of political integration, or globalization (Europa, 2015).
Currently the EU is negotiating a trade and investment partnership with the United-States (US), called TTIP (Transatlantic Trade and Investment Partnership). Which leads to further integration and globalization between the EU and the US. Negotiation is done by the EU itself, without consulting EU members (Rijksoverheid, 2015).
2.1.4 Technology the accelerator for globalization
The final aspect of globalization is technology. Technology has a great influence on globalization, while improvements in transportation causes an increase in
interconnectivity of the globe. We will briefly discuss two important examples of technological developments which caused an increase in globalization.
Transportation costs were greatly reduced by the invention of steam, this helped in the development of steamboats and trains in the nineteenth century (O'Rourke, 2001). O'Rourke explains, that between 1870 and 1913, the transportation costs reduced by 45 percentage points. This decrease in transportation costs led to an increase in
globalization. As proof, O'rouke finds that at the beginning of the twentieth century international prices for goods have converted, due to the increase in technology
(O'Rourke, 2001). During the twentieth century transportation costs continued to drop – mostly due to the invention of flight and cheaper transport flights. Leading to a tenfold increase in the ratio air to ocean shipments in 1962 (O'Rourke, 2001). Even in the past decade transport has been getting cheaper due to an increase in international
9 A recent development is the globalizing effect of the internet and social media on the worlds’ economy and the way people interact. Nowadays broadcast stations will spread their breaking news via social media so everybody with an internet connection can be informed directly (Cozma, 2013). Companies can even rise productivity and efficiency by interacting with consumers with social media. This can be seen as a way in which technology incubates globalization.
2.2 KOF-index of globalization
A proxy for overall globalization is the KOF-index, a number between 0 and 100 where 100 indicated full globalization. The Swiss Economic Institute gathered data of 207 countries for the period 1970-2013. The KOF-index covers the social, economic and political dimensions of globalization. It describes globalization as a process of creating networks among people and nations at multi-continental distances through ideas, capital, goods and information and people. The KOF-index conceptualizes the integration of national economies, cultures, governance and technologies. The KOF-index is split up in three categories: economic, political and social globalization (See appendix, KOF variables). Where economic globalization covers actual flows - as trade and foreign direct investment both as a percentage of gross domestic product (GDP) - and restrictions, which covers import tariffs and import barriers.
Social globalization has three categories: personal contacts, information flows and cultural proximity. Personal contacts measure the interaction between people who live in different countries. Information flows, on the other hand, capture the potential exchange of ideas between nations. Lastly, cultural proximity measures how closely related cultures are by looking at the trade in book as a percentage of GDP (KOF Swiss Economic Institute, 2015).
Political globalization encompasses variables such as embassies in a country and the number of international treaties.
The KOF-index is a weighted average of these three categories, where economic, social and political globalization contributes 27%, 36% and 37% respectively. For the full sets of variables and their weights see appendix (KOF variables).
Figure 2 of the KOF-index of The Netherlands shows an upward trend since 1970 until 2013. Indicating an increase in the level of globalization in The Netherlands over the same period.
10
Figure 2: KOF-index The Netherlands
Source: KOF Swiss Institute
2.3 Developments in income inequality
To research the potential effect of globalization on income inequality this section will firstly give a brief summary of the global inequality trends. Thereafter the Dutch inequality trends will be discussed. Finally, the Gini coefficient is presented.
2.3.1 Global inequality developments
During the 1960s O'Rourke finds that inequality increased. This overall increase in world inequality was mainly caused by the inequality between countries rather than inequality within countries (O'Rourke, 2001). However, according to Melchior (Melchior et al., 2000), between 1965 and 1997 the Gini coefficient decreased
indicating a decrease in inequality – this was mainly caused by the catch-up of China.
2.3.2 Dutch inequality developments
In comparison with the global inequality developments, inequality had a different development in the Netherlands. Mostly due to the fact that the Netherlands is a
developed country with a strong social safety net and a progressive income tax system – which requires the highest incomes to pay the highest percentage of their income tax (Belastingdienst, 2015). Figure 3 shows that the Gini coefficient in the Netherlands increased between 1989 and 2013, this is an indication that income inequality in The Netherlands has deepened. This is in agreement with the findings of O'Rourke (2001).
Salverda explains this increase in income inequality (2015). By sorting households from low to high income and dividing them in ten equal groups of ten percent, Salverda clearly shows a worsening of the distribution of income. The 10%
0,74 0,76 0,78 0,8 0,82 0,84 0,86 0,88 0,9 0,92 0,94 0,96 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
11 group with the highest income increased its real income by 28%, while the bottom 40% of household experienced a decrease of real income by 3 till 7%. Over the period 1977 until 2011 (Salverda, 2015).
