How inward and outward FDI relate to income inequality within
developed and developing countries: an empirical study
Marloes Korendijk S2753189
m.m.korendijk@student.rug.nl
University of Groningen Faculty of Economics and Business Msc International Economics & Business
Date: 07-01-2019
2
Abstract
World income inequality has increased over the past two centuries where Foreign Direct Investment (FDI), as an aspect of globalization, is one of the main drivers. This paper aims to diminish the ambiguousness in the literature and proposes new thoughts and findings regarding the relation between FDI and income inequality. The paper contributes to previous research by examining both inward and outward FDI simultaneously in the same regression. Besides that, developed and developing countries are researched separately. Furthermore, differences are expected by this paper regarding the relation between FDI and income inequality before and after the second unbundling. A fixed effects panel regression of an unbalanced dataset of 162 countries with the years ranging from 1970 to 2016 shows that inward FDI is related to an increase in income inequality in developed countries while outward FDI is related to a decrease in income inequality in developed countries after 1995.
3
Index
1. INTRODUCTION ... 4
2. THEORETICAL BACKGROUND ... 7
2.1 Foreign Direct Investment ... 7
2.2 Effects of Foreign Direct Investment ... 8
2.3 A New Paradigm ... 9
2.4 Effects of Inward FDI ... 10
2.5 Effects of Outward FDI ... 12
3. METHODOLOGY AND DATA ... 14
3.1 Dependent and Independent Variables ... 15
3.2 Control Variables ... 16 3.3 Descriptive Statistics ... 17 3.4 Estimation Method ... 18 4. EMPIRICAL RESULTS ... 19 4.1 Hypotheses Testing ... 20 4.2 Robustness Check ... 24 4.3 Discussion ... 28
4.4 Limitations & Future Research ... 30
6. REFERENCES ... 32
4
1. INTRODUCTION
World income inequality has increased over the past two centuries (Jaumotte, Lall, & Papageorgiou, 2013), which means that income becomes more unevenly spread. This is a concerning trend, as a more even distribution of income addresses the welfare and social concerns regarding an uneven income inequality. Social concerns may lead to social unrest which can be of international harm (Jaumotte et al, 2013). As for welfare, the government can support all citizens by reducing income inequality with certain policies. Policy and institutional differences are one of the reasons why developed countries with similar technological and productivity developments still have different patterns of income inequality (Alvaredo, Atkinson, Piketty, & Saez, 2013). For instance, taxation differences across countries can change income inequality. A strong correlation exists between the reductions in top tax rates and the increase in top 1 percent pre-tax income shares (Alvaredo et al, 2013).
One of the main drivers of income inequality is globalization (Bourguignon & Morisson, 2002). Globalization creates a larger labor market, which increases wage competition between workers. A rather skeptical view of globalization is brought to light by the American president Donald Trump. Trump argues that the Unites States of America (USA) needs a retreat of the global economy as he believes that localization is key in order to decrease the inequality within the country (Ghemawat, 2017).
5
Figure 1: Top 1% income share Source: World Inequality Database
As seen, globalization is related to an increase of within-county income inequality. One aspect of globalization is Foreign Direct Investment (FDI). While foreign direct investment is positively related to firm-level productivity and macroeconomic growth, the possible downsides regarding the issue of income inequality are neglected (Figini & Görg, 2011; Herzer & Nunnenkamp, 2013). Foreign direct investment is therefore the focus of this study and is defined as “a form of international inter-firm co-operation that involves a significant equity stake in, or effective management control of, foreign enterprises” (De Mello, 1997, p.4). Two types of FDI can be distinguished, namely inward and outward FDI. A country can engage in inward FDI when one is receiving a part of a foreign firm, and its capital, from another country. A country engages in outward FDI when one is outsourcing a part of its firm, and its capital, to another country. The source country is the country engaging in outward FDI, whereas the host country receives it as inward FDI.
FDI is related to income inequality due to a capital shift resulting from ownership of assets abroad (FDI). The movement of capital from one country to another induces an increase in relative demand for skilled labor and an accumulation of capital (Feenstra & Hanson, 1997). Consequently, the accumulation of capital positively effects the relative wages and employment of skilled labor which results in an increase in income inequality. Although income inequality increases, FDI can still have positive effects since the rising supply (due to capital investments) in the South decreases the prices of the inputs (Feenstra & Hanson, 1997).
How FDI relates to income inequality is theoretically ambiguous since literature provides arguments to expect an increase, an increase and decrease, or a decrease in income inequality (Herzer & Nunnenkamp, 2013). First, an increase in inequality may occur when the skill intensity increases in a country. The skill intensity increases in the host country since inward FDI is demanding more skilled in the host country and increases in the source country since they outsource relatively unskilled labour. Therefore, the skill premium is increased in both the developed source country and the developing host country, which will increase the income inequality in both countries (Feenstra & Hanson, 1997). Second, an additional theory based on the model of Aghion, Howitt, Howitt, Brant-Collett and García-Peñalosa (1998)
0 0.05 0.1 0.15 0.2 0.25 1970 1980 1990 2000 2010 P er ce n tile o f to tal in co m e Year
Top 1% income share
USA United Kingdom
World China
6
suggests that the demand for skilled labor will first increase but will decrease afterwards due to the learning processes and knowledge gain. These learning spill-overs will decrease the income inequality in the longer run. Third, a decrease in inequality can occur when inward FDI is reducing the relative demand for skilled labor (Markusen & Venables, 1997). Reducing the relative demand for skilled labor occurs, for instance, when both source and host countries are developed countries. In this case, the host country has the same or higher skill-intensity for its own MNE’s and will thus not see a rising demand for skilled labor with the demand of foreign firms. This indicates that theoretically, different effects are expected between developed and developing countries.
Currently, FDI is not only targeting cheap and unskilled labor, but also high-skilled labor. This change represents the second unbundling, which reflects the reduction of communication costs (Baldwin, 2006). With the reduction of communication costs, FDI opportunities increase as well. Now, FDI can be directed towards service workers, particular tasks within jobs and talented workers, whereas it was previously mainly directed towards lower skilled people. This gives additional motives for FDI, which suggest that the theoretical reasoning may not cover all possible effects anymore.
There is some empirical evidence for the distributional consequences of inward FDI for individual countries like the United Kingdom and Ireland (Figini & Görg, 1999; Taylor & Driffield, 2005). Other studies focused on the differences between developed and developing host countries and their income inequality. For instance, Figini and Görg (2011) find that inequality increases when inward FDI is present in developing host countries, whereas inequality decreases in developed host countries. Conversely, other scholars find an increase in income inequality in developed countries (Lee, 2006; Blonigen & Slaughter, 2001).
Outward FDI is considered even less in previous literature (Herzer & Nunnenkamp, 2013). Therefore, how outward FDI relates to income inequality still remains ambiguous (Baldwin, 2006). Outward FDI relates to an increase in income equality as found by some scholars (e.g. Gopinath & Chen, 2003; Jaumotte et al, 2013), whereas others found that outward FDI relates to an increase and decrease of income inequality (Herzer & Nunnenkamp, 2013). Besides, outward FDI is almost solely examined for developed countries. This neglects the possibility that different effects occur for developing countries, while a difference between both type of countries is likely. After all, inward FDI has assumingly different effects for developed and developing countries. Subsequently, developing countries need to be considered as well, since outward FDI is becoming more common in developing countries (Dunning, Kim, & Park, 2008).
