The effect of
foreign direct investment
on income inequality in
developing countries
Abstract
This paper investigates the extent to which inward FDI stock has a positive effect on income inequality in 15 developing, mainly African, countries. A panel data analysis is conducted on both annual and three-years averaged observations, in the period 1980-2004. The Gini coefficient is explained by inward FDI stock and a number of control variables. There are no robust findings of an effect of FDI, but the findings indicate that country specific factors – the value added by the agricultural sector, secondary school enrollment and openness to trade – do have a significant effect on income inequality.
BSc Thesis Economics
Charlotte Cator
February 2016
10610561
1
Statement of Originality
This document is written by Student Charlotte Cator 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
Introduction ... 3
1 Literature Review ... 4
1.1 Foreign direct investment ... 5
1.2 Gini-coefficient ... 5
1.3 Theoretical insights ... 5
1.3.1 Negative effect ... 6
1.3.2 Positive effect ... 7
1.3.3 No effect ... 8
1.4 Earlier empirical results ... 9
1.4.1 Negative effect ... 9 1.4.2 Positive effect ... 9 1.4.3 No effect ... 10 1.5. Hypotheses ... 10 2 Empirical analysis ... 11 2.1 Methodology ... 11 2.1.1 Dependent variable ... 11 2.1.2 Independent variables ... 12 2.2 Data ... 15 3 Results ... 18 3.1 Annual data ... 18 3.2 Averaged data ... 20 3.3 Assumptions ... 22 3.3.1 OLS assumptions ... 22
3.3.2 Fixed effects or random effects ... 24
3.4 Sample size and regional differences ... 26
4 Conclusion ... 27 References ... 29 Appendix 1 ... 31 Appendix 2 ... 32 Appendix 3 ... 33 Appendix 4 ... 34 Appendix 5 ... 35
3
Introduction
The last decades, the amount of worldwide foreign direct investment (FDI) has increased notably. Figure 1 shows the development of worldwide FDI flows since 1970. From 2000, the worldwide flows and the flows toward developed economies have become more volatile. This is not the case in developing countries, where the FDI inflows have been increasing steadily. According to the World Investment Report of 2015, in 2014, five of the ten countries that were the highest recipients of FDI were developing countries (UNCTAD, 2015, p. 9).
Source: UNCTAD
Foreign direct investment (FDI) is the process whereby a multinational directly acquires a foreign subsidiary to establish a long term interest (Moosa, 2002). There are several effects that FDI can have on the host country’s economy. Alfaro et al. (2004) state that FDI may benefit the host country through labor and technology spillovers, which lead to modernization and growth in the economy and thus benefit the host economy. Because of these presumed spillovers, policy makers in a large number of countries have attempted to attract more FDI flows the last decades (Wang & Wong, 2009). However, FDI can also have a negative effect on the host country. One of the concerns is that FDI may lead to worse working conditions in the subsidiaries (Brown et al., 2003). Another concern is that it may lead to higher income inequality in the recipient country (Sylwester, 2005; Tsai, 1995). If this is true, the decision to attract FDI might be affected.
4
Elaborate studies have been conducted on different (panels of) both developing and developed countries, to investigate the relationship between FDI and income inequality. The results are diverse. This might be due to different approaches, but also to missing data and inconsistent measures of income inequality, especially for developing countries. This paper investigates the relationship between FDI and income inequality empirically. It addresses the question as to what extent inward FDI has a positive effect on income inequality in developing countries.
To measure the effect of FDI on income inequality, the Gini coefficient is regressed on the stock of inward FDI in a country and other assumed determinants of income inequality. To avoid a bias that might arise from inconsistent income inequality measures, the Gini coefficient from the Standardized World Income Inequality Database is used. Following some of the influential earlier studies on this matter, a panel data analysis is conducted. The sample consists of 15 developing countries, of which most are African, during the period 1980-2004. The first part of the analysis focuses on the data on annual basis. In the second part of the analysis, the dataset is converted to three-year averages to eliminate short-term fluctuations.
The empirical analysis shows that there is no significant effect of inward FDI on income inequality. In the annual dataset, the effect is negative in all estimated models. In the averaged dataset, the effect is negative in all models except for the ones where a time variable is added. The findings indicate that the country specific factors – the value added by the agricultural sector, secondary school enrollment and openness to trade – and the time variable do have a significant effect.
The paper is organized as follows. The first section provides a literature review. Then, the methodology, data and hypothesis are presented in section 2. Section three lays out the empirical results and the fourth section concludes.
1
Literature Review
The following section provides an overview of the existing literature. Many studies have been conducted on the relationship between FDI and income inequality. This section lays out the different approaches that have been used and different conclusions that have been drawn. First, several concepts central to this paper are explained. Then, the insights that economic theory provides on the relationship are laid out. The section concludes with a review of earlier empirical research.
5
1.1
Foreign direct investment
Moosa (2002) defines foreign direct investment as the investment method whereby a multinational enterprise located in one country acquires ownership of assets of a firm in another country. The purpose of this acquisition is to control the production, the distribution and other activities, on a long term basis. This aim for a long term relationship distinguishes FDI from portfolio investment. According to Moosa, the consensus is that at least 10 percent of the total asset stock should be purchased to obtain a controlling interest.
Inward FDI is considered as the process whereby a foreign multinational invests in the home country. This implies that it is accompanied by FDI inflows from the perspective of the home country. Outward FDI on the other hand is considered as the process whereby a domestic multinational invests in a foreign country, thus generating FDI outflows from the perspective of the home country.
There are two types of FDI to be distinguished. Brownfield FDI on the one hand, is the type of FDI that is described by Moosa. Greenfield FDI on the other hand, involves the installation of a new affiliate in a foreign country by one or more multinational enterprises (Cheng, 2006). In this paper, both types of FDI are considered to be relevant and no distinction is made.
1.2
Gini-coefficient
The Gini coefficient is a widely accepted measure of income inequality, whose measurement is based on the Lorenz Curve. Gastwirth defines the Lorenz Curve as the curve that “plots the percentage of total income earned by various portions of the population when the population is ordered by the size of their incomes” (1971, p. 1037). The Gini coefficient consists of the value of the area between the Lorenz Curve and a 45-degrees line that indicates absolute equality is measured, divided by the area under the 45-degrees line (Gastwirth, 1972). A coefficient of 0 indicates perfect income equality within a country, whereas a coefficient of 100 indicates perfect income inequality.
1.3
Theoretical insights
Economic theory is inconclusive about the effect of FDI on income inequality. There are two main aspects to this effect: the demand for labor and the distribution of the spillover gains from FDI. In economic theory, a distinction is made between owners of capitalists and workers. The owners of capitalists get income from the return on capital, which is equal to the marginal
6
product of capital, whereas laborers get income from hourly wage, which is equal to the marginal product of labor. Another distinction is made between high-skilled laborers and low-skilled laborers. High-low-skilled laborers earn a higher hourly wage than low-low-skilled laborers, based on their characteristics (Martins & Pereira, 2004).
To examine the different aspects of FDI and income inequality, various models are presented. The following section lays out how these models predict either a negative, a positive or no effect.
