FDI effects on economic growth in Sub-Saharan
Africa: a comparison between low-income and
middle-income countries
MSc Economics
International Economics and Globalization Track
Yordanka Temelkova
04.07.2016
Student number: 11088699
Supervisor: Dr. K.B.T. Thio
2
Statement of Originality
This document is written by Yordanka Temelkova 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.
3
Abstract
The aim of this paper is to analyze the effect of FDI on economic growth in Sub-Saharan Africa and examine eventual differences due to income level by conducting a fixed effects regression model, utilizing data for 32 countries in the region for the period from 1995 to 2014. Inflation rate, population growth, initial level of GDP, average years of schooling, political stability, investment, government consumption and trade are control variables. The results indicate that FDI is an important growth driver. Also, macro stability, trade openness and the interaction of FDI with education are necessary conditions for growth. No significant difference between low-income and middle-low-income countries in the region in their effect on GDP growth and interaction with FDI is found. The robustness check confirmed the importance of FDI in Sub-Saharan Africa, but casts some doubts about the validity of the other conclusions.
Key words: economic growth, FDI, low-income countries, middle-income countries, Sub-Saharan Africa
4
Table of contents
1. Introduction ... 5
2. Literature review ... 10
2.1. FDI and growth worldwide ... 10
2.2. FDI and growth in SSA ... 13
3. Empirical analysis ... 15
3.1. Theoretical background ... 15
3.2. Methodology ... 18
3.3. Data ... 21
3.4. Results and discussion ... 23
4. Robustness check ... 28
5. Conclusions ... 32
References ... 34
5
1. Introduction
The economies of Sub-Saharan Africa (SSA) have been growing steadily since 1994: the total GDP reached $492 billion in 1995, $769.5 billion in 2005, continued to grow even during the world crisis with $943 billion in 2009 and reached $1170.7 billion in 2014. Currently the region is undergoing a rapid population growth and according to the forecasts, by 2035 the working age population of SSA will outnumber that of the rest of the world combined (Regional Economic Outlook, 2015). Despite the fact that the population in the region is increasing quickly, the GDP per capita is rising because the GDP is growing at a faster rate (figure 1).
Figure 1
Source: Graph constructed by author based on UNCTAD Data.
Moreover, as shown by figure 2, the growth rate in SSA was much higher after 2000 and apparently the region has begun to catch up with the rest of the world. After 2001 the GDP growth rate in SSA countries surpassed the growth rate in Europe, America and Asia (except for 2010, 2011 and 2012 when Asia had a higher rate), but at the same time the region remains the
6 poorest one in the world. Despite experiencing declining poverty rates, the number of poor people continues to rise - it is one third of the world’s extreme poor (World Bank, 2015). Therefore, understanding what is driving the recent economic growth in SSA is of great importance.
Figure 2
Source: Graph constructed by author based on UNCTAD Data.
Foreign direct investment (FDI) is one of the possible growth drivers because it is a key source of external finance for the region. FDI inflows have increased significantly during the last two decades due to the improved investment climate, governance and market access initiatives (World Investment Directory, 2008). In 1995 the inward FDI flows amounted to $4.441 billion, in 2009 they were $37.97 billion, falling down to $30.39 billion in 2010 (because of the global crisis) and recovering quickly, reaching $42.95 billion in 2014. Even the least developed countries in the region had a significant surge in FDI inflows (figure 3).
7 FDI flows increased by 5% from 2014 to 2015 in SSA while other regions were faced with a decline, for example Latin America and the Caribbean (Global Investment Report 2015, UNCTAD). Moreover, the intraregional FDI increased significantly – from less than 10% in 2003-2008 to 18% in 2009-2013 (Regional Economic Outlook, 2015). Although FDI in SSA has risen, its share in the worldwide FDI and in the developing countries’ FDI remains small. In 2014, for example, the FDI in SSA accounted for only 5.46% of the FDI in the developing countries.