7
with more understanding behind the contradicting theories and findings of the relation. The paper examines the relation between FDI and income inequality for developed and developing countries separately, as theory expect that developing countries may sooner see an increase of skill-intensity and thus income inequality than developed countries. Further, the paper distinguishes before and after the second unbundling while examining how FDI relates to income inequality since the relation is hardly researched in this century. Therefore, a time gap exists in the literature, which means that potential changes of FDI effects after the second unbundling are not researched extensively. Concluding, this gap in literature is filled by researching how inward and outward FDI relate to income inequality in developed and developing countries before and after the second unbundling.
Section 2 explains the mechanisms behind FDI and the effects it could have on income inequality. Section 3 discusses the empirics and data which consists of 162 developed and developing countries ranging from 1970 to 2016. Section 4 provides the results and discussion of the regression analyses and concludes that, for developed countries, inward FDI relates to an increasing income inequality, whereas outward FDI relates to a decrease in income inequality after 1995. Furthermore, the inclusion of both types of FDI diminishes the specific effects of inward FDI. Section 5 briefly concludes the paper.
2. THEORETICAL BACKGROUND
2.1 Foreign Direct Investment
This paper takes a step back by first explaining why and how firms engage in FDI before it is discussed how FDI relates to income inequality.
Firms have four strategy options (Helpman, Melitz and Yeaple, 2004). The least productive firms leave the industry when their profits are negative. Low-productive firms choose to serve the domestic market while more productive firms serve both domestic and foreign markets, consequently, by exporting their products to increase their profits. Only the most productive firms within an industry choose to invest in foreign markets. This FDI decision is made when the trade costs are higher than the entire relocation of (a part of) the firm. Thus, when the advantages of relocating firms’ activities outweigh the advantages of economies of scale, firms decide to participate in FDI. Therefore, outward FDI is predominantly present for developed countries with highly productive firms, while the share of outward FDI in developing countries has increased as well during the late 20th century (Dunning, Narula, & Van Hoesel,
1996).
8
2.2 Effects of Foreign Direct Investment
One of the main explanations for the distributional consequences of FDI is given by the North-South model by Feenstra and Hanson (1997). The model explains the effect of distribution considering a world economy with only North and South as countries. North and South were based on the view of North and South America. Both countries have skilled and unskilled labor whereas the wage differs between the two types of labor. South produces inputs that are relatively intensive in the use of unskilled labor, whereas North produces inputs that are relatively intensive in the use of skilled labor. The unskilled and relatively cheap labor in South is causing North to invest in vertical FDI. The model states that labor is mobile between the skill categories, which means that through education and training one can move from being an unskilled worker to a skilled worker. This mobility of labor implies that the supply of each type of labor is reacting to the relative wage (Feenstra & Hanson, 1997).
Now, imagine a shift from input production from North to South. When North decides to outsource lower skilled labor inputs production to the South, South needs to increase its skills in order to be able to produce these skilled inputs. Hence, relatively skilled labor in a host country can be relatively unskilled labor in the source country. This shift leads to an increase in relative demand for skilled labor in South. North has given away inputs of relatively lower skilled labor, which means that it will also see a skill intensity upgrade and therefore an increase in the relative demand for skilled labor (Feenstra & Hanson, 1997).
When the demand for skilled labor increases relative to unskilled labor, wages change too. If the demand for skilled labor increases, wages go up until the demand is fulfilled with supply and moves to the equilibrium. The demand for unskilled labor is relatively less, ergo, the wages of unskilled labor are lower than the wages of skilled labor. When wages go up for skilled labor, while wages for unskilled labor stay the same or even decrease, incomes will end up being unequally distributed due to the increasing difference of the wages.
Thus, vertical FDI from North to South leads to an increase in the relative demand for skilled labor in both North (source country) and South (host country). This results into higher wages for skilled labor and therefore, wage inequality rises in both North as South (Feenstra & Hanson, 1997).
9
line with the above reasoning of the North-South model where an increase in skill premium leads to a higher income inequality.
Thus, on the one hand, if a host country has a HQ of an MNE, it implies that inward FDI requires less high skilled labor than the country uses for its own HQ services which relatively decreases the demand of skilled labor. On the other hand, when the host country does not have a HQ, it implies that the inward FDI requires more skilled labor than the country is currently using which concerns an increase in demand for skilled labor. Hence, the former observes a decrease in income inequality, whereas the latter notices an increase in income inequality.
Additionally, literature suggests a non-linear relation between inward FDI and inequality (e.g. Figini & Görg, 2011; Herzer, Hühne, & Nunnenkamp, 2014). Hence, the income inequality increases and decreases again (Taylor & Driffield, 2005). As discussed, with an increase of FDI, the demand for skilled labor increases. However, the demand for skilled labor decreases afterwards due to the gained knowledge and learning processes. With the rising and falling demand, income inequality increases in the short run and decreases in the longer run (Herzer et al, 2014). The rising and falling demand relates to an inverted U-shape and is based on the endogenous growth model (Aghion et al, 1998). The model predicts that the wage for skilled workers will increase while the wage for unskilled workers falls towards zero due to the introduction of a general purpose technology (GPT) (Aghion et al, 1998). When introducing a new technology, firms will innovate using skilled labor. This will rise the demand and thus the wage for skilled labor, which will lead to income inequality. When firms are adjusted to the new technology, wage inequality decreases again because all firms have moved to this new technology, leaving the demand for unskilled labor at zero (Figini & Görg, 2011). In this case, inward FDI is the mechanism for introducing new technologies in a developing host country by acting as a role model for domestic firms. The theory might be applicable to developed countries as well as the inverted U-shape is empirically found for both developing countries (Figini & Görg, 2011) and developed countries (Chintrakarn, Herzer & Nunnenkamp, 2012).
In sum, there is theoretical ambiguity on how FDI relates to income inequality. Following Feenstra and Hanson (1997), the inequality of the host and source country increases because of the increased relative demand of skilled workers. This holds under the assumption that FDI is mainly vertical and FDI is only regarding developed source countries and developing host countries (Herzer et al, 2014). Additionally, the GPT model argues that host countries will follow an inverted U-shape by means of FDI-induced spillovers (Aghion et al, 1998). However, according to Markusen (1995), developed host countries’ inequality decreases since they already have high skilled labor from their own HQ services. Therefore, the relative demand of skilled labor decreases when a foreign plant enters.
2.3 A New Paradigm
10
paradigm represents the first unbundling that stands for the geographic separation of consumption and the production of goods. Hence, competition was between factories and sectors solely, while service workers were not taken into account (Baldwin, 2006). On the other hand, the new paradigm stands for trading ideas and being task focused. The new paradigm represents the second unbundling where the competition is between workers performing the same tasks in different nations. With the falling costs of communication, moving goods and moving people, the second unbundling also included service and more skilled workers. Hence, almost every task and person has direct international competition now (Baldwin, 2006).