1.3.1 Negative effect
FDI flows to developing countries can have a negative effect on income inequality in different ways. Brown, Deardorff and Stern (2003) consider the first mechanism by which inward FDI can lead to less income inequality. The mechanism assumes that income inequality is the result of differences between return to capital and return to labor, and that the former is higher than the latter. The simplest model that explains this mechanism, is a model in which all countries produce the same good with both capital input and labor input. If a country faces inward FDI, there is an inflow of capital which leads to an increase in domestic output. Due to the increased capital stock, the marginal product of labor increases, which implies an increase in the wage. In contrast, the marginal product of capital decreases because of the increased capital stock, which implies a decrease in the return on capital. As labor income increases and capital income deceases, income inequality decreases.
Another mechanism that can explain a negative effect of inward FDI on income inequality concerns an increase in the demand for unskilled labor. If a developing country faces inward FDI, this will affect the labor demand in that country. This can be an increase in the relative demand for either skilled or unskilled labor. Ucal et al. (2014) state that the Heckscher-Ohlin model predicts a negative effect of FDI on income inequality, by generating an increase in the relative demand for unskilled labor. The Heckscher-Ohlin model explains trade on the basis of the differences in production factor endowments. It also examines the redistribution of welfare that is the result of trade. It assumes two countries that produce two goods, and that the countries open up to trade at a given moment in time. The results of opening up to trade are stated by the Stolper-Samuelson theorem, which is based on the Heckscher-Ohlin model. It states that trade, and thus FDI, exploits the relatively abundant production factor in the host country. It thereby benefits the owners of the abundant production factor, while it hurts the owners of the scarce production factor. Ucal et al. (2014) state that, if developed countries are abundant in skilled labor and developing countries are abundant in unskilled labor, FDI in
7
developing countries would lead to higher demand for unskilled labor in the host country. This means that inward FDI leads to lower income inequality.
A negative effect of FDI on income inequality is also presented by the Kuznets curve. According to Kuznets, industrial development initially induces higher income inequality, up to a certain peak. After the development has reached this point, further development decreases inequality. This relationship is formalized by the inverted-U shaped Kuznets curve (Nielsen & Anderson, 1997). This is also the main insight of the modernization hypothesis, which states that sufficient output must first be generated before it can be distributed equally (Tsai, 1995). This implies that the short-term effect of FDI is positive, whereas the long-term effect is negative.
Finally, FDI can reduce income inequality by creating economic opportunities for those who would not have had these without FDI (Sylwester, 2005). These opportunities are generated by the spillover effects that FDI may have on the host country, which are the following. Firstly, there is the possibility that FDI improves the capital formation and increases employment in the host country. Secondly, FDI may stimulate exports of manufactured goods and generate technology spillovers. Furthermore, FDI may bring several resources into the recipient country, such as management expertise and established brand names (Zhang, 2006). The effect of these spillovers on income inequality depends on the distribution of the spillover benefits. If the lower income workers are able to capture (part of) the benefits, the income inequality may decline (Sylwester, 2005).
1.3.2 Positive effect
There are different ways in which FDI can have a positive effect on income inequality in the recipient country. The first possibility is through labor demand. Feenstra and Hanson introduced an adjusted Heckscher-Ohlin model in 1996 that indicates a rise in the demand for skilled labor as a result of inward FDI (Brown, Deardorff, & Stern, 2003; Ucal et al., 2014). This model can be used to explain what happens when there is an FDI flow from a developed country toward a developing country. It considers two countries – a developing country and a developed country – in which different goods are produced with capital, skilled and unskilled labor. The goods with the lowest skill-intensity are exclusively produced in the developing country. In case of a capital flow from the developed to the developing country, the production possibilities expand in the recipient country and contract in the donor country. In both countries, the average skill intensity increases, since the production in the developed country is more skill-intense than the production sectors in the developing country. To expand their production after the capital
8
inflow, the developing country starts to produce relatively skill-intensive products, which increases the demand for relatively skilled labor in the developing country. Hence, FDI flows to developing countries lead to an increase in the demand for skilled labor. Likewise, Herzer and Nunnenkamp (2013) state that FDI can lead to increased demand for skilled labor, since the production of multinationals is more skill-intensive than the production of domestic companies in a developing host country. Sylwester (2005) and Tsai (1995) state that an increase in the relative demand for skilled labor results in more income inequality, through higher wages for skilled labor and lower wages for unskilled labor.
Besides the effect of increased demand for skilled labor, there is the aspect of the spillover benefits from FDI that are described in section 1.3.1. If these benefits are mostly captured by workers that already earned a high income, the income inequality would increase (Sylwester, 2005). Basu and Guariglia (2007) formulate a model in which the benefits of foreign capital can only be captured by those who have sufficient human capital to operate the modern manufacturing process. Consequently, the poor can only benefit from foreign capital inflows if they are able to generate enough human capital and become entrepreneurs. The view that the poor might not have the ability to capture the benefits of FDI is also supported by the dependency hypothesis, which states that FDI leads to the formation of a “labor elite”. This elite consists of workers that are active in multinational firms, earning much higher wages than workers in the domestic sectors. Furthermore, because of their international employer, the workers in the multinationals might be offered better financial opportunities, such as investment and saving opportunities (Tsai, 1995).
1.3.3 No effect
Brown, Deardorff and Stern (2003) provide an explanation of how income inequality may be unaffected by inward FDI. They consider the Heckscher-Ohlin model with a factor price equalization mechanism. This mechanism implies that the factor prices across the countries will equalize after opening up to trade. In the case of a small country, the changes in output due to a capital inflow do not change world prices. As long as this country keeps producing both goods, there is no change in factor prices. For a large country, a capital inflow results in an increase in total domestic output. This alters world prices, and leads to a decrease in capital in the other country. If the latter produces both goods before and after the change, there is no change in factor prices. This means that there is no change in wages or return on capital effect. Hence, income inequality is unaffected.
9
1.4 Earlier empirical results
As section 1.3 pointed out, economic theory provides explanations for the mechanism of international trade (and FDI) and income inequality, but the explanations are not conclusive. In reality, the effect of FDI on income inequality may have more dimensions than the economic models include. Therefore, the empirical analyses that have been conducted earlier are considered. The empirical results are diverse. This section first considers studies that find a negative effect, then the ones that find a positive effect follow. The section concludes with an overview of studies that do not find a significant effect.
1.4.1 Negative effect
Several studies find a negative relationship between FDI and income inequality. Herzer and Nunnekamp (2013) conduct a study on eight European countries in the period from 1980 to 2000. They find a negative average effect of inward and outward FDI on income inequality in the long-run, although short-run effects appear to be positive. They also find large cross-country differences, with some European countries exhibiting positive long-run effects. Chintrakarn et al. (2012) conduct a study on the effect of inward FDI on income inequality in the US. For the US as a whole, they find a negative effect. But there is heterogeneity across states, with 21 of 48 states showing a positive relationship.