Figure 3
Source: Graph constructed by author based on UNCTAD data
The positive impact of FDI on economic growth is associated with a transfer of technologies and skills, an increase in employment, competition and access to global markets (OECD, 2001). However, FDI is a part of the globalization process and as such is being opposed by anti-globalists who point at the possible negative effects: jeopardized national independence, more monopolistic and oligopolistic firms in the recipient country and social unrest (Moosa, 2002). That is why FDI’s impact on growth in the region is a significant, but yet a controversial topic.
8 The empirical results in the literature are not unanimous because some authors argue that there is a positive relation between FDI and GDP, others state that there is a positive impact only when certain conditions are present (a threshold for human capital, trade openness, etc.) and others find a negative effect.
According to the World Bank classification of countries by income, all of SSA countries are in the groups of low-income, lower-middle-income and upper-middle-income economies (except for Seychelles which is the only high-income economy in the region, but due to its small territory and the fact that it has the smallest population of any independent African state, most studies do not include it when a research about SSA is done. Here the same approach is followed. Also, Equatorial Guinea is not included since from 2007 it is considered a high-income economy). For the purposes of this paper, lower-middle-income and upper-middle-income countries are grouped in one category – middle income economies (see Appendix Table A12).
Furthermore, when one takes a closer look at the difference between low-income and income countries in regards to FDI, the bulk of FDI inflow was received by the middle-income economies: for the period of 1990-2014, they have benefited from 68% of FDI in the region (figure 4). In addition, South Africa and Nigeria account for 58% of the total inward FDI stock in the same period. Among the top 10 FDI recipients, only three countries are low-income countries: Mozambique, Liberia and United Republic of Tanzania. This is in line with Michalowski, 2012, who stated that the top five FDI recipients3 account for approximately 70% of total FDI inward stock in SSA (Michalowski, 2012). They are all middle-income countries. However, the proportion changes when the average FDI as a share of GDP for the same period is considered. Although low-income countries have received a lower amount of FDI, when it comes to FDI as a percentage of GDP, both economy categories have approximately the same percentage number, even low-income countries have a slightly higher percentage (figure 6). This indicates that FDI is an important finance source, thus possible economic growth driver in both income groups.
2 Table A1, Appendix lists the countries included in the regression. The countries that were dropped due to missing
observations are: Angola, Burkina Faso, Cabo Verde, Chad, Comoros, Djibouti, Eritrea, Ethiopia, Guinea, Guinea-Bissau, Madagascar, Nigeria, Sao Tome and Principe, Somalia and South Sudan.
9
Figure 4
Source: Graph constructed by author based on World Bank Data
Nevertheless, the studies so far group the countries by institutional quality, for example, but do not focus on the income differences. Therefore, there is still room for a further analysis and this paper’s research question is “What are the FDI effects on economic growth in SSA comparing low-income and middle-income countries?”. This may provide further insights and contribute to the state of knowledge.
The data set is obtained from the World Bank Data: World Development Indicators, Education Statistics (Barro-Lee Dataset) and Worldwide Governance Indicators, for the period 1995-2014 since this is the period when a surge in FDI was observed. A fixed effects regression model is performed based on the paper of Sukar, Ahmed and Hassan (2007), expanding their model by the inclusion of additional control variables and the use of a greater number of SSA countries. The dependent variable is GDP per capita growth, the independent one is FDI as a percent of GDP and the control variables are inflation rate, investment, among others.
10 The remainder of the thesis is structured as follows. Chapter II provides a review of the existing literature about the impact of FDI on economic growth, Chapter III explains the methodology, data and results, Chapter IV shows the robustness check while the conclusions are presented in Chapter V.
2. Literature review
In the last few decades, the initial anti-FDI views in the developing countries have gradually changed to favorable ones since people started to realize the positive technology effects and increased possibilities for the local markets associated with FDI (new knowledge, greater demand for local goods, etc.). Therefore a lot of countries have introduced policy measures aimed at attracting FDI (Aoki and Todo, 2008). This inevitably has led to a significant interest in testing empirically the relation between FDI and economic growth worldwide (see 2.1.) and in SSA, in particular (see 2.2.). Therefore, this chapter aims to highlight some of the papers that explore the impact of FDI on economic growth, focusing on the different models used and the obtained results and conclusions.