The new paradigm can change the proposed effects of theoretical views regarding the relation between FDI and income inequality. For instance, the North-South model proposes that developed source countries have an increasing demand of skilled labor due to outward FDI. Today, outward FDI can be about cherry-picking, that is, employing the most talented high skilled people. However, when outward FDI is aiming for high skilled labor, the relative demand of skilled labor of the North will not increase. With a decrease in the demand of skilled labor, income inequality can reduce as well. Another example regarding inward FDI is the GTP model (Aghion et al, 1998). This model predicts an increase and decrease in income inequality. Different from the GTP model, empirical results with the most recent data saw solely a decrease (Figini & Görg, 2011) or an increase (Jaumotte et al, 2013) of income inequality in developed countries. However, while the data used in these papers is more recent, it still hardly covers the decades after the second unbundling. This results that the potential effects of FDI after the second unbundling are still unknown. In sum, the second unbundling may result into a different relation between FDI and income inequality since previous literature generally based their findings on theory and data before the second unbundling.
2.4 Effects of Inward FDI
Receiving FDI can have both positive and negative effects on income inequality for a developing host country. The effects of inward FDI depend on the absorptive capacity of the host country (Mihaylova, 2015). When a country has lower levels of human capital and economic development, FDI tends to increase the income inequality, whereas this effect diminishes when the degree of education and GDP per capita increases. In line with the GTP model of Aghion et al. (1998), when FDI leads to technological spill-overs, a host country can expect an increase and decrease of income inequality (Figini & Görg, 2011). Figini and Görg (2011) find that wage inequality increases in developing countries with inward FDI stock, but this effect diminishes with the further increases in FDI.
Besides a more equal income distribution, foreign direct investment could also lead to bankruptcy and job losses (Lee & Vivarelli, 2006). For instance, Tsai (1995) found that inward FDI was harmful for the income distribution in East and Southeast Asia. Thus, domestic firms can learn and catch up through an increase in FDI, but they can also be destroyed which will lead to an increase in income inequality that will remain unequal.
11
and technology of the multinationals. Countries can catch up by decreasing the relative demand of skilled labor when transforming from unskilled labor to skilled labor. However, it is unknown how and if countries adjust to these external FDI shocks (Goldberg & Pavcnik, 2007). Therefore, it is still unclear how many years one needs to consider to observe the potential inverted U-shape effect. Furthermore, it is unknown if the inverted U-pattern always occurs. For instance, several studies found a lack of the sectoral labor reallocation, which points out that labor mobility is not a given fact for developing countries. The lack of labor reallocation leads towards the incapability of finding empirical evidence for an inverted U-shape pattern for developing countries (Tsai, 1995; Gopinath & Chen, 2003; Jaumotte et al, 2013).
To sum up, there is little support that developing economies can cope with labor reallocation and thus the mobility of labor within sectors (Chiquiar, 2008; Goldberg & Pavcnik, 2007; Topalova, 2010). Some research saw an increase and decrease (inverted U-shape) of income inequality in developing countries (e.g. Figini & Görg, 2011), whereas Tsai (1995), Gopinath and Chen (2003), and Jaumotte et al. (2013) only find an increase in income inequality due to inward FDI. Thus, inward FDI positively relates to income inequality and might see a negative relation afterwards. To clarify, if FDI positively relates to income inequality, it means that an increase in FDI relates to an increase in income inequality. Two hypotheses are proposed as a result of the ambiguity of potential longer-term effects.
Hypothesis 1a: Inward FDI is positively related with income inequality within a developing country.
Hypothesis 1b: Inward FDI has an inverted U-shape relation with income country inequality within a developing country.
As for developed host countries, inward FDI could have different effects as well. Recall that a HQ of an MNE is considered to be most skill intensive and generally situated in developed countries (Markusen & Venables, 1997). Alongside, developed countries also have foreign plant operations (inward FDI). The inward FDI demands relatively less skilled labor, since the foreign plant is less skill intensive as the country’s own HQ. Therefore, inward FDI in a developed country leads to less income inequality (Markusen, 1995).
However, empirical evidence for the relation between inward FDI and income inequality does not show this effect exclusively as stated by Markusen (1995). Some evidence is found that income inequality decreases in developed countries like Ireland (Figini & Görg, 1999) and the United States (Chintrakarn et al, 2012). On the contrary, Taylor and Driffield (2005) find an increase in income inequality in the United Kingdom, since foreign investors were paying more wages in specific jobs and industries. This leads to wages bidding up within sectors and eventually into a larger difference in wages between industries. Arguably, the data of Taylor and Driffield (2005) only has a time span of 10 years, which might be too short to see the inverted U-shape pattern of increasing and decreasing inequality. Lastly, Blonigen and Slaughter (2001) were inconclusive whether inward FDI had a positive or negative effect on the distribution of income in the US.
12
this, previous research investigates the relationship between inward FDI and income inequality mostly for specific countries. Since these effects are contradicting, it remains inconclusive what the relation between inward FDI on income inequality is for developed countries in general. Therefore, in order to research this empirically, two opposing hypotheses are proposed. On the one hand, following the reasoning of Markusen (1995), we expect that the demand for labor related to the inward FDI is lesser skilled than the high skilled labor used by the host country’s own MNE’s. Therefore, the demand for skilled labor will relatively decrease and thus the income inequality will decrease as well. On the other hand, we assume that inward FDI has negative effects for the labor market, which leads to an increase in income inequality. A negative effect is, for instance, foreign firms paying higher wages which increases the income inequality (Taylor & Driffield, 2005). In a more formalized matter, the hypotheses for developed countries are:
Hypothesis 2a: Inward FDI is negatively related with income inequality within a developed country.
Hypothesis 2b: Inward FDI is positively related with income inequality within a developed country.
2.5 Effects of Outward FDI
For developed countries, the expansion of domestic firms’ operations abroad and therefore the outsourcing of production stages to predominantly low-income countries is raising concerns about the consequences this could have for not only the host country but also for the source country (Becker et al, 2005). The expansion of domestic operations is mainly based on reducing production costs. Besides cost considerations, market expansion is a main motive for outward FDI (Becker et al, 2005). Market expansion is a form of horizontal FDI, implying that activities at the home country are independent of activities in the host country (Head & Ries, 2002). The firms that operate globally are therefore present in both developed and developing countries. The effects of outward FDI also depend on the developments across industries and firms in a country. For instance, the effects of income inequality depend on the firms’ ability to cope with productivity, the skill intensities and whether the scale of domestic operations in- or decreases due to outward FDI (Herzer & Nunnenkamp, 2013). This makes the theoretical framework for outward FDI ambiguous.
13
Alongside, most papers generally do not consider outward FDI effects since they mostly focus on inward FDI. Only a few papers empirically investigated outward FDI in relation to income inequality (e.g. Gopinath & Chen, 2003; Jaumotte et al, 2013; Herzer & Nunnenkamp, 2013). First, Gopinath and Chen (2003) identify an increase in income inequality with outward FDI. Second, Herzer and Nunnenkamp (2013) find that the effects of outward FDI first increased but eventually decreased income inequality in European countries. Lastly, research by Jaumotte et al. (2013) found an increase in income inequality within developed countries because of the reducing employment opportunities in the relatively lower skill sectors. An overview of the papers that empirically examines inward/outward FDI and income inequality used in this paper can be found in table A1 in the Appendix.