1.4.2 Positive effect
A number of studies find a positive effect of inward FDI on income inequality. Alderson and Nielsen (1999) analyze an unbalanced panel data set, consisting of 88 countries in the period from 1967 to 1994. They find a positive effect of FDI on income inequality. They also find that the balance of inward and outward FDI has a positive effect on income inequality. This leads to their conclusion that relative dependence on FDI induces higher income inequality. Tsai (1995) analyses the income distribution within less developed countries during the 1970s and 1980s. He finds a positive effect for South East Asian countries only and therefore addresses the importance of distinguishing different regions in models that analyze the effect of FDI. Basu and Guariglia (2007) conduct a panel data analysis for 119 developing countries over the period 1970-1999, observing that FDI promotes income inequality, particularly in a setting where low-skilled workers are unable to reap benefits of the modern FDI-based technology. Choi (2006) analyses the effect on income inequality of inward, outward and total FDI stock for 119 countries during the period 1993-2002. He finds a positive and significant effect of FDI on
10
income inequality, which is stronger for outward FDI than for inward FDI. Figini and Görg (2011) conduct an analysis for 100 countries (both OECD and non-OECD) over the period 1980-2002. For the developing countries, they find that FDI increases income inequality, but the effect is diminishing as FDI stocks increase further. For developed countries, however, they find that FDI decreases income inequality. Figini and Görg suggest that the reason for the different results is that the effect of the technology spillovers on the economy depends on the level of development in the host country. Herzer, Hühne and Nunnenkamp (2014) find a positive effect of inward FDI on income inequality in Latin-American countries, during the period from 1980 to 2000. Reuveny and Li (2003) conduct a study on the relationship between economic openness, democracy and income inequality for 69 countries over the period 1960-1996. One of their findings is that FDI promotes income inequality, while trade reduces income inequality. Driffield and Taylor (2000) find that FDI increases demand for skilled labor in the United Kingdom, leading to higher income inequality.
1.4.3 No effect
There are studies that do not find a significant effect of FDI on income inequality. Sylwester (2005) conducts a study on a cross section of developing countries between 1970 and 1989. He finds that FDI is positively related to economic growth, but that there is no significant effect of FDI on income inequality. Te Velde and Morrissey (2004) find mixed results among their data set of five East Asian countries, no statistical effect of FDI on income inequality is found. However, when controlling for domestic factors, they find that FDI reduced income inequality in Thailand during the period of 1985-1998. Ucal et al. (2014) analyze FDI and income inequality in Turkey during a period from1970 to 2008. They make a distinction between the long-run and the short-run effects, although they do not determine specific periods of time. They find a small positive short-run effect that is significant at a level of 10%. In the long-run, they do not find a significant effect.
1.5. Hypotheses
The majority of the studies reviewed find that inward FDI has an effect on income inequality. It is however questionable whether this effect is positive or negative, since the results are ambiguous. Neither theory nor empirical analysis is conclusive about the direction of the effect. But as section 1.4 on earlier empirical results points out, relatively more studies find a positive effect, especially the ones focusing on developing countries. Therefore, in this paper, the
11
following hypothesis is tested: inward FDI has a positive effect on income inequality in developing countries.
2
Empirical analysis
2.1 Methodology
In this section, the methodology and data for the empirical analysis are presented. A panel data analysis is conducted on 15 countries for the period from 1980 to 2004. The following model is estimated:
GINIit = β0+ β1FDIit+ β2AGRIit+ β3SCHOOLit+ β4TRADEit+ β5PCGDPit+ β6PCGDPGit+ β7TIMEit+ εit
Subscript i indicates the country and subscript t denotes the year. The model consists of the following variables:
GINIit : Gini coefficient
FDIit : inward FDI stock (% of GDP)
AGRIit : value added to GDP by the agricultural sector (% of GDP)
SCHOOLit : secondary school enrollment, 10 years prior to the Gini coefficient
observation
TRADEit : openness to trade:
(X+M) GDP
PCGDPit : the natural logarithm of per capita GDP
PCGDPGit : per capita GDP, annual growth (%)
TIMEt : indicator of time
εit ~ N(0, σ2) : normally distributed error term
2.1.1 Dependent variable
The Gini-coefficient is used as the dependent variable (GINI), because this coefficient is used as the explained variable in almost all reviewed studies. As section 1.2 points out, the value of the Gini coefficient lies between 0 and 100, which means that the coefficient is bounded. Reuveny and Li (2003) transform the coefficient into an unbounded coefficient before using it as a variable, because OLS requires an unbounded dependent variable. They regress the
12
unbounded variable and later do a robustness check on the bounded variable. In both regressions, they find the same significant variables. Furthermore, studies that use the bounded coefficient as dependent variable, encounter no difficulties. Therefore, the bounded Gini coefficient is used as dependent variable in this paper.
Various researchers have approached the question by dividing labor into skilled and unskilled labor, to see how FDI changes the wage differentials among these two categories (e.g. Driffield and Taylor, 2000; te Velde and Morrissey, 2004). However, Figini and Görg (2011) state that there are several theoretical reasons for using general Gini coefficients, instead of making a distinction between skilled and unskilled labor only. One of the reasons they give is that both absolute and relative employment in both sectors change due to FDI. This effect would not be captured when only focussing on skilled and unskilled sectors. They also state that the simple wage gap between skilled and unskilled labor cannot capture the difficult interaction between FDI, innovation and the labor structure of a firm: innovation affects all workers, thereby changing wages in both the skilled and the unskilled sector. Furthermore, there may be wage dynamics within the sectors that are not captured when only separating skilled from unskilled workers. For example because multinationals may not invest in an entire sector, but rather in that specific parts of the sector in which the host country has a comparative advantage. For these reasons, general Gini coefficients are used in this analysis, instead of separating skilled labor from unskilled labor only.
2.1.2 Independent variables
The Gini coefficient is regressed on various explanatory variables. According to Lin, Kim and Lee (2015) it is important to control for structural economic shifts, due to internal and external economic disturbances. Therefore, the regression contains variables to control for country specific effects and both economic and non-economic development in the host country.
FDI
To indicate the volume of foreign direct investment in the host country, FDI is included as a variable. FDI is measured by the inward stocks, as a percentage of GDP. Following Figini and Görg (2011), Chintrakarn et al. (2012) and Herzer et al. (2014), FDI stocks instead of FDI flows are used, since stocks represent long-run effects more effectively. Economic theory is not conclusive about the effect of FDI on the Gini coefficient. Earlier empirical results are inconclusive as well, although there is a tendency towards a positive effect, especially for developing countries. It is therefore expected that the coefficient of FDI is positive.
13 AGRI
The first control variable that is included as a proxy for the economic structure in the host country, is AGRI, which is estimated by the value added by the agricultural sector as a percentage of total GDP. According to Tsai (1995), there is common agreement that the economic structure is an important determinant of income inequality. Since agriculture is the principal economic activity in most developing countries, this sector is chosen as the indicator for economic structure. Tsai uses the percentage of the labor force that is active in the agricultural sector. However, data on this percentage are scarcely available for developing countries, so the value added by the agricultural sector is used instead. Tsai states that it is widely agreed upon that income inequality is increasing in the share of the labor force that is active in the agricultural sector. The result differs when the value added by the sector is used as a proxy. Ahluwalia (1976) for example, includes the value added as a proxy for economic structure. He conducts an empirical study on economic growth and inequality. He estimates a time series model with income distribution as dependent variable and he finds a negative effect of the value added by the agricultural sector on income inequality. Ahluwalia explains this effect by stating that a decrease in the value added by the agricultural sector is usually the result of economic development. This economic development has a twofold effect. On the one hand, it implies a shift of income from the middle to the upper sector, which increases income inequality. On the other hand, there is a population shift from the lowest income group toward the nonagricultural, modern sectors. In his empirical study, the latter effect appears to be larger, which means that the total effect on income inequality of value added by the agricultural sector is negative. Although Ahluwalia uses a different measure of income inequality, his findings of a negative effect may indicate that AGRI has a negative effect on the Gini coefficient as well.