Figure 3: Gini coefficient of The Netherlands
Source: AMCIS and OECD
2.4 Gini coefficient
Income inequality is defined by the income gap between the highest incomes and the lowest. The common used measure of income inequality is the Gini coefficient. The Gini coefficient is equal to the area between the 45 degree line and the Lorenz curve, where the 45 degree line implies perfect equality and the Lorenz curve displays the income distribution, see figure 4. The Gini coefficient is bound by zero and one. When the Lorenz curve is equal to the 45 degree line the Gini coefficient is equal to zero, indicating perfect equality. Gini indicates perfect inequality when it approaches one. Both extremes equally unlikely. When the Gini coefficient approaches one, the income is distributed more unevenly and unequally, which is the case for The Netherlands (Celik & Basdas, 2010).
0,23 0,24 0,25 0,26 0,27 0,28 0,29 0,3 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
12
Figure 4: Lorenz curve
Source: Wikipedia
The Lorenz curve describes the relation between the percentage of people and the percentage of income those people receive, on the vertical and horizontal axis respectively. The line of equality displays complete equality, everyone has the same level of income. In reality levels of income differ, causing the Lorenz curve to shift inward. The more the top incomes differ from low incomes the steeper the Lorenz curve becomes and the bigger the area which represents the Gini coefficient.
2.5 Theoretical framework
There are many studies which have researched the relationship between globalization and inequality. There are different approaches and different definitions used to describe and measure concepts as globalization and income inequality, this section will focus on studies that use the Gini coefficient, KOF-index or identify important variables in relation to income inequality or globalization.
13
2.5.1 Relationship globalization and income inequality
Jaumotte divides globalization in two parts: trade and financial globalization (2013). He finds that trade integration and financial globalization have grown rapidly in the last two decades, while the former Sojvet-Union became increasingly more integrated in the global trading system. Another cause was the doubling of international financial assets, from 1990 to 2004 (Jaumotte, 2013). Jaumotte's model uses the log of the Gini
coefficient as the dependent variable. For variables of trade globalization Jaumotte uses various measures of trade openness, such as the average tariff rate and the ratios of non-oil imports and non-non-oil exports to GDP. For financial globalization Jaumotte uses the Chinn-ito index (index measuring a country's degree of capital account openness) to describe financial openness and different types of financial liabilities to GDP ratio's. Technology is quantified by the log of the total capital of ICT cooperation’s divided by the total stock. Jaumotte’s major finding is that technological change has had the most influence on income inequality. Financial globalization exacerbated inequality but trade globalization decreased inequality.
In agreement with Jaumotte, Celik and Basdas find that an increase in trade liberalization leads to a decrease in inequality in developing countries (2010). They conduct a panel data analysis in which they divided countries in three categories: developed, developing and miracle countries. The miracle countries used in this study are China, India, Korea, Malaysia, Singapore and Thailand.
Dreher uses the KOF-index as a proxy for globalization. Controlling for several variables and by looking at OECD countries. Dreher finds that industrial wage and household income inequality rise with globalization. He concludes that on average a one-point increase in the KOF-index leads to an in industrial wage inequality by 0.16% and household income inequality by 4% (2008).
Dollar & Kraay identified a group of developing countries that have had large increases in trade volumes and sharp cuts in import tariffs (2004). Included in this group are China, India and multiple other large countries. Which in total amount for more than half of the population of developing and globalizing economies. These countries had a higher rate of tariff cuts and a higher increase in trade volumes than developed
countries. Leading to an average growth of 5% for the selected countries, 2.2% growth for rich countries and a 1.4% growth for non-globalisers. According to Dollar & Kraay this is evidence that globalizing countries are catching up with the developed countries. As a result of cross-country regression Dollar & Kraay find that trade volumes have a strong positive relation with growth rates. They additionally find that a sharp decrease in absolute poverty, between 1984 and 2004, led to less overall inequality. The final conclusion is that open trade regimes lead to faster growth and poverty reduction, according to Dollar & Kraay (2004).
Borraz and Lopez research the connection between globalization and income inequality within Mexico (2007). For globalization they used several definitions, such
14 as a trade openness proxy and number of workers working for foreign companies. Borraz and Lopez conclude that an increase in globalization caused a decrease in income inequality (2007).
2.5.2 Other determinants of income inequality
Celik and Basdas conclude that foreign direct investment (FDI) and capital flows between countries have different effects. They find FDI outflows worsen income inequality while FDI inflows cause income to become more equally distributed (2010). But looking at total FDI Pillai finds that an increase in FDI leads to more inequality (2011).
Pillai uses multiple determinants of income inequality as independent variables such as the level of democracy, foreign direct investment, trade intensity, lagged Gini coefficient and growth in GDP per capita (2011). Pillai concludes that a more
democratic political system decreases the amount of inequality. On the other hand, he concludes that the per capita GDP variable tends to have an inverse relationship with inequality. Borraz and Lopez make use of a lagged Gini coefficient (2007).
Mocan finds that decomposing unemployment into structural and cyclical components reveals that an increase in structural unemployment decreases the shares of the bottom sixty percent of the population, and it increases the income share of the top ten percent in the United States (1999). He concludes that the US government should focus on combatting cyclical unemployment while a reduction in current unemployment may help generate a reduction in long-term unemployment (1999).