Developing countries were generally not involved in outward FDI up until 1995 (Gopinath & Chen, 2003). This may be because of the first unbundling, where FDI was still about trading jobs within sectors and aiming for cost reductions instead of market expansion. Since developing countries can get less cost advantages to outsource labor abroad, it was not common for developing countries to engage in FDI (Dunning et al, 1996). Therefore, we expect that developing countries are less engaged in FDI before the second unbundling. Due to the expected differences concerning the effects and motives of FDI, the unbundling periods are separated in the next hypotheses.
Hypotheses 3a and 3b are solely for the period before the second unbundling. We propose contradicting hypotheses since the findings are inconclusive regarding the effects of outward FDI on the source country’s income inequality. Furthermore, since developing countries were not often engaged in outward FDI before 1995, the hypotheses are regarding developed countries. For hypothesis 3a, we argue that FDI can lead to a decrease in income inequality. Following Head and Ries (2002) who state that outward FDI may have diminishing effects of skill intensity, we state that the skill intensity in the source country together with the diminished skill upgrading eventually leads to skill downgrading. Skill downgrading consequently leads to less demand differences between high and low skilled workers, ergo, less income inequality. Hypothesis 3b follows the reasoning of the North-South model (Feenstra & Hanson, 1997) where outward FDI is related to an increase in inequality due to the increased skill intensity of the source country. In a formalized way:
Hypothesis 3a: Outward FDI is negatively related with income inequality within a developed country before the second unbundling.
Hypothesis 3b: Outward FDI is positively related with income inequality within a developed country before the second unbundling.
14
employment opportunities in lower skilled labor (Jaumotte et al, 2013). Now, high skilled labor is affected too, which equalizes the employment opportunities for all types of labor. Therefore, low-skilled workers are not harmed more than high-skilled workers (Baldwin, 2006). Outward FDI of a developed country towards a similar level host country leads to a lower demand of high-skilled labor, since skill-intensive activities are relocated abroad. This will lower the skill intensity in the source country. The diminished skill upgrading eventually leads to skill downgrading in the source country (Head & Ries, 2002). Recall, when a rise in skill intensity increases income inequality (Feenstra & Hanson, 1997), skill downgrading can stop the rise of skill intensity and thereby narrow the income distribution. The knowledge gained from outward FDI also allows the country to increase its productivity and average wages, which leads to a decrease in inequality (Herzer & Nunnenkamp, 2013). Therefore, we expect that outward FDI of a developed country directed towards more similar countries is related to less inequality within the source country. In a formalized matter:
Hypothesis 4: Outward FDI is negatively related with income inequality within a developed country after the second unbundling.
As for developing countries, hardly any research is done for the effects of outward FDI and income inequality. The share of outward FDI of developing countries is lower than developed countries, but has increased during the late 20th century (Dunning et al., 1996). Since 2000, there has been a growth in outward FDI in especially Asian countries (Dunning et al., 2008). This growth happened since global markets integrate even more with the current wave of globalization (second unbundling). Growth in a developing country is reached by an increase of skill intensity which results into an increase of competitiveness (Dunning et al., 2008). For instance, being competitive can be reached via cost reduction focused vertical FDI where relatively low skilled labor is outsourced to lower income host countries. However, an increase in skill intensity is connected to an increase in income inequality according to the North-South model (Feenstra & Hanson, 1997). This is in line with the reasoning as stated for hypothesis 3b. Thus, it is expected that outward FDI of developing countries will increase their skill intensity which will also increase income inequality. More formally:
Hypothesis 5: Outward FDI is positively related with income inequality within a developing country after the second unbundling.
3. METHODOLOGY AND DATA
15
distinguishes between developed and developing countries by means of the country classifications of the United Nations (2014). The classification of countries is based on their development by measuring per capita gross national income (GNI). Besides a broad country range, a large time frame is necessary since the latter hypotheses are highlighting the effects before and after the second unbundling. Especially from the mid 1990’s, the international outsourcing of the new paradigm took off which encourages firms to outsource more and more stages of production (Dunning et al, 1996; Baldwin & Evenett, 2015). Therefore, 1995 is chosen as a starting point of the second unbundling. In total, we have an unbalanced dataset of 36 developed countries and 128 developing countries with data ranging from 1970 until 2016.
3.1 Dependent and Independent Variables
The dependent variable of this study is income inequality. Income inequality within a country is measured with the Gini and is mostly used in previous studies to measure income inequality. The Gini index ranges from 0 to 100 where 0 means perfect equality and 100 perfect inequality. Data for the Gini indexes is from the Standardized World Income Inequality Database (SWIID). SWIID is one of the richest datasets on income inequality (Solt, 2016) and used by scholars like Mihaylova (2015). SWIID is based on the World Income Inequality Database (WIID) and supplemented by other sources. The database consists of country and year estimates of summary measures of income distributions, especially the summarized Gini coefficient. Furthermore, compared to WIID, SWIID covers 12 more countries and a broader year range (Jenkins, 2015). Thus, SWIID is a combination of several previously popular datasets to create a bigger coverage and comparability than other datasets. A list of the countries and their summary statistics concerning the Gini is available in Table A3 in the Appendix. In this table, each country is listed together with the total amount of observations for that particular country, the mean of the Gini and the minimum and maximum of the Gini. For instance, Australia has 45 years of observations, an average Gini of 30.44 and a minimum of 27.3.
Our independent variables are the shares of inward and outward FDI in the country. There are two measures of FDI, namely FDI stocks and FDI flows. FDI stock is the value of the share of capital from the parent enterprise together with the net debt of affiliates to the enterprise. FDI flows are the transactions recorded during the year accounted for, whereas the FDI stocks show the accumulated value held at the end of the period (here in years). This research focusses on FDI stocks since this measure captures the long-run effects more effectively than the annual FDI flows (Figini & Görg 2011; Herzer & Nunnenkamp, 2013)
16
of GDP is based on current U.S. dollars. This means that the prices are valued based on the currency for that particular year and are not corrected for price inflation or deflation effects. By including year fixed effects in the model, we partly solve this problem. While constant prices could be argued as a preferable option since this measure does consider price inflation effects, consistency throughout the dataset is most important. Therefore, all measures are taken in current U.S. dollars.
Thus, since we are interested in the effects of both inward and outward FDI, we have two independent variables Inward FDI stock and Outward FDI stock. The independent variables are measured as the share of GDP of the particular country. Although FDI stocks are chosen to be the best capturer of long-run effects, FDI flows is analyzed as well. Examining the FDI flows, as an alternative measure of FDI, accounts as a robustness check for the findings (Herzer & Nunnenkamp, 2013).
3.2 Control Variables
The literature includes several control variables when researching the relations between FDI and income inequality. The control variables that are included in this study are discussed now.
GDP per capita is the GDP divided by the midyear population of the country and
reflects the level of development of a country. Economic development is playing a role in the income inequality within a country, and therefore it is necessary to control for GDP. Including GDP per capita makes sure that FDI does not show the impact of the level of economic development on inequality solely (Figini & Görg, 2011). The effect of economic development is hypothesized to be in line with the Kuznets’ inverted U-shape (Kuznets, 1955). Kuznets (1955) hypothesizes that income inequality increases at the early stages of development but declines when a certain stage of development is reached. The increase in income inequality is more noticeable due to the increase in GDP per capita (Mihaylova, 2015). Therefore, an increase in GDP is first related to an increase in income inequality. This first control variable is acquired from the World Development Indicators from the World Bank and is measured in current US$. Concluding, GDP per capita positively relates to income inequality.