SCHOOL
The secondary school enrollment is included to control for the effect of human capital. Secondary school enrollment is often used as a measure for human capital formation, as it is considered a consistent measure across countries (Barro, 1989). The relationship between human capital and income inequality is ambiguous. Some theories state that an increase in the share of youth completing secondary school enlarges the pool of skilled workers in the labor force. This in turn decreases the skilled labor premium, thereby decreasing income inequality (Herzer et al., 2014; Alderson & Nielsen, 1999; Figini & Görg, 2011; Tsai, 1995). But there are other studies that suggest a negative effect. Martins and Pereira (2004) find that the return to schooling increases over the wage distribution. This means that the effect of extra schooling on
14
wage depends on the characteristics that determine the position in the wage distribution. They explain this by stating that there may be an interaction between schooling and other factors that affect wage differentials. If these factors have a larger impact on the wages of laborers with higher levels of schooling, the return to schooling may increase over the wage distribution. The results of a study conducted on education, inequality and economic mobility in South Africa by Hertz (2001) are in line with this reasoning. Hertz finds that the logarithm of expected earnings is convex in years of schooling. This implies that even if educational attainment becomes more equally distributed, this may be associated with greater inequality. Lemieux (2006) finds mixed results in his study on the effect of schooling for male hourly wages in the US. He finds no significant effect of secondary education on wage inequality, but post-secondary education has a positive and significant effect on wage inequality.
When including school enrollment in the analysis, it must be taken into account that there is a longer term effect, due to a substantial time lag. This means that there is a time-match problem in the regression. To overcome this problem, Tsai (1995) uses values of SCHOOL from 10 years prior to the year of the corresponding Gini-coefficient. This approach is also used in this analysis.
TRADE
Following Figini and Görg (2011) and Basu and Guariglia (2007) among others, the trade openness is included as a variable. It is captured by the total amount of imports and exports as a fraction of GDP: 𝑋+𝑀
𝐺𝐷𝑃. The reason for including trade openness into the equation can be found
within the Stolper-Samuelson theorem, which was laid out in section 1.3.1. It can be stated that developing countries are relatively well endowed with unskilled labor. Since trade benefits the owners of the abundant factor, it would benefit workers and hurt the owners of capital. Hence trade would decrease income inequality (Reuveny & Li, 2003; Figini & Görg, 2011). This expectation is also supported by the study Chakrabarti (2000) conducts. He analyzes a sample of 73 countries in 1985. He finds that greater trade participation leads to less income inequality. This is not because countries with less inequality for reasons others than trade tend to participate in trade more. Rather, growth promotes subsequent growth and an increase in initial income, which both lead to lower income inequality.
15 PCGDP
To indicate economic development, the natural logarithm of per capita GDP is included. The choice for per capita GDP instead of GDP is based on several studies (e.g. Tsai, 1995; Choi, 2006; Figini and Görg, 2011). According to Tsai, there is a problem with comparability of per capita GDP values across countries, due to the use of official exchange rates that fail to reflect the purchasing power parity. For that reason, PCGDP is measured in international dollars, representing the purchasing power parity. The expectation is that the coefficient of PCGDP is negative, since Tsai, Choi, Figini and Görg all find significant negative effects. This implies that a larger country tends to have less income inequality.
PCGDPG
Following different studies (e.g. Choi, 2006; Ucal et al., 2014), the annual percentage growth of per capita GDP is included in the equation as well. Choi (2006) states that fast growing countries tend to have a more equal income distribution. Both Choi (2006) and Ucal et al. (2014) find significant negative effects for GDP growth. Therefore, the expectation is that the effect of PCGDPG is negative.
YEAR
To control for trends and time specific changes that are not captured by other variables, a variable YEAR is included.
Two different analyses are conducted on the dataset. First, the annual observations are analyzed. Then, the data is converted to three-year averages, to eliminate short-term fluctuations. In both cases, a panel data analysis using fixed effects is done, to investigate the impact of the independent variables on the dependent variable over time, while controlling for time-invariant effects.
2.2 Data
The empirical study is focused on countries that are indicated as low income countries by the World Bank and UNU-WIDER. The following countries are selected on the basis of the availability of the Gini coefficient:
16
Figure 2: Countries included in sample
The data on the Gini coefficients are retrieved from the Standardized World Income Inequality Database, a database that intends to overcome the incomparability of the measures across different datasets. It provides both the gross Gini coefficient, whose measurement is based on gross income, and the net Gini coefficient, whose measurement is based on net income. To avoid the bias that might arise due to different tax policies across countries, gross Gini coefficients are used instead of net Gini coefficients.
Data on inward FDI stocks as percentage of GDP are retrieved from the database of the United Nations Conference on Trade and Development (UNCTAD). The data on per capita GDP in current international dollars are mainly taken from the World Bank. For the countries that miss observations in the World Bank dataset, observations from the Humanitarian Data Exchange (HDX) are used instead. Data on the value added by the agricultural sector are also taken from the World Bank, as is almost all of the data on trade. Only for the trade in Ethiopia, data from UNCTAD is used. For secondary school enrollment, data from either the World Bank or United Nations Educational, Scientific and Cultural Organization (UNESCO) is used. Also in this case, there is both a gross and a net measure. The gross enrollment ratio measures the number of children enrolled in secondary education, regardless of age, and divides this to the number of children who officially should be enrolled in secondary education. The net enrollment ratio on the other hand, measures the children enrolled in secondary education that are in the corresponding age group, and divides this to the total number of children in the corresponding age group. For this analysis, the gross enrollment ratio is used, because any child that is enrolled in secondary education might affect income inequality ten years later, despite his or her age during the education.
Before analyzing the effect of the independent on the dependent variable, the correlations between the explanatory variables are evaluated. These correlations are shown in Table 1 and Table 2 for the annual and the averaged sample respectively.
Burkina Faso Madagascar Niger
Burundi Malawi Sierra Leone
Ethiopia Mali Tanzania
Guinea Mozambique Uganda
17
Table 1: Correlations Annual Sample
GINI FDI AGRI SCHOOL TRADE PCGDP PCGDPG TIME
GINI 1 FDI 0.2304 1 AGRI -0.3480 -0.0493 1 SCHOOL 0.1319 -0.1312 -0.1980 1 TRADE 0.3672 0.2200 -0.4220 0.3291 1 PCGDP 0.1070 -0.1251 -0.3663 0.3216 0.1484 1 PCGDPG -0.0702 -0.0429 -0.0336 0.0098 0.0473 0.0225 1 TIME -0.1505 0.1005 -0.2456 0.1364 0.2097 0.3797 0.2058 1
In case of high correlations between the explanatory variables, multicollinearity problems may arise. The correlations in Table 3 are all around |0.4| or lower. This leads to the conclusion that from these correlations, no multicollinearity problems should be expected.