Well-being and health is an important determinant of income inequality according to Lynch (et al, 2000). He concludes that differences in health are related to income inequality. An increase in well-being and health, by giving everyone the same level of healthcare for example, will decrease differences in income inequality.
2.5.3 Conclusion
To summarize, certain variables are essential as explanatory variables of income inequality and cannot be disregarded in an analysis. Important variables are GDP, FDI inflow and outflow, trade openness and financial globalization. The effect of social and political globalization should be taken in account when researching overall globalization and unemployment and health are determinants of income inequality.
2.6 Hypothesis
This chapter identified variables and definitions that are important when researching the effect of globalization on income inequality. An increase in trade openness is expected to decrease income inequality, while FDI outflow is expected to worsen income
15 inequality by driving Gini upward. But there are contrasting results, an increase in globalization can both worsen and better income distribution. Technology is an important accelerator for globalization. The common used variable for income inequality is the Gini coefficient and an often used variable for globalization is the KOF-index. There are three ways globalization occurs, through politics, economy and sociocultural. But there are certain conditions that cause income inequality:
unemployment, an unequalising tax system and unequal healthcare services. To find a clear relation between globalization and income inequality, our hypothesis is the following:
A change in globalization affects income inequality in The Netherlands. Or in terms of variables: a change in the KOF-index affects the Gini coefficient in The Netherlands.
16
3 Data
This chapter presents the dependent and independent variables used in this paper.
3.1 Dependent variable
The Gini coefficient is used as the dependent variable, because it is the common measure for income inequality in the literature. The Gini coefficient lies between zero and one which means that Gini is bounded, where a coefficient of one indicates total inequality and zero complete equality.
We constructed the Gini coefficient for the period starting from 1989 until 2013. The Amsterdam Centre for Inequality Studies (AMCIS) Gini database contributes the most to our Gini set, from 1989 to 2010. The last three years, 2011 until 2013 is added by using the Gini from an OECD dataset. Both data sets are constructed in a similar way, as can be concluded from the overlapping years.
Gini is tested for a unit root problem, which can cause statistical interference to a time series variable. The Dicky-Fuller test for unit roots the null-hypothesis is rejected. Meaning no unit root is assumed (See appendix Unit root).
While income inequality does not radically change year by year, but rather evolves over time, it is needed to check for autocorrelation. While lags can be highly correlated with variables at time t. In the case of Gini there is autocorrelation with the first lag. To solve this problem, the first lag of Gini is added to the model (See appendix Autocorrelation).
3.2 Independent variables
The Gini coefficient is regressed on multiple explanatory variables. These variables represent a wide range of effects. But due to very high correlation (See
appendix complete correlation) between other explanatory variables and the KOF-index and the small number of observations, some had to be discarded, see figure 5 for the remained explanatory variables.
The first and most important independent explanatory is the KOF-index, which is a proxy for globalization. As discussed in section 2.2, the KOF-index is an overall index encompassing a vast variety of variables. In the original dataset of Dreher, the KOF-index is a number between 0 and 100 (2008). In this thesis the index is rescaled to a number between 0 and 1, which allows for an easier interpretation of the estimated coefficient.
The second explanatory variable is a dynamic term of Gini, lag Gini. This is included in the model to compensate for the autocorrelation, discussed in the previous section.
17 Unemployment (Ut), from CBS Statline, is the third independent variable.
Because an increase in unemployment results in an increase in income inequality (Mocan, 1999).
Log of life expectancy (Worldbank) is added as a proxy for health and living standards. Literature suggests that an increase in health or living standards decreases income inequality (Lynch, 2000). The natural logarithm is taken because life
expectancy in the period of 1989-2013 has an upward trend, which might indicate a fixed growth rate over time and the natural logarithm removes this growth rate.
The final explanatory variable added to the model is GDP growth (Worldbank). While literature suggests that GDP or GDP per capita is an important factor in income inequality (Pillai, 2011), we use GDP growth because nominal GDP and GDP per capita are highly correlated with the KOF-index and GDP has a unit root which can be
removed by taking the growth rate (See appendix, complete correlation).
The literature defines the level of democracy as an explanatory variable for income inequality. But this paper assumes that the level of democracy has not changed in the period 1989-2013, such that it is absorbed in the constant.
Figure 5: correlation table
GINI NewKOF LagGini U GDPGrowth LNLifeExpectancy
GINI 1 NewKOF 0.6627 1 LagGini 0.5944 0.7185 1 U -0.478 -0.5644 -0.3494 1 GDPGrowth -0.3618 -0.3047 -0.5824 -0.0053 LNLifeExpectancy 0.7206 0.7515 0.7765 -0.2336 -0.5359 1
18
4. Estimation Method and methodology
This chapter will use the data and construct an estimation method and the methodology. A time series analysis is conducted on The Netherlands for the period 1989-2013.