Trade openness must be considered because predominantly, trade and technology are
seen as the most common explanations for wage inequality (Taylor & Driffield, 2005). Trade can be seen as the trade openness of a country. Trade openness is the sum of exports and imports of goods and services measured as a share of the gross domestic product of the country. Increased trade openness in a developing country will lead to an increase in wages of the abundant low-skilled labor compared to the high-skilled workers and as a result a reduction in income inequality (Jaumotte et al, 2013). The variable is measured as trade relative to GDP and is extracted from the World Development Indicators from of the World Bank. Trade openness is expected to be negatively related to income inequality.
Technology is another driver of income inequality (Jaumotte et al, 2013). Technological
17
Indicators of the World Bank. Unfortunately, World Bank solely had data from 1989 onwards. We expect that technology is positively related to income inequality.
Education is added as a control since an increase of education and access to education
can help inhabitants of a country cope with the changes that come along with globalization and technical change (Jaumotte et al, 2013). Education is needed to induce the transmission from lower to higher skilled labor and consequently the equalizing of income inequality within a country. This variable is derived from the World Bank of the Statistics Indicators. Education measures the ratio of people enrolled in secondary education to the total population. More enrolment of secondary education should mean higher supply of skilled labor and consequently less income inequality. Thus, education is expected to be negatively related of income inequality.
Corruption is the last control in this paper and is explained in more depth since this
control is generally not taken into account in this literature. Corruption is seen as a measure to research the supporting institutional system of a country since corruption is interfering with the core functions of the government such as the allocation of resources, stabilization of the economy and the redistribution of income (Gupta Gupta, Davoodi, & Alonso-Terme, 2002). Therefore, corruption has a negative effect on income inequality. So far, only a few studies try to link corruption to income inequality (e.g. Li, Xu, & Zou, 2000; Gupta et al, 2002; Gyimah-Brempong, 2002). For instance, Gyimah-Brempong (2002) finds that African countries are coping with high-income inequality and suggests that income inequality can be improved by reducing corruption in the country. More general, an increase in income inequality for developing countries is seen if corruption is present (Gupta et al, 2002). Furthermore, the importance of having a supporting institutional system or corruption in connection to the relation between FDI and income inequality is hardly researched. Corruption may also be linked to FDI since, the higher the corruption in a country, the harder it is to conduct business and thus the less appealing inward FDI is. Therefore, due to the previous findings in literature, corruption is taken into account in this paper to control for the role this could have on income inequality. Corruption is measured by means of the Corruption Perception Index (CPI) of Transparency International. This is a score per country per year on how corrupt the public sectors of a country are perceived to be. The score is given by MNE managers who have experience in doing business in that particular country. A country is the least corrupt with a score of 100. Thus, when the corruption of the country increases, the variable decreases. We expect that the higher the corruption, the higher the income inequality. Therefore, this corruption measure is expected to be negatively related to income inequality. Since a proper dataset is not online for the corruption perception index, all data are manually taken from the website of Transparency International to assemble a dataset. The data is available from 1995 onwards and can thus only be taken into account for the latter part of our study.
3.3 Descriptive Statistics
18
twice as large as the mean of outward FDI stock. This difference emphasizes that, in general, inward FDI is a larger part of a countries’ GDP share than outward FDI. Fourth, the FDI flows have lower shares than FDI stocks, but also show that inward FDI flows are twice as large as outward FDI flows. Recall, FDI stocks are an accumulation of FDI flows which explains the share differences between both measures. Finally, the standard deviation for most variables is larger than the mean, indicating that the data has a wide range of observations. This might suggest that some variables are not normally distributed.
3.4 Estimation Method
For each hypothesis, several variants of equation 1 are used. In general, both inward and outward FDI are examined over a period of time in order to observe the possible effect FDI could have on a country. Examining outward and inward FDI together is unique in the literature. Previous research (e.g. Figini & Görg, 2011) acknowledges the possibility that outward FDI has an effect on income inequality as well. However, to our knowledge, no paper tests both inward and outward FDI in the same regression. Given the theoretical arguments and the empirical findings, the effects of outward FDI can amplify or diminish the effects of inward FDI. Therefore, this paper argues that outward FDI is linked to inward FDI in such a way that neglecting one of them can cause an incomplete view upon the effects of FDI on income inequality.
Additionally, we want to test if there is an inverted U-shape detectable for rising and falling inequality. Thus, following the theoretical discussion based on the GTP model in our previous section, it is assumed that income inequality can first increase but can decrease in the longer run. Therefore, the equation has additional squared variables of FDI which allow us to include the potential nonlinearity. The equation is depicted below.
𝐺𝑖𝑛𝑖𝑖𝑡= 𝛼 + 𝛽1𝑙𝑛𝐼𝐹𝐷𝐼 𝑠𝑡𝑜𝑐𝑘𝑖𝑡−1+ 𝛽2𝑙𝑛𝐼𝐹𝐷𝐼 𝑠𝑡𝑜𝑐𝑘𝑖𝑡−12 + 𝛽3𝑙𝑛𝑂𝐹𝐷𝐼 𝑠𝑡𝑜𝑐𝑘𝑖𝑡−1+
𝛽4𝑙𝑛𝑂𝐹𝐷𝐼 𝑠𝑡𝑜𝑐𝑘𝑖𝑡−12 + 𝛽5𝑙𝑛𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝜆𝑖+ 𝜆𝑡+ 𝜀𝑖𝑡 (1)
The model specifics are explained now. The dependent variable, income inequality, is the Gini, which is measured per country i in time t. The independent variables are IFDI stock
(Inward FDI stock) and OFDI stock (Outward FDI stock) and both variables in the squared term, IFDI stock2 and OFDI stock2, to account for non-linearity. All independent variables are
Table 1: Descriptive statistics of the variables
Variable Description Observations Mean Standard
deviation
Gini Gini index of inequality 4,738 38.06 8.55
Inward FDI stock FDI inward stock relative to GDP 4,101 37.88 100.36 Outward FDI stock FDI outward stock relative to GDP 3,438 19.27 68.55 Inward FDI flow FDI inward flows relative to GDP 4,557 3.76 13.79 Outward FDI flow FDI outward flows relative to GDP 3,536 1.84 10.20 GDP per capita GDP per capita in current US$ 4,576 8,797.40 14,462.13
Trade Openness Trade relative to GDP 4,411 78.10 54.17
Technology High-technology exports relative to all manufacturing
exports 2,895 10.45 12.43
Education Ratio of people enrolled in secondary education to the
total population 3,541 73.16 31.21
19
in natural logarithms, as will be explained later. The four variables are all measured per year (t
) and country (i ) individually with a lag of one year (t-1) to solve a potential endogeneity
problem. This is in line with previous research which aims to decrease potential endogeneity and causality issues with the use of lagged variables of FDI (Tsai, 1995; Figini & Görg, 2011; Herzer & Nunnenkamp, 2013; Herzer et al, 2014). One can also lag all variables on the right hand side of the equation with one year, but the differences were minimal between this and solely lagging FDI variables. Therefore, we choose to follow previous research. Next, 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡
are all controls as previously discussed in natural logarithms per year and country, 𝜆𝑖 and 𝜆𝑡 are
country and year fixed effects and 𝜀𝑖𝑡 represents the error term per year and country. The year fixed effects controls for effects such as inflation and deflation, whereas the country fixed effects controls for the country specific non-observed variables. Non-observed variables could be the country’s tariff regulation, population or geographic location.