2.1. FDI and growth worldwide
When FDI increases GDP, this leads to a larger market and in turn, it attracts more FDI (Li and Liu, 2004) or market reforms in host countries can affect GDP and FDI at the same time (Makki and Somwaru, 2004), which means that an endogeneity problem exists and different econometric models are used to avoid it. All authors discussed here have used as an independent variable either FDI to GDP ratio or FDI levels and as a dependent one GDP level or growth, but only some of them have found a positive effect of FDI on growth. In most of these papers, FDI stimulates economic growth when certain conditions in the host country are present, meaning that the interaction terms of FDI and given variables have statistically significant positive coefficients.
Borensztein et al. (1998) found that FDI increases GDP when a minimum threshold level of human capital4 is present, after controlling for initial income, government consumption, political instability (measured as political assassinations per capita per year and a dummy for
4
Proxied by level of educational attainment – average years of male secondary schooling constructed by Barro and Lee (1993).
11 wars on the national territory) and quality of institutions. The utilized panel data is for 69 developing countries for the period 1970-1989 and a seemingly unrelated regressions technique (SUR) was conducted to estimate the regressions. In addition, an instrumental variable (IV) estimation with lagged values of FDI, a lag value of total GDP and other variables5 as instruments was used in order to control for endogeneity problems. However, the results were similar to those obtained previously. Jyun-Yi and Chih-Chiang (2008) used cross-sectional data for 62 countries for the period 1975-2000, performing three instrumental threshold regressions analyses (Two-Stage Least Squares (2SLS) method and Generalized Method of Moments (GMM) estimation for the coefficients) with the initial GDP, human capital and trade volume as threshold variables. In order to cope with endogeneity, the log of population, bureaucracy, corruption and institutional quality were used as instruments. Their results are in line with the Borenzstein’s conclusion about the necessary higher level of human capital since they found that FDI exerts a greater positive effect in countries with a higher level of education. Moreover, they state that the FDI impact on economic growth in higher income countries is positive, despite the negative coefficient of the interaction term between FDI and initial GDP because the coefficient of FDI itself is far greater and positive. But the interaction terms between FDI and FDI itself in regards to the trade volume threshold are insignificant.
However, two papers contradict the conclusions of Jyun-Yi and Chih-Chiang (2008) about FDI and trade. Balasubramanyam, Salisu and Sapsford (1996) used a sample of 46 countries in the period 1970-1985 and performed a generalized instrumental variable method of estimation in order to control for simultaneous equation bias. The countries were grouped in two categories: countries that have export promoting strategies and countries with import substituting ones, based on the ratio of the imports to GDP, where a high ratio indicates an export-oriented country and an import substituting one otherwise. The authors concluded that in the first group of countries the growth enhancing effects of FDI are stronger than in the latter group. Thus, the results show that countries with more liberal trade regimes and fewer impediments benefit more from FDI. Moreover, the coefficient of domestic investment is insignificant and it is stated that FDI is more important to economic growth. Makki and Somwaru (2004) performed an IV estimation (Three Stage Least Squares) using data for 66 countries in the period of 1971 to 2000
5
Log value of area, measures for political stability and institutions, continental dummies for East Asia and South Asia and the other explanatory variables.
12 utilizing the average value of the variables for every decade and creating a system of three equations (one for every decade where the dependent variable is the mean value of GDP growth rate per capita). Then a SUR model is used as well as an IV estimation due to possible endogeneity problems. Their results are in accordance with the previous paper since they argue that FDI has a positive impact on GDP through its interaction with the trade term.
Alfaro, Chanda, Kalemi-Ozcan and Sayek (2003) used OLS as well as an IV model6 (thus dealing with endogeneity) and data for 49 countries in the period 1975-1995 and found that the precondition for the FDI growth enhancing effect is the development of the financial system. Six different measures7 were used as a proxy for it and all models confirm the significance of the interaction between FDI and financial markets while in most regression FDI alone is insignificant.
Nair-Reichert and Weinhold (2001) used data for 24 developing countries in the period 1971-1995, constructing a Mixed Fixed and Random coefficient approach in order to avoid homogeneity assumptions about the coefficients of the lagged dependent variables and allow for heterogeneity as well as to control for country-specific characteristics. The result shows that FDI impact on growth is highly heterogeneous, but on average FDI is growth-enhancing.