Table 2: Correlations Averaged Sample
GINI FDI AGRI SCHOOL TRADE PCGDP PCGDPG TIME
GINI 1 FDI 0.3417 1 AGRI -0.1580 0.2349 1 SCHOOL -0.0676 -0.0621 -0.0473 1 TRADE 0.3251 0.1330 -0.2201 0.2322 1 PCGDP -0.1933 0.0319 -0.2770 0.2069 -0.0294 1 PCGDPG -0.3290 -0.2614 0.0784 -0.0602 -0.1405 0.1204 1 TIME -0.1521 0.1219 -0.2278 0.1312 0.2764 0.5364 0.4362 1
Table 2 shows that also in the averaged sample, most of the correlations between the variables are low. The highest correlations are 0.5364, between TIME and PCGDP, and 0.4362, between TIME and PCGDPG. Because of the low correlation between all other variables, it can be expected that no multicollinearity problems would arise in the analysis.
18
3
Results
In this section, the results of the empirical analysis are provided. First, the results of the analysis using annual data are provided. Then, the results of the analysis on the averaged data are laid out.
3.1
Annual data
The results of the analysis on the annual dataset are provided in Table 3. For each estimated coefficient, the standard error is provided in parentheses. The table also shows the R-squared, the coefficient of determination, which indicates how much of the variance in the dependent variable is explained by the variance in the independent variables. Below that, the F-values, indicating the significance of the model, are provided. These F-values are based on clustered standard errors. This generates 15 observations, equal to the number of countries included in the dataset. The reason for clustering is that there may be correlation between the standard errors within countries, but not across countries.
In the first model that is estimated, FDI is the only explanatory variable that is included. The estimated effect of FDI is negative in this model, indicating that income inequality decreases as inward FDI stock increases. However, this negative effect is insignificant, as is the model as a whole. This implies that more independent variables should be added. In the second model, per capita GDP and per capita GDP growth are included. This model is significant at the level of 0.05, but per capita GDP growth is the only significant variable and the R-squared of the model is low. Model 3 investigates the effect of FDI and the country specific factors (AGRI, SCHOOL and TRADE) on the Gini coefficient. This model has a higher R-squared, but only two variables are significant at the level of 10 per cent. The coefficient of FDI is negative in both the second and the third model, but significant in neither. The estimated coefficients of AGRI and SCHOOL are negative and positive respectively, and both significant.
The fourth model includes both the country specific factors and the GDP-related variables. The difference in the R-squared values of the third and the fourth model is small, and both GDP-variables are insignificant. When testing the hypothesis that the joint effect of PCGDP and PCGDPG is zero, an F-value of 0.12 (with a p-value of 0.8908) is found. It therefore seems reasonable to estimate a model that does not include these variables. The control variables that are included in the fifth model are the country specific factors and the time variable. Again, the coefficient of FDI is negative but insignificant and the other variables are all significant. The coefficient of TRADE is positive, against the expectations. The
R-19
squared of the model increases to 0.375 and the model is significant. The final model that is estimated is the model including all explanatory variables. This generates a negative, insignificant coefficient for FDI, but AGRI, SCHOOL, TRADE and TIME are all significant. The whole model is significant too, and the model has an R-squared of 0.425.
Table 3: Results analysis annual data
Robust standard errors in parentheses below the estimated coefficients, significance indicated as follows: * if p < 0.10, ** if p < 0.05, *** if p < 0.01. F-value indicates the significance of the model, with the degrees of freedom below the value. The indicated R-squared is the value of the within R-squared.
From the analysis on annual data, it can be concluded that the variable of interest – FDI – has a negative coefficient in all models, which implies that the Gini coefficient decreases with higher levels of inward FDI stocks. The effect is however insignificant. The effect of AGRI is negative in all models, but insignificant when both the GDP variables and the time variable are
Dependent variable GINI Independent variables (1) (2) (3) (4) (5) (6) FDI -0.283 -0.481 -0.582 -0.563 -0.205 -0.169 (0.36) (0.36) (0.36) (0.37) (0.26) (0.25) AGRI -0.269* -0.292** -0.284** -0.150 (0.14) (0.13) (.011) (0.10) SCHOOL 0.345* 0.396** 0.748** 0.603** (0.18) (0.17) (0.29) (0.22) TRADE 0.094 0.102 0.143** 0.119** (0.06) (0.07) (0.05) (0.05) PCGDP 8.735 -2.161 13.256** (6.06) (5.10) (5.14) PCGDPG -0.254** -0.007 .018 (0.12) (0.10) (0.10) TIME -0.776*** -1.097*** (0.24) (0.20) Constant 63.412*** 7.771 67.538*** 81.465** 69.118*** -15.709 (3.03) (39.21) (5.96) (32.18) (4.40) (34.13) N 184 184 157 157 157 157 R-squared 0.021 0.102 0.222 0.224 0.375 0.4255 F-value 0.625 2.180** 4.280*** 5.738*** 24.165*** 18.17*** df (1, 14) (3, 14) (4, 14) (6,14) (5, 14) (7, 14)
20
included. The effect of SCHOOL is positive in all models and this result is robust, since it is significant at either the 10- or the 5 percent level in all models. TRADE is significant in the models that include the time variable. It has, against the expectations, a positive coefficient in all models.
Per capita GDP is significant only in the model that includes the time variable, where it has a strong positive effect, which was not expected. This might be due to correlation between PCGDP and TIME. The effect of per capita GDP growth is negative in the second and fourth model. When the time variable is included however, the coefficient of PCGDPG is positive, against the expectations. The effect of the time variable is negative and significant in both models in which it is included. These results indicate that there is a decrease in the Gini coefficient over time, which is the effect of time-varying factors that are not captured by the included explanatory variables.
3.2
Averaged data
The results of the analysis on the averaged data set are provided by Table 4. The coefficients are provided with the standard errors in parentheses below. The table also provides the R-squared and F-value of the model based on clustered standard errors.
The first model consists of one explanatory variable: FDI. The coefficient of FDI is negative, but insignificant, as is the model as a whole. Adding per capita GDP and per capita GDP growth generates a model that is significant at the level of 10 percent. The effect of FDI is still negative and insignificant. PCGDP has a positive effect that is significant at the level of 10 percent.
The third model investigates the effect of FDI in combination with the country specific factors. This model is significant at the level of 1 percent, with AGRI and FDI both having a negative effect that is significant at the level of 10 percent. In the fourth model, the variables of both the second and the third model are included. This generates a significant model with an R-squared of 0.432. The effect of FDI is still negative and insignificant and per capita GDP growth is significant now at the level of 5 per cent. The coefficient of PCGDPG is negative, which was expected based on earlier empirics.
In the fifth model, the Gini coefficient is explained by FDI, the country specific variables and the time variable. The difference between the values of the R-squared in the fourth and the fifth is large. This implies that TIME explains a large part of the variation in the dependent variable. In the fifth model, the effect of FDI is still insignificant, but now slightly positive.
21
Table 4: Results analysis averaged data
Robust standard errors in parentheses below the estimated coefficients, significance indicated as follows: * if p < 0.10, ** if p < 0.05, *** if p < 0.01. F-value indicates the significance of the model, with the degrees of freedom below the value. The indicated R-squared is the value of the within R-squared.