4.1 Estimation method
The following models are estimated:
𝐺𝐼𝑁𝐼𝑡 = β0+ β1𝑁𝑒𝑤𝐾𝑜𝑓𝑡+ β2𝐿𝑎𝑔𝐺𝑖𝑛𝑖𝑡+ εt (1) 𝐺𝐼𝑁𝐼𝑡 = β0+ β1𝑁𝑒𝑤𝐾𝑜𝑓𝑡+ β2𝐿𝑎𝑔𝐺𝑖𝑛𝑖𝑡+ β3𝑈𝑡+ εt (2) 𝐺𝐼𝑁𝐼𝑡 = β0+ β1𝑁𝑒𝑤𝐾𝑜𝑓𝑡+ β2𝐿𝑎𝑔𝐺𝑖𝑛𝑖𝑡+ β3𝑈𝑡 + β4𝐿𝑛𝐿𝑖𝑓𝑒𝐸𝑥𝑝𝑒𝑐𝑡𝑎𝑛𝑐𝑦𝑡+ εt (3) 𝐺𝐼𝑁𝐼𝑡 = β0+ β1𝑁𝑒𝑤𝐾𝑜𝑓𝑡+ β2𝐿𝑎𝑔𝐺𝑖𝑛𝑖𝑡+ β3𝑈𝑡 + β4𝐿𝑛𝐿𝑖𝑓𝑒𝐸𝑥𝑝𝑒𝑐𝑡𝑎𝑛𝑐𝑦𝑡+ β5𝐺𝐷𝑃𝑔𝑟𝑜𝑤𝑡ℎ + εt (4)
Where subscript t denotes the time in years. For the meaning of the variables see figure 6.
Figure 6
GINIt Gini coefficient for The Netherlands at time t.
NewKOFt KOF-index for The Netherlands divided by a hundred at time t. Ut Total unemployment as percentage of the working force in The
Netherlands at time t.
GDPgrowtht GDP growth in The Netherlands at time t.
LnLifeExpectancyt Natural logarithm of life expectancy at birth for The Netherlands at time t.
LagGINIt Dynamic term of Gini, GINI t-1 εt Normally distributed error term
19
4.2 Methodology
The empirical analysis will be done by using a beta regression and a OLS-regression. Both methods will be examined on their applicability on regressing the model stated in above in this chapter.
4.2.1 Beta regression
Cribari-Neto discussed beta regression, named after the beta distribution (2009). The motivation to use a beta regression lies in the flexibility of the model, because it uses the beta law. This law allows the beta distribution to assume a number of different shapes, depending on the combination of the parameter values, including right- and left-skewed or the flat shape of the uniform density. The beta regression allows for
nonlinearities and variable dispersion. Moreover, the regression parameters are
interpretable in terms of the mean of the dependent variable and the model is naturally heteroskedastic and easily accommodates asymmetries.
The original intent of the beta regression is to model random binominal variables (Cribari-Neto, 2009). But a recent development in beta regression has lead researchers to regress on continuous random variables which are bound by zero and one, such as: rates, proportions and indices – such as the Gini index for inequality. The beta
regression is interpreted by looking at the margins, where a one percent increase in an independent coefficient predicts a coefficient percent increase in the dependent variable (Stata, 2016).
Figure 7 shows the beta distribution. The parameter ϕ is known as the precision parameter since, for fixed µ, the larger ϕ the smaller the variance of y and ϕ-1 is a dispersion parameter. In the beta distribution µ is bound by zero and one. General version of the model:
𝐿𝑒𝑡 𝑦 = (𝑦1, … , 𝑦𝑛)𝑇𝑏𝑒 𝑎 𝑟𝑎𝑛𝑑𝑜𝑚 𝑠𝑎𝑚𝑝𝑙𝑒, 𝑤ℎ𝑒𝑟𝑒 𝑦
𝑖~ β(µ𝑖, 𝜙𝑖), 𝑖 = 1, … , 𝑛
Suppose the mean and the precision parameter of yi satisfies the following functional relation:
𝑔1(µ𝑖) = η1i= 𝑓1(𝑥𝑖𝑇; 𝛽) 𝑔1(ϕ𝑖) = η2i= 𝑓2(𝑧𝑖𝑇; θ)
Where β = (β1, … , βk )T and θ = (θ1, … , θh)T are vectors of unknown regression parameters which are assumed to be functionally independent, η1i and η2i are
predictors, 𝑥𝑖𝑇 and 𝑧𝑖𝑇are observations on q1 and q2 known covariates. We assume that the derivative matrices X = ∂η1/∂β and Z = ∂η1/∂β have rank k and h, respectively.