4. EMPIRICAL RESULTS
The data is analyzed with the help of a fixed effect panel regression with robust standard errors. In order to use panel regression, certain assumptions have to be met. Therefore, several steps are taken to test each assumption before running the actual regressions. The first step is the detection of unusual and influential data. All merged datasets together resolved in a dataset of over 4700 observations. Since we are interested in effects over time, all countries, 10 in total, with only one observation were dropped from the dataset. Scatterplots of the data detected some outliers. To provide a general view on the real world, the data was not trimmed since each dropped observation is still a bit of information that one is neglecting otherwise. However, as a robustness check, data was trimmed in such a way to provide an almost perfectly balanced dataset. These results are taken as a third robustness check.
The next step is to detect for normality of residuals. As suggested in the descriptive statistics, most variables have a larger standard deviation than the mean which suggests that these are not normally distributed. As also shown in their histograms, most variables are indeed not normally distributed. The only variable which is already normally distributed is the dependent variable, Gini. To solve the normality issue, variables must be taken in natural logarithms (Herzer et al, 2014). Therefore, all variables are transformed to natural logarithms except for the Gini. A problem arises with taking logs of the inward and outward flows due to some negative observations in the variables. One cannot take logs of a negative observation and therefore these observations were dropped from the variables regarding FDI flows in order to transform them to natural logarithms and thereby create a more normal distribution of the variables. In total, 228 and 404 observations, respectively, were dropped from the approximate 4500 inward FDI flows and 3500 outward FDI flows observations in total. The FDI flows are used in the robustness check, as discussed later on.
20
Next, multicollinearity is calculated by the VIF test (Mansfield & Helms, 1982). Using a threshold of 4, every variable is proven to not have issues with multicollinearity except for GDP per capita. We added all controls one by one and included this control on a later stage. In this way, we saw the possible effects GDP per capita has on the outcomes. The problems with multicollinearity are not too severe since the variable does not get omitted when added. As for inward and outward FDI, Herzer and Nunnenkamp (2013) saw a high correlation between both types of FDI stock and therefore decided not to include this simultaneously in the regression models to avoid collinearity problems. However, with the natural logarithm of the variables, the correlation between both types of FDI is reduced which means that simultaneous inclusion in a regression is possible. Therefore, we were able to include both in the same regression.
Lastly, the assumption for linearity is solved by allowing for non-linearity with the squared terms of both inward and outward FDI. The assumption checks resulted in some amendments of the model. The regression models are specified like described in the previous section and most independent variables are added separately to detect the effects of the variables more specifically.
4.1 Hypotheses Testing
The results for hypotheses 1 and 2 are seen in Table 2, hypothesis 3 in Table 3 and hypotheses 4 and 5 in Table 4. Every table starts with the regression of the main control variables. Two controls are added for the final two hypotheses since we only have data on the controls corruption and technology from 1995 and 1989 respectively. Each next column in the table adds an independent variable to show the strength of the influence on the particular independent variable.
21
Columns 6 to 10 of Table 2 try to clarify the ambiguousness of the literature by determining whether the share of inward FDI of a developed country is related to a decrease (hypothesis 2a) or increase (hypothesis 2b) of income inequality. For each column, regardless of the inclusion of additional variables or outward FDI, inward FDI has a positive and significant effect upon income inequality (p<0.01). With an adjusted R-square of 0.426, the last column (10) has the best explanatory value. The positive coefficient of 0.947 indicates that an increase in the share of inward FDI in a developed country is related to an increase in income inequality, supporting hypothesis 2b. Since all variables, except for the Gini, are measured in logs, the coefficient cannot be interpreted directly. Now, every percentage change of the variable is given instead of every unit change. To know the actual change of the Gini, one needs to multiply a coefficient with the log of the wanted percentage change. For instance, since the inward FDI stock variable has a coefficient of 0.947, a 10 percent increase in the share of inward FDI will increase the Gini with 0.947 times the natural logarithm of 1.1, thus with 0.039. This illustrates that although a significant relation is found, it is small. Additionally, in column 9, we see marginal evidence that outward FDI relates to a decrease in income inequality (p<0.1).
Table 2: Hypotheses 1 and 2
Panel Fixed Effect Regressions With Robust Standard Errors
Sample of Developing and Developed Countries solely. Dependent variable is Gini
Developing Countries Developed Countries
1 2 3 4 5 6 7 8 9 10 L.ln Inward FDI stock 0.204* 0.138 0.389 0.751*** 0.789*** 0.947*** (0.117) (0.225) (0.401) (0.184) (0.155) (0.170) L.ln Outward FDI stock 0.106 0.09 0.165 -0.142 -0.312* -0.077 (0.088) (0.091) (0.105) (0.154) (0.181) (0.216) L.ln Inward FDI stock2 -0.067 0.029 (0.087) (0.023) L.ln Outward FDI stock2 0.022* -0.108*** (0.012) (0.035) ln Education -0.511 0.751 0.996 1.061 1.307 -1.883 -0.008 -0.65 0.095 -0.016 (0.697) (0.797) (0.996) (1.034) (1.048) (1.726) (1.482) (1.474) (1.487) (1.505) ln Trade Openess 0.741 0.381 0.478 0.378 0.236 -0.813 -0.905 0.116 -0.586 -1.311 (0.700) (0.962) (1.140) (1.101) (1.067) (1.162) (0.896) (0.771) (0.861) (0.860) ln GDPPC 1.112 1.626** 1.151 1.162 1.061 0.763 -0.285 0.394 -0.15 -1.022 (0.685) (0.818) (0.958) (0.965) (0.945) (0.879) (0.714) (0.877) (0.709) (0.797)
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Country fixed
effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Constant 32.902*** 25.651** 28.070** 27.826** 28.148** 30.596*** 30.029*** 23.786** 27.442** 38.027*** (6.493) (9.878) (11.647) (11.781) (11.512) (10.210) (10.878) (11.595) (10.826) (10.898) R-squared 0.121 0.172 0.198 0.199 0.209 0.235 0.387 0.318 0.415 0.453 Adjusted R-squared 0.1 0.154 0.177 0.177 0.186 0.201 0.359 0.287 0.387 0.426 Observations 2169 1831 1538 1533 1533 1165 993 993 981 981 Notes:
(i) Columns 1-5 can answer hypothesis 1, Columns 6-10 can answer hypothesis 2 (ii) Significant at * p<0.10, ** p<0.05, *** p<0.01
22
Furthermore, with the significant and the negative sign of Outward FDI stock2, outward FDI is
related to a decrease in income inequality (p<0.01) in column 10.
Table 3 gives insights whether an increase in outward FDI of a developed country is related to a decrease (hypothesis 3a) or an increase (hypothesis 3b) of the income inequality within the country before the second unbundling. As previously discussed, these are the years before 1995. In Table 2, outward FDI had a marginal significant effect of decreasing income inequality for the entire timeframe of the dataset (column 9). Now, solely including the years up until 1995, only the squared term of outward FDI is marginally significant (p<0.1) indicating an eventual decrease in income inequality. Therefore, we cannot find support hypothesis 3a and hypothesis 3b.