On the other hand, some studies found that there is no solid link between FDI and GDP. Carcovic and Levine (2002) used GMM estimation (72 countries, 1960-1995) in order to control for potential bias due to endogeneity and country specific effects. The results contradict the findings of the other authors since it is stated that FDI does not have a significant positive effect on economic growth. In addition, the interactions between FDI and education, financial development, trade and GDP per capita level were included in separate four regressions. Contrary to Borenzstein, Balasubramanyam, Alfaro, Chanda, Kalemi-Ozcan and Sayek (2003), no robust significance of the interaction terms was found. Chloe (2003) used a dataset for 80 countries in the period 1971-1995 and used a VAR model and Granger causality tests concluding that although there is reverse causality between FDI and growth, the effects from growth to FDI are more evident.
6 Legal origin variables and creditor rights are used as instruments for the financial markets while lagged values of
FDI and the exchange rate are the instruments for FDI.
7
Liquid liabilities of the financial system, Commercial-central bank assets, Private sector credit, Bank credit, Stock market liquidity and Capitalization.
13
2.2. FDI and growth in SSA
Esso (2010) examined the link between FDI and economic growth in the period from 1970 until 2007 in 10 SSA countries utilizing the Pesaran (2001) cointegration approach as well as the Toda and Yamamoto (1995) procedure for a non-causality test when conducting an unrestricted error correction model and VAR, respectively. His results are mixed because when the dependent variable is GDP per capita, FDI significantly causes economic growth in Angola, Cote D’Ivoire and Kenya, but when the left-hand-side variable is the ratio of FDI to GDP, then growth causes FDI in Liberia and South Africa. Tomasz Michalowski (2012) used data for 34 Sub-Saharan African countries for two periods (1981-1990 and 2001-2010) and the linear regression’s result showed no or a weak positive link between FDI and GDP growth in the region. Nonetheless, both papers do not consider endogeneity problems. Furthermore, Adams (2009) used panel data for 42 SSA countries in the period 1990-2003 and performed a dynamic fixed effects model with lagged values of GDP growth (thus controlling for endogeneity). The result is a lack of a positive FDI impact on economic growth. In line with most studies of FDI and growth, Adams (2009) made the conclusion that the FDI impact is determined by country-specific conditions – the level of absorption capacity, the policy environment and the FDI targeting (human capital8, political risk, domestic investment, trade9 variables were included).
However, another set of literature found a positive link between FDI and GDP in SSA. Seetanah and Khadoroo (2007) and Ndambendia and Njoupouognigni (2010) found that FDI is conducive to growth. The former conducted a static panel data estimation with random effects and a dynamic panel data regression with GMM estimation for 39 SSA countries in the period 1980-2000 while the latter used data for 36 SSA countries, 1980-2007 and performed dynamic panel data estimation techniques based on auto-regressive distributed lags models. The one step GMM estimator in the first paper is used in order to control for endogeneity. The second one used mean group estimator, pooled mean group estimator and dynamic fixed effects estimator aiming to avoid heterogeneity issues. Seetanah and Khadoroo (2007) found that not only FDI is significant and growth-enhancing, but also domestic private investment, public investment and
8
Proxied by secondary school enrollment.
14 the employment level. Ndambendia and Njoupouognigni (2010) stated that the domestic saving rate and the growth rate of the labor force affect positively GDP growth.
Moreover, Adeleke (2014) utilized panel data from 1996 to 2010 for 31 SSA countries, conducting three different regression techniques – pooled OLS, fixed and random effects, to estimate the interaction of FDI and governance on growth. All three regressions show a positive relation between FDI and GDP and a positive link between the interaction term and GDP. However, the author pointed out that governance in SSA is weak and thus is not able to effectively enhance growth. In addition, a dummy variable for landlocked and coastal countries is used and the result clearly showed that the countries that are not landlocked stimulate economic growth stronger. However, endogeneity issues were not addressed. Another paper where endogeneity was not accounted for is the paper of Johnson (2006). He did a cross-section analysis as well as panel data models for 90 countries in the period 1980-2002. The economies were divided into two subsamples – developing and developed economies, and two dummy variables for Sub-Saharan Africa and East Asia economies were introduced. Five-year average data was used in order to avoid short business cycle fluctuations and OLS, fixed and random effects estimation techniques were used, but the latter two did not change considerably the results. Performing the Hausman test, the random effects method was preferred. The results showed that FDI is growth-enhancing in the developing countries, but not in the developed ones. The dummy for SSA had a negative coefficient which indicates the poor development in the region.