In the sixth model, all explanatory variables are included. This generates a highly significant model with an R-squared of 0.736. The effect of FDI is again positive and insignificant. AGRI, TRADE and TIME do have significant coefficients. The effect of TRADE is positive, which was not expected. Testing whether PCGDP and PCGDPG have a joint significant effect generates an F-value of 1.63 and a p-value of 0.25. This indicates that their joint effect is insignificant. Testing for the significance of the joint effect of PCGDP, PCGDPG and FDI generates an F-value of 1.13, with a p-value of 0.39. This implies that the joint effect of these three variables is insignificant too.
In general, it can be concluded from the analysis that FDI has an insignificant effect on the Gini coefficient. In the majority of the estimated models, the effect is negative, although it
Dependent variable GINI Independent variables (1) (2) (3) (4) (5) (6) FDI -0.841 -1.054 -1.165* -0.911 0.016 0.014 (0.58) (0.67) (0.51) (0.65) (0.31) (0.33) AGRI -0.512* -0.564** -0.402** -0.272** (0.25) (0.22) (0.14) (0.09) SCHOOL 0.203 0.347 0.846* 0.697 (0.29) (0.25) (0.38) (0.38) TRADE 0.162 0.219 0.398*** 0.354*** (0.12) (0.16) (0.07) (0.08) PCGDP 11.906* -6.702 12.072 (5.95) (9.25) (7.38) PCGDPG -0.753 -0.476** 0.092 (0.47) (0.18) (0.09) TIME -3.987*** -4.781*** (0.92) (0.66) Constant 68.259*** -8.210 80.813*** 120.371* 67.438*** -8.759 (4.73) (38.13) (8.29) (54.77) (5.85) (43.97) N 44 44 42 42 42 42 R-squared 0.091 0.259 0.389 0.432 0.707 0.736 F-value 2.115 3.419* 10.028*** 20.468*** 78.569*** 511.294*** df (1, 14) (3, 14) (4, 14) (6,14) (6, 14) (5, 14)
22
is positive in the two models that include TIME. The GDP variables seem to have no significant effect on income inequality. SCHOOL has a positive effect, which is insignificant in the majority of the models. AGRI has a significant negative effect in all models, which was not expected. TRADE is significant in the models that include the time variable. The effect of trade openness is positive, in contrast with the expectations. TIME has a significant and negative effect.
3.3
Assumptions
In this section, the validity of the assumptions underlying the analysis is evaluated. First, the OLS assumptions are discussed. Then, the differences between a panel data analysis with either fixed or random effects are laid out.
3.3.1 OLS assumptions
To estimate the model, several assumptions are made. For a panel data analysis with fixed effects, the Ordinary Least Squares assumptions must hold. These are the following:
1. Linearity
2. No large outliers 3. εit ~ N(0, σ2) :
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 explanatory variables and the Gini coefficient is linear. If this assumption is not met, non-linear data is fitted into a linear relationship, which yields wrong estimates. To see whether linearity is a reasonable assumption, Appendix 1 and 2 provide scatter plots of the residuals against the predictor variables, which show no clear non-linear pattern. Appendix 3 and 4 provide plots of the augmented partial residuals with non-linearity test results below. For the annual data, the plots show large deviations from linearity. Only TRADE and PCGDPG have a p-value larger than 0.05, which means that only these variables satisfy the linearity condition. Therefore, Appendix 5 provides the distributions of all explanatory variables. It can be seen that there are large deviations from normality, which may be the result of too little observations. For the averaged data, the deviations from linearity are smaller. For all variables, the non-linearity test generates a p-value
23
larger than 0.05. This suggests that the assumption of linearity is met for the averaged data, but not for the annual data.
Figure 3 and 4 show plots of the residuals against the fitted values. It can be stated that the variance appears to be roughly constant, although there are several outliers in both plots. To control for these outliers, all models are estimated with the robust option and the second assumption is met.
Figure 3: Residuals plot, annual data Figure 4: Residuals plot, averaged data
The third assumption concerns the error term. First, it has to be normally distributed. The distribution of the error term for both analyses can be found in Figures 5 and 6. Both figures show that the residuals are not normally distributed, especially in the middle range of the data. The deviations may be the results of the small sample size.
Figure 5: Normality of the residuals, annual data Figure 6: Normality of the residuals, averaged data
Another assumption is made regarding the conditional mean of the error, which means that the expectation of the error term is zero, given the value of the independent variables. Since the OLS estimation method is designed to generate residuals with an expectation of zero, it cannot
-2 0 -1 0 0 10 20 R e si d u a ls 50 55 60 65 70 Fitted values -2 0 -1 0 0 10 20 30 R e si d u a ls 50 60 70 80 Fitted values 0 .0 1 .0 2 .0 3 .0 4 .0 5 D e n sit y -20 -10 0 10 20 30 Residuals Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 2.9164
Kernel density estimate
0 .0 1 .0 2 .0 3 .0 4 .0 5 D e n sit y -20 -10 0 10 20 30 Residuals Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 3.5330
24
be tested whether this assumption holds. However, Figure 3 and 4 indicate that there is no clear pattern, so there is reason to expect that the assumption holds.
Furthermore, the assumption of homoscedasticity is made. This implies that the variance of the error term is constant. To test this, a Breusch-Pagan/Cook-Weisberg test is performed, which tests the null-hypothesis that the variance of the error term is constant. For the annual data, the test generates a Chi-squared value of 2.52 with a p-value of 0.1127, which means that the null-hypothesis is not rejected. For the averaged data, the test generates a Chi-squared value of 1.38 and a p-value of 0.2394. Again, the null-hypothesis is not rejected. This means that it can be assumed that the variance of the error term is constant in both the annual data set and the averaged data set.
In summary, it can be concluded that the assumption of linearity and the assumption of normality of the error term do not hold. This may be due to the small number of analyzed countries. All but one of these countries are located in the same continent. This may lead to small regional difference, which may have a consequence for the interpretation of the coefficients. This is discussed in section 3.4.
3.3.2 Fixed effects or random effects
When analyzing a panel data set, a regression with either fixed effects or random effects is used. With a fixed effects regression, time-invariant factors are being controlled for, whereas a random effects regression assumes that the variation across countries is random and uncorrelated with the dependent and independent variables. To determine whether the analyses have to be performed using fixed effects or random effects, a Hausman test is performed. The null hypothesis of this test is that a random effects model is preferred. For the annual data set, this test generates a Chi-squared value of 7.22 and a corresponding p-value of 0.4068. For the averaged data, the test generates a Chi-squared value of 9.31, with a p-value of 0.2314. This means that the null hypothesis is not rejected and a random effects model can be used in both cases. Because of these test results, the sixth model that was estimated with fixed effects for both the annual and the averaged data, is estimated again with random effects. The results of the regressions are presented in Table 5.
Comparing the results of the models based on the annual data with fixed and random effects –Model (6) in Table 3 and Model (i) in Table 5 respectively – it can be stated that all coefficients have the same direction, except for the coefficient of the constant. The constant is insignificant in both models, but positive in Model (i). Another difference is that the coefficient of AGRI is significant at the level of 5 per cent in Model (i), whereas it is insignificant in Model
25
(6). The effect of FDI is negative and insignificant in both models. The effects of SCHOOL, TRADE and PCGDP are significant at the 1%-level, instead of the 5%-level.