Table 3: Hypothesis 3
Panel Fixed Effect Regressions With Robust Standard Errors
Sample of Developed Countries and Observations Before 1995. Dependent variable is Gini Before 1995
1 2 3 4 5
L.ln Inward FDI stock 0.380** 0.324** 0.657*
(0.162) (0.157) (0.337)
L.ln Outward FDI stock 0.209 0.321 0.481
(0.360) (0.526) (0.509)
L.ln Inward FDI stock2 0.044
(0.030)
L.ln Outward FDI stock2 -0.178*
(0.102) ln Education -2.357 -1.269 -2.306 -1.374 -1.47 (2.040) (1.761) (1.726) (1.838) (1.826) ln Trade Openess -0.109 -1.028 -0.839 -1.596 -2.283 (1.742) (1.948) (1.795) (2.188) (2.213) ln GDPPC 1.622 -0.73 -0.769 -1.118 -1.806 (1.423) (1.589) (1.555) (1.701) (1.738)
Year fixed effects Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes
Constant 23.311 41.144** 45.150** 46.577** 55.613** (14.642) (18.988) (17.429) (19.724) (20.623) R-squared 0.174 0.304 0.308 0.345 0.368 Adjusted R-squared 0.122 0.251 0.256 0.291 0.311 Observations 473 313 313 302 302 Notes: (i) Significant at * p<0.10, ** p<0.05, *** p<0.01 (ii) Standard errors in second row
(iii) R-squared is the within R-squared
23
Table 4 provides insights for hypotheses 4 and 5 and is similar to Table 2, but with data solely after 1995. The controls technology and corruption are included now as the data for these controls was available from 1995 as well. Hypothesis 4 argues that an increase in outward FDI is related to a decrease in income inequality within a developed country after the second unbundling. Therefore, hypothesis 4 is based on the results on the right half of Table 4. Since column 9 is showing a negative coefficient on outward FDI, we find evidence for the hypothesis. Thus, the findings suggest that outward FDI is related to a decrease in income inequality within a developed country (p<0.05). However, the hypothesis is only supported in column 9 and not throughout all columns. The effect of outward FDI becomes insignificant when it is measured solely or when the squared terms of both types of FDI are added. Therefore, we can only partly support hypothesis 4. Additionally, inward FDI has a positive and significant effect when there is controlled for outward FDI as well (columns 9 and 10). Besides this, technology seems to have a positive and marginally significant effect on income inequality (p<0.1). This is in line with the theoretical arguments that more technological change will increase the difference between high and low-skilled wages and thus increase income inequality.
24
Table 4: Hypotheses 4 and 5
Panel Fixed Effect Regressions With Robust Standard Errors
Sample of Developing and Developed Countries after 1995. Dependent variable is Gini
Developing Countries Developed Countries
1 2 3 4 5 6 7 8 9 10 L.ln Inward FDI stock 0.482 0.319 -0.706 0.312 0.621*** 1.205** (0.386) (0.428) (0.831) (0.206) (0.217) (0.549) L.ln Outward FDI stock 0.194 0.146 0.105 -0.244 -0.463** -0.374 (0.132) (0.135) (0.158) (0.206) (0.204) (0.248) L.ln Inward FDI stock2 0.206 -0.077 (0.152) (0.083) L.ln Outward FDI stock2 -0.019 -0.005 (0.041) (0.052) ln Education 1.761* 1.880* 1.697 1.773* 1.625 -1.411 -1.202 -1.448 -1.116 -1.35 (0.967) (0.967) (1.041) (1.045) (1.074) (1.003) (0.976) (0.953) (0.900) (0.894) ln Trade Openess -0.206 -0.785 -0.465 -0.583 -0.471 -0.136 -0.455 0.039 -0.039 -0.244 (0.732) (1.463) (1.533) (1.579) (1.563) (0.736) (0.797) (0.796) (0.824) (0.911) ln GDPPC -0.016 -0.107 -0.344 -0.315 -0.18 -0.459 -0.563 -0.426 -0.377 -0.259 (0.641) (0.812) (0.836) (0.851) (0.859) (0.748) (0.725) (0.715) (0.648) (0.859) ln Technology 0.06 0.045 0.068 0.055 0.047 0.506 0.542 0.648* 0.608* 0.533 (0.062) (0.059) (0.078) (0.075) (0.074) (0.370) (0.341) (0.367) (0.323) (0.318) ln Corruption 1.204* 1.199* 1.165* 1.148* 1.066 -0.438 -0.391 -0.385 -0.376 -0.531 Year fixed
effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Country fixed
effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
(0.670) (0.662) (0.696) (0.682) (0.715) (1.025) (0.996) (0.958) (0.805) (0.825) Constant 31.919*** 33.478*** 35.834*** 35.148*** 35.520*** 41.280*** 41.557*** 40.279*** 37.388*** 37.993*** (8.770) (11.473) (12.108) (12.106) (12.633) (9.801) (9.541) (9.456) (8.596) (10.075) R-squared 0.3 0.319 0.324 0.329 0.341 0.174 0.189 0.191 0.235 0.246 Adjusted R-squared 0.281 0.299 0.303 0.307 0.317 0.141 0.154 0.157 0.202 0.211 Observations 967 955 883 883 883 676 669 668 668 668 Notes:
(i) Columns 1-5 can answer hypothesis 4, columns 6-10 can answer hypothesis 5 (ii) Significant at * p<0.10, ** p<0.05, *** p<0.01
(iii) Standard errors in second row
(iiii) All controls included, timeframe 1995-2016 (iiiii) R-squared is the within R-squared
4.2 Robustness Check
Three different robustness checks are done to support the results, namely using different year lags, using FDI flows instead of stocks and using a more balanced and smaller dataset. All corresponding tables for the robustness checks are included in the appendix.
25
years. To stay close to previous research (e.g. Figini & Görg, 2011), only a lag of one year is chosen to show in the tables.
Second, the same analyses are done with FDI flows instead of stocks. As discussed previously, taken logs is necessary to meet the normal distribution assumption. Since logs cannot be taken from negative observations, all negative observations are deleted. Table A4 shows the results for the first two hypotheses. Recall, for developing countries, it is found that inward FDI stock is related to an increase in income inequality. Therefore, hypothesis 1a is partly supported for FDI stock while hypothesis 1b cannot be supported. However, for inward FDI flows, one can find support for neither hypothesis 1a nor hypothesis 1b due to the lack of significant results. This shows that the partly found evidence for the effect of inward FDI stock on income inequality is not robust over the two different measures of FDI. Table A4 only shows a positive and marginally significant effect of GDP per capita on income inequality for developing countries. This finding is in line with the expectations of the control.
For developed countries, inward FDI flow has a positive and marginally significant effect on income inequality throughout columns 7-10 of Table A4 (p<0.1). This result is in line with the results of inward FDI stock, supporting hypothesis 2b. Therefore, the support for hypothesis 2b is robust over both inward FDI stock and inward FDI flows. Outward FDI flows have a negative and significant effect on income inequality for developed countries. The latter two columns even show a negative effect with a one percent significance level which indicates that outward FDI flows are related to a decrease in income inequality. This is elaborated later when the hypotheses regarding outward FDI are discussed.