Sukar, Ahmed and Hassan (2007) also conducted a fixed and random effects model and pooled cross section effect using data for 12 SSA countries in the period of 1975-1999. The independent variable is FDI as a percent of GDP, the dependent one is the growth rate of real GDP per capita. Also, the one-period lagged logarithm of real GDP per capita and the logarithm of real GDP per capita in the initial year of the considered period are included on the right-hand side of the equation. In addition, control variables are inflation rate, population growth, trade openness (measured as the sum of real exports and imports as a share of GDP), investment growth and government consumption (expressed as percentage of GDP). The performed Hausman test suggests that the random effects model is more suitable in their case than the fixed effects one. The results indicate that FDI has a positive coefficient of 0.004, significant at the 10% level.
15 Furthermore, using SUR estimation, data from 1980 to 2000 and for 47 African countries, Lumbila (2005) stated that there is an overall positive impact of FDI on growth. However, this is dependent on country characteristics – government consumption, trade openness, population growth, level of population’s education (proxy for human development), which are used as control variables. Moreover, in order to control for endogeneity Lumbila performed an IV estimation using FDI lagged values as instruments and the coefficient of FDI remained positive.
The papers discussed in this chapter indicate that neither unanimity about the impact of FDI on GDP growth, nor a unanimous agreement about the variables that affect positively or negatively the economic growth exist in the empirical literature.
3. Empirical analysis
This chapter presents a theoretical background in order to explain what variables are included in the regression and why (section 3.1.). It describes the model specification (section 3.2.), provides details about the data gathering and transformation (section 3.3.) and ends with the results and their discussion (section 3.4.).
3.1. Theoretical background
Economic growth models have the same basic structure, which includes a production function consisting of physical capital, labor and technology/knowledge. The different growth models are grouped in two categories: neoclassical growth theories and endogenous growth theories. In the neoclassical framework, FDI and other variables affect only the level of income and in the long run economic growth does not depend on economic conditions10. GDP per capita growth is due to the technology progress (which here accounts for productivity), which is exogenous and cannot be explained. However, economic activities like the industrial innovations and the human capital accumulation lead to a technological change. Therefore, technology should be considered an endogenous variable and in order to analyze the policy effects on growth, the endogenous growth models were developed (Aghion and Howitt, 2009). In these models, the variables that have an effect upon the Research & Development (R & D)11 activities, also affect
10
They affect the steady-state level of output, but not the long term growth rate (Barro et.al., 1995).
11
16 the long-term growth. Thus, most recent empirical studies base their approach on the endogenous growth theory; however, the idea about the conditional convergence12 is taken from the neoclassical theory. The conditional convergence expresses the idea that countries with the same technology and fundamental determinants (depreciation rate, population growth rate, etc.) will converge to the same steady state (Barro and Sala-i-Martin, 2004). Therefore, one can say that the empirical research usually combines different aspects of both growth theories.
The data for the physical capital is scarce and usually unreliable in the developing countries, therefore in the empirical studies the level of initial GDP per capita is used as a proxy for it while the dependent variable is economic growth measured by GDP per capita (Barro and Sala-i-Martin, 2004). The reason behind the use of per capita terms is the following: if two countries have the same level of GDP, but the population in one of them is greater than in the other, then if the countries were compared by GDP level, this would lead to the conclusion that they have same level of income and development, which is wrong. Once GDP per capita is used, the difference can be seen properly.
As Barro and Sala-i-Martin (2004) explain, the empirical framework encompasses two types of independent variables: initial levels of state variables13 (the stocks of human capital and physical capital) and control (environmental)14 variables like the economy openness (measured as the ratio of export and imports to GDP). However, the empirical studies differ in the choice of a set of control variables. In the next paragraphs the most commonly chosen variables (also the ones used in this paper’s regression model) and the logic behind including them in the models are explained.