Table 5: Results analysis random effects
Dependent variable GINI Independent variables (i) Annual data (ii) Averaged data FDI -0.094 0.097 (0.20) (0.27) AGRI -0.165** -0.298** (0.08) (0.14) SCHOOL 0.561*** 0.474* (0.20) (0.25) TRADE 0.124*** 0.353*** (0.05) (0.08) PCGDP 10.478*** 10.477* (3.89) (5.43) PCGDPG 0.037 0.026 (0.09) (0.22) TIME -0.997*** -4.136*** (0.17) (0.81) Constant 2.501 3.277 (26.73) (23.01) N 157 42 R-squared 0.4229 0.7275 Wald Chi squared 147.52*** 965.96*** df 7 7
When comparing the fixed and random effects models based on the averaged data –Model (6) in Table 4 and Model (ii) in Table 5 respectively – it can again be concluded that the only coefficient that has a different direction is the coefficient of the constant. All other coefficient
Robust standard errors in parentheses below the estimated coefficients, significance indicated as follows: * if p < 0.10, ** if p < 0.05, *** if p < 0.01. The indicated R-squared is the value of the within R-squared. Wald Chi squared indicates the significance of the model, with the degrees of freedom below the value.
26
have the same direction. In Model (ii) both SCHOOL and PCGDP are significant at the level of 10%, whereas neither of these variables is significant in Model (6). The effect of FDI is positive and insignificant in both models.
In summary, it can be stated that a random effects analysis generates models with more significant variables than a fixed effects analysis. But in both analyses, FDI has an insignificant effect on the Gini coefficient.
3.4
Sample size and regional differences
In this paper, only a small number of countries is investigated, due to lack of consistent measures of the Gini coefficient. All but one of the analyzed countries are African countries. This implies that regional differences, which can be expected in a sample with countries in various continents, may be small or even negligible. Hence, certain mechanisms that are specific to FDI in African countries may play a role.
It would therefore be insightful to analyze the structure of the FDI stocks to indicate whether there is a clear pattern of certain industries toward which the FDI is mainly directed. Unfortunately, specific data on the structure of the FDI flows toward the countries that are investigated in this paper are not available. Figure 7 provides an overview of the inward FDI stock in African countries, structured by industry. The figure shows that within Sub-Saharan Africa, about 35% of the FDI stocks are held by the primary sector. This sector consists of all sectors that exploit natural resources directly, such as the agricultural sector, the mining sector, the hunting sector (United Nations, 2005). The other fraction of total FDI stock is held by either the service industry (20%) or the manufacturing sector (45%). If more detailed data on FDI by sector would be available, it could be indicated whether the structure of the investment matters for the effect on income inequality.
27
Figure 7: Africa's inward FDI stock by sector and by sub region (2012)
4
Conclusion
This paper investigates the relationship between inward FDI and income inequality in developing countries. First, the insights that economic theory and earlier empirical research provide are reviewed. The main findings are that both theory and earlier empirical results are inconclusive about the effect of inward FDI on income inequality. To investigate the effect, an empirical analysis is conducted by analyzing a panel data on both an annual and a three years averaged basis. In the analysis on the annual data, the coefficient of FDI is negative and insignificant in all estimated models. The analysis on the averaged data set generates an insignificant negative effect in the majority of the models. The effect is negative and significant at the 10%-level in the model that includes the country specific factors. In the two models that include a time variable, the effect of FDI is positive and insignificant. In both analyses, the effect of the value added by the agricultural sector is negative and significant in the majority of the estimated models, as was expected. The effect of secondary school enrollment is positive and significant in all models for the annual analysis. Although the effect is positive in the averaged data analysis too, the results were not robust there. The coefficient of openness to trade has an unexpected direction. It has a positive effect in both the averaged and the annual data analysis, which is significant in all models that include a time variable as well. There are
28
no robust findings on the effect of per capita GDP and per capita GDP growth on income inequality, whereas the time variable has a significant negative effect on income inequality. From the empirical analysis, it can be concluded that the hypothesized positive effect is not supported in this sample.
The unexpected results may be the result of the violation of some of the OLS assumptions. A larger sample may solve this problem. If more standardized Gini coefficients are available in the future, a suggestion for further research would be to conduct the analysis on a larger data set, with more geographical diversity and more observations for each country. It would also be interesting to investigate whether the structure of FDI flows influences the effect on the Gini coefficient. If more data become available on the FDI flows by sector, it can be indicated whether flows toward certain sectors have a different effect on income inequality than flows toward other sectors.
29
References
Ahluwalia, M. (1976). Inequality, Poverty and Development. Journal of Development Economics, 3, 307-342.
Alderson, A., & Nielsen, F. (1999). Income Inequality, Development, and Dependence: A Reconsideration. American Sociological Review, 64(4), 606-631.
Alfaro, L., Chanda, A., Kalemli-Ozcan, S., & Sayek, S. (2004). FDI and economic growth: the role of local financial markets. Journal of International Economics, 64(1), 89-112. Barro, R. (1989). Economic Growth in a Cross Section of Countries. NBER Working papers
no. 3120, 1-49.
Basu, P., & Guariglia, A. (2007). Foreign Direct Investment, Inequality, and Growth. Journal of Macroeconomics, 29(4), 824-839.
Brown, D. K., Deardorff, V. A., & Stern, M. R. (2003, April). The Effects of Multinational Production on Wages and Working Conditions in Developing Countries. NBER Working Paper 9669.
Chakrabarti, A. (2000). Does Trade Cause Inequality? Journal of Economic Development, 25(2), 1-22.
Cheng, Y.-M. (2006). Determinants of FDI Mode Choice: Acquisition, Brownfield, and Greenfield Entry in Foreign Markets. Canadian Journal of Administrative Sciences, 23(3), 202-220.
Chintrakarn, P., Herzer, D., & Nunnenkamp, P. (2012). FDI and Income Inequality: Evidence from a Panel of U.S. States. Economic Inquiry, 50(3), 788-801.
Choi, C. (2006). Does Foreign Direct Investment Affect Domestic Income Inequality? Applied Economic Letters, 13(12), 811-814.
Driffield, N., & Taylor, K. (2000). FDI and the labor market: a review of the evidence and policy implications. Oxford Review of Economic Policy, 16(3), 90-103.
Figini, P., & Görg, H. (2011). Does Foreign Direct Investment Affect Wage Inequality? An Empirical Investigation. The World Economy, 34(9), 1455-1475. doi:10.1111/j.1467-9701.2011.01397.x
Gastwirth, J. (1971). A General Definition of the Lorenz Curve. Econometrica, 39(6), 1037-1039.
Gastwirth, J. (1972). The Estimation of the Lorenz Curve and Gini Index. The Review of Economics and Statistics, 54(3), 306-316.
Hertz, T. N. (2001). Education, income inequality and economic mobility in South Africa. Doctoral Dissertations Available from Proquest, Paper AAI3027205.
Herzer, D., & Nunnenkamp, P. (2013). Inward and outward FDI and income inequality: evidence from Europe. Review of World Economics, 149(2), 395-422.