The link functions g1: (0, 1) → ℝ and g2: (0, ∞) → ℝ are strictly monotonic and twice differentiable. A number of different link functions can be used to adjust for the
20 right distribution function. For example the logit specification g1 (µ) = log{µ/(1 − µ)} and the probit function g1(µ) = Φ−1 (µ), where Φ(·) denotes the standard normal distribution function. Beta regression allows for flexible modelling of each of the parameters that index the distribution using parametric terms involving nonlinear or linear predictors, smooth nonparametric terms and random effects, due to the
Generalized Additive Models for Location Scale and Shape (GAMLSS) framework by Rigby and Stasinopoulos (2007). To estimate maximum likelihood is approached through a Newton-Raphson scoring algorithm with the backfitting algorithm for addictive components (Cribari-Neto, 2009).
Figure 7: Beta distribution
Description: Probability density functions for beta distributions with varying parameters µ = 0.10, 0.25, 0.50, 0.75, 0.90 and ϕ = 5 (left) and ϕ = 100 (right). Source: Cribari-Neto (2009)
4.2.2 Ordinary least-squares regression
Our second method will make use of ordinary least-squares. A common used method of regressing. A time series regression with multiple predictors has three assumptions (Stock, 2012):
1. Linearity
2. No large outliers 3. εit ~ N(0, σ 2 ) :
i. Normality of the error term
ii. Conditional mean of the error term is zero. iii. Homoscedasticity
The first assumption is that the relationship between the Gini coefficient and the explanatory variables is linear. If this assumption does not hold, non-linear data is fitted into a linear relation, which can cause wrong estimates. Scatter plots of the residuals
21 against explanatory variables show no clear linear pattern (See appendix Linearity). The linearity assumption does not hold.
Figure 8 shows a plot of the residuals to the fitted values. It shows that variance appears to be roughly constant and there are no outliers. The second assumption of no large outliers holds.
The third assumption involves the error term. To check for a normal distribution figure 9 plots the residuals against a normal distribution. Residuals are fairly normally distributed and since there are no outliers, normality of the error is assumed. Regarding the conditional mean of the error term, which is expected to be zero, figure 8 shows no clear skewed patron. There is reason to expect that this assumption holds. The last assumption regarding the error term is homoscedasticity. This assumes that the variance of the error term is constant. To test this a Breusch-Pagan/Cook-Weisberg test is done. All four models were tested and the null-hypothesis is rejected, meaning that
heteroscedasticity must be taken in account, therefore robust standard errors are assumed (See appendix hettests).
Figure 8: Residuals Figure 9: Normality
0 20 40 60 80 1 0 0 D e n sit y -.01 -.005 0 .005 .01 Residuals -. 0 1 -. 0 0 5 0 .0 0 5 .0 1 R e si d u a ls .265 .27 .275 .28 .285 Fitted values
22
5. Results
This chapter will present the results of the empirical analysis. First the beta regression will be discussed. Then the OLS regression results will be given. In both models 24 observations are used since the dynamic term of Gini, LagGini leads to a drop of the first observation.
5.1 Beta regression
The first model includes LagGini and NewKOF, see figure 10. Both the KOF-index and Lag Gini are positive but insignificant. The model might suffer from an omitted variable bias, so we add additional explanatory variables. In the second model U, unemployment is included. But again no explanatory variable is significant at a 10% or 5% significant margin.
The natural logarithm of life expectancy, LNLifeExpectancy is added to the third model. Resulting in two 1% significant independent variables: U and
LNLifeExpectancy. Since the sign of U is negative unemployment has its negative relation with income equality. Which is in disagreement with Mocan (1999), who expects unemployment to have positive relation income inequality. A possible reason for this odd result is the strong social security net in The Netherlands, where minimum wage is replaced by social benefits, which can increase an individual’s level of income. But this effect be should investigated further to determine the true cause.
Surprisingly life expectancy at birth has a positive effect on income inequality. This is in disagreement with the literature (Lynch, 2000). A possible reason for this relation is although life expectancy is a proxy for health and well-being. The pensionable age in The Netherlands is 65 years, in our data, and most people do experience a drop in income as they become elderly. But this effect should be further investigated as well.
Lastly GDPgrowth was added in the fourth model. While GDP growth itself does not impact Gini significantly, unemployment and life expectancy still have their significant relation with the Gini coefficient.
Figure 11 and 12 show the marginal effect of the variables in the third and fourth model. NewKOF is not significant so no statements can be made regarding marginal effect. But a 1% increase in unemployment will decrease Gini by 0.009% and improve equality. A 1% increase in the natural logarithm of life expectancy is expected to increase Gini by 0.01% in the fourth model.
To conclude. Beta regression does not yield satisfying results. Unemployment and life expectancy were strongly significant, but have a complete opposite relation with income inequality than expected from literature. A possible cause for these odd results is the small number of observations.