Table A5 shows significant results that support hypothesis 3. The preferred measure of FDI, FDI stock, could not find support for the hypotheses concerning a relation between outward FDI and income inequality for developed countries before 1995. However, the results of outward flow do show a negative significant effect (p<0.05). This illustrates that an increase in outward FDI is related to a decrease in income inequality which would support hypothesis 3a. However, this cannot be said without precaution, since 200 negative observations were deleted from the sample. A negative observation of FDI flows means that in some cases, there were more inward than outward FDI flows, leading to a negative net outward FDI flow. Thus, although Table A5 would support hypothesis 3a, the results are not robust for both FDI stocks and flows.
26
As for developing countries, columns 6 to 10 of Table A6, outward FDI flow does have a positive effect on the Gini. The results based on columns 6 to 10 suggests that an increase in outward FDI flows is related to an increase in income inequality. This finding supports hypothesis 5. However, we were not able to find support for hypothesis 5 with outward FDI stocks. Therefore, hypothesis 5 is marginally supported only if outward FDI flows are used (p<0.1).
In sum, using FDI flows gave some contradicting or additional support for the hypotheses. For instance, with FDI flows, hypothesis 3a and 5 could be supported while hypothesis 1a, 1b, 2a, and 4 could not. Therefore, the partial support of hypothesis 1a and 4 in the analysis with FDI stocks are not robust. The only robust finding is hypothesis 2b since this hypothesis is supported with the use of both FDI stocks as FDI flows. As discussed before, however, FDI stocks are embracing the longer run effects better and therefore this measure is preferred. Additionally, due to the use of logs, it was necessary to delete some observations which might affect the results.
The third robustness check is based on smaller datasets. First, data was trimmed at a 1 and 99 percent percentile and at a 5 and 95 percent percentile to detect the potential differences with the unbalanced dataset. However, the results were similar to the untrimmed data. Another approach was a further reduction of the dataset by aiming for a balanced set of countries and years. First, the dataset was divided into developed and developing countries. Next, all observations with missing values for the independent variable, FDI, were dropped from the database. Then, to get a balanced dataset, all countries were dropped which did not have all observations from the selected years, in both cases 1980-2015. These years were chosen to maximize the time range while still keeping the most countries in the dataset. In total, 13 developed countries and 14 developing countries are analyzed. Three tables corresponding to this robustness check are shown in the appendix (Table A7, A8 and A9).
27
was already evident for these developing countries. Therefore, income inequality decreases as well with a further rise of GDP per capita.
Columns 6 to 10 of Table A7 show the results for developing countries after the second unbundling threshold of 1995 (hypothesis 5) and can be compared to column 1 to 5 of Table 4. Both the general analysis and this robustness check fail to find support for hypothesis 5, since no significant effects are found for outward FDI for developing countries. Alongside, trade openness and GDP per capita still have a significant negative effect on income inequality. Furthermore, inward FDI stock appears to be weakly significant (p<0.1) when taken together with outward FDI stock. However, this effect disappears when one controls for potential non-linearity which shows that the effect of inward FDI is not strong. Thus, the hypotheses concerning developing countries (1a, 1b and 5) cannot be supported which is mostly in line (except for 1a) with the main analysis.
Table A8 provides the results on developed countries for the total range of 1980-2015 and for 1980-1994. In the main analysis, hypothesis 2b is partly supported whereas hypothesis 3 cannot be supported. This robustness check finds partial evidence to support hypothesis 2b since only column 5 of Table A8 shows a significant and positive effect of inward FDI stock on the increase in income inequality. Therefore, the support of hypothesis 2b is only partly robust. The latter five columns of Table A8 show again that hypothesis 3 cannot be supported.
Lastly, Table A9 shows how outward FDI relates to income inequality for developed countries after the second unbundling threshold. This provides as a robustness check for hypothesis 4. Column 5 shows a negative significant effect of outward FDI stock on income inequality (p<0.05). This indicates that outward FDI stock is related to a decrease in income inequality, providing partial support for hypothesis 4. This partial support is in line with the previous findings and therefore the support for hypothesis 4 is robust in the smaller dataset. Furthermore, marginal evidence is shown for the U-shape effect indicating that the income inequality will increase again in the long run (p<0.1). Besides, trade openness and corruption both have a significant effect on income inequality. Trade Openness is positive and marginally significant in the final column (p<0.05), which suggests that trade openness of a developed country is related to the increase in income inequality. This is contrary to the expectations by previous scholars. However, it is in line with the reasoning of Herzer and Nunnenkamp (2013), who argue that an increase in trade relates to an increase in wage inequality for countries with abundant skilled labor. Hence, abundant skilled labor is generally related to developed countries. Additionally, corruption has a negative and marginally significant relation with income inequality (p<0.1). Thus, when the corruption measure increases, the country is less corrupt in its perception which relates to a decrease in income inequality which is in line with previous expectations.
28
4.3 Discussion
Most importantly, the paper finds partly supports for hypothesis 2b and hypothesis 4. Hypothesis 2b is supported and is partially robust to the robustness checks. Hypothesis 4 was partly supported, and partly robust to the third robustness check (smaller and balanced dataset). Therefore, the main findings are that, for developed countries, inward FDI is related to an increase in income inequality and outward FDI after 1995 is related to a decrease in income inequality. Hypothesis 2b is supported with the coefficient of inward FDI of 0.947 and hypothesis 4 is partly supported with the coefficient of outward FDI of -0.463. The interpretation of the coefficients is less straight forward because of the use of natural logarithms. Hence, a 10 percent increase in inward FDI is related to an increase in 0.039 of the Gini whereas a 10 percent increase in outward FDI is related to a decrease in 0.019 of the Gini. This shows that the relation, although significant, does not have a large effect on income inequality.
Where the effect of inward FDI (hypothesis 2b) is also found by some scholars, (Taylor & Driffield, 2005; Gopinath & Chen, 2008; Jaumotte et al, 2013), the effect of outward FDI (hypothesis 4) is not found previously. Only three papers investigated the relation between outward FDI and income inequality and saw either an increase (Gopinath & Chen, 2008; Jaumotte et al, 2013) or an increase and decrease (Herzer & Nunnenkamp, 2013). We argue that the effect of outward FDI changed with the second unbundling due to the increased competitiveness for different tasks and jobs. Where FDI was first mainly between developed and developing countries, FDI is directed towards similar countries more and more as well. Inward FDI is now also directed towards services and high skilled labor. Since these changes of FDI possibilities occurred roughly after 1995, we argue that the effects of FDI differ as well. Literature could not foresee these changes of FDI possibilities since the previous findings are generally based on data before the 21st century. Moreover, to our knowledge, this paper is first to link the second unbundling with the effects of income inequality. To conclude, the paper finds that outward FDI is related to a decrease in income inequality and argues that this can be explained due to the changes in labor mobility and FDI opportunities connected to the second unbundling. Therefore, we take the first step in a new development for both theoretical reasoning as empirical findings.
Next, we are not able to find an inverted U-shape for the relation between FDI and income inequality. Since our dataset is rather extensive with the amount of years and countries taken into account, we suspect that the found inverted U-shapes by previous research are country or region specific. We did, however, find that inward FDI is related to an increase of income inequality solely. In order to decrease income inequality again, policy makers should be focused on education to enhance the skills and technological improvement necessary to keep up with the foreign firms.