Herzer, D., Hühne, P., & Nunnenkamp, P. (2014). FDI and Income Inequality - Evidence from Latin American Economies. Review of Development Economics, 18(4), 778-793. doi:10.1111/rode.12118
Lemieux, T. (2006). Post-secondary Education and Increasing Wage Inequality. American Economic Review, 96(2), 195-199.
Lin, S.-C., Kim, D.-H., & Lee, Y.-H. (2015). Financial Development and the FDI-Inequality Nexus. Contemporary Economic Policy, 33(3), 513-534.
Martins, P., & Pereira, P. (2004). Does education reduce wage inequality? Quantile regression evidence from 16 countries. Labour economics, 11(3), 355-371.
Moosa, I. A. (2002). Foreign Direct Investment: Theory, Evidence and Practice. Palgrave. Nielsen, F., & Anderson, A. (1997). The Kuznets Curve and the Great U-Turn: Income
Inequality in U.S. Counties, 1970 to 1990. American Sociological Review, 62(1), 12-33.
30
Reuveny, R., & Li, Q. (2003). Economic Openness, Democracy, and Income Inequality - an Empirical Analysis. Comparative Political Studies, 36(5), 575-601.
Sylwester, K. (2005). Foreign Direct Investment, Growth and Income Inequality in Less Developed Countries. International Review of Applied Economics, 19(3), 289-300. Te Velde, D. W., & Morrissey, O. (2004). Foreign Direct Investment, Skills and Wage
Inequality in East Asia. Journal of the Asia Pacific Eocnomy, 9(3), 348-369.
Tsai, P.-L. (1995). Foreign direct investment and income inequality: Further evidence. World Development, 23(3), 469-483. doi:10.1016/0305-750X(95)00136-Z
Ucal, M., Haug, A. A., & Bilgin, M. H. (2014). Income Inequality and FDI: evidence with Turkish data. Applied Economics, 48(11), 1030-1045.
UNCTAD. (2015). World Investment Report 2015. 1-15. Retrieved December 2015 United Nations. (2005). Economic Development in Africa - Rethinking the Role of Foreign
Direct Investment. New York and Geneva: United Nations.
Wang, M., & Wong, S. (2009). Foreign direct investment and economic growth: the growth accounting perpective. Economic Inquiry, 47(4), 701-710.
Zhang, K. (2006). Foreign direct investment and economic growth in China: A panel data study for 1992-2004. Conference of WTO, China and Asian Economies, UIBE, Beijing, China., (pp. 1-18).
31
Appendix 1
Annual data: residuals plotted against explanatory variables of Model 6
-2 0 -1 0 0 10 20 30 R e si d u a ls 0 10 20 30 40 FDI -2 0 -1 0 0 10 20 30 R e si d u a ls 10 20 30 40 50 60 AGRI -2 0 -1 0 0 10 20 30 R e si d u a ls 0 10 20 30 40 SCHOOL -2 0 -1 0 0 10 20 30 R e si d u a ls 20 40 60 80 100 TRADE -2 0 -1 0 0 10 20 30 R e si d u a ls 5.5 6 6.5 7 7.5 lnpcgdp -2 0 -1 0 0 10 20 30 R e si d u a ls -20 -10 0 10 20 PCGDPG
32
Appendix 2
Averaged data: residuals plotted against explanatory variables of Model 6
-2 0 -1 0 0 10 20 R e si d u a ls 0 10 20 30 FDI -2 0 -1 0 0 10 20 R e si d u a ls 20 30 40 50 60 AGRI -2 0 -1 0 0 10 20 R e si d u a ls 0 10 20 30 40 SCHOOL -2 0 -1 0 0 10 20 R e si d u a ls 30 40 50 60 70 TRADE -2 0 -1 0 0 10 20 R e si d u a ls 6 6.5 7 7.5 lnpcgdp -2 0 -1 0 0 10 20 R e si d u a ls -5 0 5 10 PCGDPG
33
Appendix 3
Annual data: augmented partial residual plot with F-value and p-value of non-linearity test below. F-value = 6.38 F-value = 2.08 p-value = 0.000 p-value = 0.035 F-value = 2.29 F-value = 0.61 p-value = 0.020 p-value = 0.787 F-value = 2.28 F-value = 0.66 p-value = 0.020 p-value = 0.743 -2 0 -1 0 0 10 20 30 Au g me n te d co mp o n e n t p lu s re si d u a l 0 10 20 30 40 FDI -3 0 -2 0 -1 0 0 10 20 Au g me n te d co mp o n e n t p lu s re si d u a l 10 20 30 40 50 60 AGRI -2 0 -1 0 0 10 20 30 Au g me n te d co mp o n e n t p lu s re si d u a l 0 10 20 30 40 SCHOOL -1 0 0 10 20 30 Au g me n te d co mp o n e n t p lu s re si d u a l 20 40 60 80 100 TRADE -2 1 0 -2 0 0 -1 9 0 -1 8 0 -1 7 0 Au g me n te d co mp o n e n t p lu s re si d u a l 5.5 6 6.5 7 7.5 lnpcgdp -2 0 -1 0 0 10 20 30 Au g me n te d co mp o n e n t p lu s re si d u a l -10 0 10 20 30 PCGDPG
34
Appendix 4
Averaged data: augmented partial residual plot with F-value and p-value of non-linearity test below. F-value = 2.18 F-value = 1.72 p-value = 0.061 p-value = 0.137 F-value = 1.58 F-value = 0.26 p-value = 0.174 p-value = 0.981 F-value = 1.55 F-value = 0.77 p-value = 0.185 p-value = 0.642 -2 0 -1 0 0 10 20 30 Au g me n te d co mp o n e n t p lu s re si d u a l 0 10 20 30 FDI 10 20 30 40 50 60 Au g me n te d co mp o n e n t p lu s re si d u a l 20 30 40 50 60 AGRI -3 0 -2 0 -1 0 0 10 Au g me n te d co mp o n e n t p lu s re si d u a l 0 10 20 30 40 SCHOOL 0 10 20 30 40 50 Au g me n te d co mp o n e n t p lu s re si d u a l 30 40 50 60 70 TRADE 1 5 0 1 6 0 1 7 0 1 8 0 1 9 0 Au g me n te d co mp o n e n t p lu s re si d u a l 6 6.5 7 7.5 lnpcgdp -2 0 -1 0 0 10 20 Au g me n te d co mp o n e n t p lu s re si d u a l -5 0 5 10 PCGDPG
35
Appendix 5
Annual data: distribution of explanatory variables
0 .0 2 .0 4 .0 6 .0 8 D e n sit y 0 10 20 30 40 FDI Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 2.2876
Kernel density estimate
0 .0 1 .0 2 .0 3 .0 4 D e n sit y 0 20 40 60 80 AGRI Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 3.4633
Kernel density estimate
0 .0 2 .0 4 .0 6 D e n sit y 0 10 20 30 40 SCHOOL Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 2.3186
Kernel density estimate
0 .0 1 .0 2 .0 3 D e n sit y 20 40 60 80 100 TRADE Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 4.1164
Kernel density estimate
0 .2 .4 .6 .8 1 D e n sit y 5.5 6 6.5 7 7.5 lnpcgdp Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 0.1221
Kernel density estimate
0 .0 5 .1 .1 5 D e n sit y -20 -10 0 10 20 PCGDPG Kernel density estimate Normal density
kernel = epanechnikov, bandwidth = 1.0257