23
Figure 10: Beta regression
Figure 11 Margins of beta regression third model
Delta-method dy/ex Standard Error z P>Z [95% Confidence Interval] NewKOF -0.00178 0.04674 -0.04 0.97 -0.0933862 0.08983 U -0.00895 0.003351 -2.67 0.008 -0.0155148 -0.00238 LNLifeExpectancy 1.046875 0.269832 3.88 0 0.5180145 1.575736 LagGini -0.01166 0.062044 -0.19 0.851 -0.1332653 0.109942
Figure 12 Margins of beta regression fourth model
Delta-method dy/ex Standard Error Z P>Z [95%Confidence Interval] NewKOF 0.001175 0.045341 0.03 0.979 -0.08769 0.090043 GDPGrowth -0.00034 0.000928 -0.36 0.717 -0.00215 0.001482 U -0.0091 0.003356 -2.71 0.007 -0.01567 -0.00252 LNLifeExpectancy 1.020399 0.290542 3.51 0 0.450947 1.589851 LagGini -0.01821 0.05837 -0.31 0.755 -0.13261 0.096193
24
5.2 OLS regression
The OLS regression has the same setup as the beta regression, see figure 13, so interpretation and comparison of results is easier.
In the first model two explanatory variables are included: NewKOF and
LagGini. Yielding the same results as the beta regression. Both variables are positively related with income inequality but not significant. The R2 of this model equals 0.47. The R2 indicates how much of the variance in the dependent variable is clarified by the variance in the explanatory variables. A high R2 is generally an indication that a model improves.
Unemployment is added to the second model but does not result in any significant explanatory variables. Whereas adding the natural logarithm of life
expectancy does yield significant independent variables. Unemployment is negatively related with Gini at a 5% significant level, so an increase in unemployment causes a decrease in Gini and an equalizing effect on income levels. Life expectancy is significant at a 1% level and is positively related to Gini. As a result of adding life expectancy the R-squared has increased to 0.62. In the fourth model GDPgrowth is added. While GDP growth itself is not significant, U and LNLifeExpectancy still are significant as in the third model.
The effect of explanatory variables on Gini is best described by looking at marginal effects. Figure 13 shows that a unit increase in unemployment is associated with a change in Gini of -0.002. An increase of 1% in the natural logarithm will increase Gini by 0.0024.
To conclude, OLS regression yields comparable results as the beta regression described above. Which is unsatisfying while the results are in contrast with literature.
25
6. Conclusions
This paper investigates the relation between globalization and income inequality. The theoretical framework of the model is based on the literature, where
important variables for income inequality and globalization are identified such as trade openness, financial, social and political globalization and common research methods.
As a proxy for income inequality the Gini coefficient is used. After discarding several independent variables to prevent multicollinearity, five explanatory variables remain: KOF-index, unemployment, GDP growth, life expectancy and lag Gini.
We use two different estimation methods, beta regression and OLS regression. Beta regression is designed to regress indices or ratios which are bound by zero and one. Beta regression can adept its parameters to ensure it fits the estimated model. OLS regression is the most standard estimation method and used as a robustness check.
Our most important conclusion, regarding the hypothesis, is that the KOF-index does not influence the Gini coefficient in all estimated models. According to these results globalization does not affect income inequality in The Netherlands.
From the remaining four explanatory variables only life expectancy and
unemployment are significant at a 5% level. In contrast with the literature, we conclude that an increase in unemployment causes income to be more evenly spread and an increase in life expectancy, a proxy for living standards and health, increases income inequality. A possible reason for these odd results may be the small number of
observation. A suggestion for further research would be to analyse a dataset with more observations for The Netherlands. Furthermore, it will be interesting to look at more countries using a panel data set-up.
26
References
Articles
Bechtel, G.G. (2014) Does Globalization Mitigate Income Inequality? Journal of data science, 12, 197-215.
Bhugwati, J. & Srinivasan, T.N. (1999) Trade and poverty in poor countries. The American economic review, 90(2), 180-183.
Borraz, F. & Lopez, J.E. (2007) Has Globalization Deepened Income Inequality in Mexico? Global Economy Journal, 7(1), 1-55.
Çelik, S. and Basdas U. (2010) How Does Globalization Affect Income Inequality? A Panal Data Analysis. International Advances in Economic Research, 16, 358-370.
Cozma, R. & Chen, K. (2013) What’s in a tweet? Journalism Practice, 7(1), 33-46. Cribari-Neto, F. & Achim, Z. (2009) Beta Regression in R. Research Report Series /
Department of Statistics and Mathematics WU Vienna University of Economics and Business, 98.
Dollar, D. & Kraay, A. (2004) Trade, Growth and Poverty. The economic journal, 114
(493). F22-F29.
Dreher, A. and Gaston, N. (2008) Has globalization increased inequality? Review of International Economics, 16(3). 516–536.
Jaumotte, F., Lall, S. & Papageorgiou C. (2013) Rising Income Inequality: Technology, or Trade Financial globalisation? IMF Economic Review. 61(2). 271–309. Lynch, J.W., Smith, G.D., Kaplan, G.A. & House, J.S. (2000) Income inequality and
mortality: Importance to health of individual income, psychosocial environment, or material conditions. British Medical Journal. 320. 1200-4.
Mocan, H.N. Structural Unemployment, Cyclical Unemployment, and Income Inequality. The Review of Economics and Statistics. 81(1). 122-134.
O'Rourke, K. H. (2001) Globalization and Inequality: Historical Trends. NBER working paper 8339.
Pillai, A. (2011) Impact of Political Regime and Economic Openness on Income inequality: A Tale of Low-Income and OECD Countries. The IUP Journal of Financial Economics, 9(4). 60-75.
Rigby, A.R. & Stasinopoulos, D.M. (2007) Generalized Additive Models for Location Scale and Shape (GAMLSS) in R. Journal of statistical software, 23(7). Taylor, J.B. (2009) The Financial Crisis And The Policy Responses And The Policy
Responses: An Empirical Analysis Of What Went Wrong. NBER working paper 14631.
Books
Beck, U., Sznaider, N., & Winter, R. (2003) Global America?: The Cultural Consequences of Globalization. Liverpool: Liverpool University Press. Brawley, M. R. (2008) The politics of globalization: gaining perspective, assessing
consequences. Canada: Broadview press.
Chomsky, N. (2010) Hopes and Prospects. London: Penguin Group.
Lange De, T. (2007) Staat, markt en migrant: De regulering van arbeidsmigratie naar Nederland 1945-2006. Den Haag, Boom Juridische uitgevers.
Melchior, A., Telle, K. and Wiig, H. (2000) Globalisation and Inequality: World Income Distribution and Living Standards. 1960-1998. Studies on Foreign Policy Issues. Oslo: Royal Norwegian Ministry of Foreign Affairs.
27 Salverda, W. (2014) Hoe ongelijk is Nederland? Een verkenning van de ontwikkeling en
gevolgen van economische ongelijkheid. Amsterdam: Amsterdam University Press.
Stock, J. H. & Watson, M. W. (2012) Introduction to Econometrics. Third edition. International edition. Boston, Pearson.
Websites
Belastingdienst. (2015) Inkomstenbelasting. [Online] Available from:
http://www.belastingdienst.nl/wps/wcm/connect/bldcontentnl/belastingdienst/pri ve/inkomstenbelasting/. [Assessed 21st of April 2015].
Europa. (2015) De geschiedenis van de Europese Unie. [Online] Avialable from:
http://europa.eu/about-eu/eu-history/index_nl.htm. [Assessed 21st of April 2015]. KOF Swiss Economic Institute. (2015) Kof as a mediator. [Online] Available from:
https://www.kof.ethz.ch/en/about-us/. [Assessed 20th of April 2015]. OECD. (2015) Inequality. [Online] Available from: http://www.oecd.org/social/
inequality.htm. [Assessed 20th of april 2015].
Rijksoverheid. (2015) TTIP: handelsovereenkomst tussen Europese Unie en de Verenigde Staten. [Online] Available from: http://www.rijksoverheid.nl/ onderwerpen/ttip-handelsovereenkomst-tussen-europese-unie-en-de-verenigde -staten. [Assessed 21st of April 2015].
Stata. (2016) Betareg – Betaregression. [Online] Available from: http://www.stata.com /manuals14/rbetareg.pdf [Assessed 14th of June 2016].
UNESCO. (2016) Globalisation. [Online] Available from: http://www.unesco.org /new/en/social-and-human-sciences/themes/international-migration /glossary/globalisation. [Assessed 26th of June 2016].
Vluchtelingenwerk (2016) Bescherming in Nederland. [Online] Available from: http://www.vluchtelingenwerk.nl/feiten-cijfers/cijfers/bescherming-nederland [Assessed 17th of June 2016].
Wikipedia (2016) A typical Lorenz curve. [Online] Available from:
https://en.wikipedia.org/wiki/Lorenz_curve#/media/File:Economics _Gini_coefficient2.svg [Assessed 29th of June 2016].
28
Appendix
KOF variables
29
Unit root
DF-GLS test for GINI indicates that the optimal number of lags, while testing for unit roots is 1 lag (SC & MAIC) or 0 lags (Ng-Perron criterion).
Applying this to the Dicky-Fuller test, using a drift term for Gini gives:
*P-Value = .0012<.05. Which rejects the unit root null-hypothesis.
Min MAIC = -8.965057 at lag 1 with RMSE .0049426 Min SC = -10.27317 at lag 1 with RMSE .0049426 Opt Lag (Ng-Perron seq t) = 0 [use maxlag(0)]
1 -2.342 -3.770 -3.509 -3.100 2 -1.798 -3.770 -3.349 -2.949 3 -1.502 -3.770 -3.178 -2.781 4 -1.517 -3.770 -3.033 -2.625 5 -1.701 -3.770 -2.950 -2.512 6 -1.012 -3.770 -2.965 -2.471 7 -1.266 -3.770 -3.114 -2.532 8 -0.811 -3.770 -3.432 -2.725 [lags] Test Statistic Value Value Value DF-GLS tau 1% Critical 5% Critical 10% Critical Maxlag = 8 chosen by Schwert criterion
DF-GLS for GINI Number of obs = 16
_cons .1380128 .0399952 3.45 0.002 .0550678 .2209578 L1. -.4972506 .1450072 -3.43 0.002 -.797977 -.1965241 GINI D.GINI Coef. Std. Err. t P>|t| [95% Conf. Interval] p-value for Z(t) = 0.0012
Z(t) -3.429 -2.508 -1.717 -1.321 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Z(t) has t-distribution Dickey-Fuller test for unit root Number of obs = 24 . dfuller GINI, drift regress lags(0)
30 *P-Value = .0498<.05. Which rejects the unit root null-hypothesis.
Autocorrelation
Looking at the correlogram above indicating autocorrelation between Gini and its lags, it is clear that the first lag is outside the 95% confidence band which indicates that there is autocorrelation between Gini and Gini t-1.
_cons .0831111 .0478194 1.74 0.098 -.0166385 .1828607 LD. -.2331108 .174967 -1.33 0.198 -.5980856 .1318639 L1. -.2988097 .1730693 -1.73 0.100 -.6598258 .0622065 GINI D.GINI Coef. Std. Err. t P>|t| [95% Conf. Interval] p-value for Z(t) = 0.0498
Z(t) -1.727 -2.528 -1.725 -1.325 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Z(t) has t-distribution Augmented Dickey-Fuller test for unit root Number of obs = 23 . dfuller GINI, drift regress lags(1)
31 Complete correlation L N G D P p e r C a ~ a 0 . 7 7 5 4 0 . 9 1 0 2 0 . 8 0 6 6 -0 . 4 6 5 1 -0 . 4 5 7 6 0 . 9 3 5 0 0 . 9 9 9 9 0 . 9 8 8 6 1 . 0 0 0 0 L N P O P 0 . 7 3 4 3 0 . 9 0 4 2 0 . 8 0 3 5 -0 . 3 6 0 3 -0 . 5 0 5 1 0 . 9 4 4 3 0 . 9 9 1 1 1 . 0 0 0 0 L N G D P 0 . 7 7 1 6 0 . 9 1 0 6 0 . 8 0 7 1 -0 . 4 5 3 7 -0 . 4 6 3 6 0 . 9 3 7 1 1 . 0 0 0 0 L N L i f e E x p e ~ y 0 . 7 2 0 6 0 . 7 5 1 5 0 . 7 7 6 5 -0 . 2 3 3 6 -0 . 5 3 5 9 1 . 0 0 0 0 G D P G r o w t h -0 . 3 6 1 8 -0 . 3 0 4 7 -0 . 5 8 2 4 -0 . 0 0 5 3 1 . 0 0 0 0 U -0 . 4 7 8 0 -0 . 5 6 4 4 -0 . 3 4 9 4 1 . 0 0 0 0 L a g G i n i 0 . 5 9 4 4 0 . 7 1 8 5 1 . 0 0 0 0 N e w K O F 0 . 6 6 2 7 1 . 0 0 0 0 G I N I 1 . 0 0 0 0 G I N I N e w K O F L a g G i n i U G D P G r o ~ h L N L i f e ~ y L N G D P L N P O P L N G D P p ~ a ( o b s = 2 4 ) . c o r r G I N I N e w K O F L a g G i n i U G D P G r o w t h L N L i f e E x p e c t a n c y L N G D P L N P O P L N G D P p e r C a p i t a
32
Hettests
Prob > chi2 = 0.0143 chi2(2) = 8.50 Variables: NewKof LagGini Ho: Constant variance
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . hettest NewKof LagGini
Prob > chi2 = 0.0154 chi2(3) = 10.40 Variables: NewKof LagGini U Ho: Constant variance
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . hettest NewKof LagGini U
Prob > chi2 = 0.0091 chi2(4) = 13.48
Variables: NewKof LagGini U LNLifeExpectancy Ho: Constant variance
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . hettest NewKof LagGini U LNLifeExpectancy
Prob > chi2 = 0.0105 chi2(5) = 14.96 GDPGrowth
Variables: NewKof LagGini U LNLifeExpectancy Ho: Constant variance
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . hettest NewKof LagGini U LNLifeExpectancy GDPGrowth
33 Linearity .2 5 .2 6 .2 7 .2 8 .2 9 .82 .84 .86 .88 .9 .92 NewKOF
GINI Fitted values
lowess GINI NewKOF
.2 5 .2 6 .2 7 .2 8 .2 9 .25 .26 .27 .28 .29 LagGini
GINI Fitted values
34 .2 5 .2 6 .2 7 .2 8 .2 9 4 6 8 U
GINI Fitted values
lowess GINI U .2 5 .2 6 .2 7 .2 8 .2 9 -4 -2 0 2 4 6 GDP-Growth
GINI Fitted values
35 .2 5 .2 6 .2 7 .2 8 .2 9 4.34 4.36 4.38 4.4 LNLifeExpectancy
GINI Fitted values