The state variables are the initial level of GDP per capita and human capital while the control ones are FDI, inflation, population growth, government consumption, investment, trade and political stability. The initial level of GDP per capita is added in the model in logarithmic form, thus the coefficient in the regressions not only is a proxy for physical capital, but it also shows the convergence rate - the reaction of the growth rate to a proportional change in the initial level of GDP per capita (Barro and Sala-i-Martin, 2004). Thus, countries with a lower starting
12 There are also theories about absolute and club convergence, see Aghion and Howitt, 2009. 13
State variables “describe the evolution of the state of the economy” (Barro and Sala-i-Martin, 2004, p.605).
17 GDP will grow at a faster rate. Additionally, it controls for past economic conditions (Makki and Somwaru, 2004). The most commonly used proxy for human capital is education measured as average years of schooling or enrolment ratios.15 Technology advancement is a human capital intensive activity and since growth is dependent on innovation, it is also dependent on the level of education (Aghion and Howitt, 2009). Thus, a negative coefficient for the initial GDP and a positive one for the education variable are expected.
Moreover, the growth theories state that capital accumulation is one of the economic growth driving forces and since FDI affects it directly, one should expect that FDI has an impact upon economic growth (Moosa, 2002). Furthermore, the ratio of investment to GDP captures the effect of the saving rate (Barro and Sala-i-Martin, 2004) while the degree of trade (economy openness) is included because trade “facilitates more efficient production of goods and services
by shifting production to countries that have comparative advantage in producing them” (Makki
and Somwaru, 2004, p. 2). Therefore, positive coefficients for these variables are expected. Additionally, government consumption measures expenses that do not have a direct effect on productivity. However, these expenses distort private decisions. Therefore, the higher the government consumption, the lower the steady-state value of output (Barro and Sala-i-Martin, 2004) and a negative coefficient is anticipated.
Macroeconomic stability matters since it is one of the conditions that attract investors and inflation rate is used as a measure. The higher the inflation rate, the higher the implied instability, thus a negative coefficient is expected. In addition, political stability is directly related to the previous one since it is associated with enhanced property rights and thus provides and incentive to invest (Barro and Sala-i-Martin, 2004). If it is not present, investors are not willing to invest in the country. Therefore, a positive coefficient should be expected.
Furthermore, the neoclassical model predicts a negative coefficient for the population growth variable16, while some endogenous growth models17 predict a positive coefficient, the explanation being that the population growth leads to a bigger market and an increased supply of
15
Hanushek (2013) states that cognitive skills is what matters and quality of schooling is the best measurement for the impact of education on economic growth. However, due to data limitations about quality of schooling, the other variables are widely used.
16 The explanation being that when the population grows, having all other variables constant, the income must be
distributed among more people. Thus population growth decreases the level of GDP per capita.
18 potential researchers. The assumption in such models is that there are constant returns to scale of new knowledge creation (thus economic growth is increased when population size has increased because it supplies more researchers). However, as Jones (1995) argues, the empirical research does not confirm the idea about “the scale effect of increased population”18,19. Since scale effects do not reflect reality properly, an additional assumption was introduced leading to the growth rate being independent of population. This assumption is the insight that when population increases, the greater variety of products reduces the effectiveness of research aimed at quality improvement because it spreads it “more thinly over a larger number of different sectors, thus
dissipating the effect on the overall rate of productivity growth” (Aghion and Howitt, 2009, p.
97). However, Jones (1995) developed a model called semi-endogenous growth model20, in which the key assumption is that there are diminishing returns to scale of knowledge since with the increase of knowledge it becomes more difficult for someone to come up with a new idea. Therefore, the growth rate depends on the R&D growth rate which, in turn, is dependent on the population growth rate, “reflecting an intuitive link between innovations and scientists:
inventions require inventors” (Jones, 1995, p. 760). As Prettner and Prskawetz (2010, p.16) state, “whether population growth or population size foster or hamper economic growth strongly depends on the modelling framework”. Hence, as generally accepted in most papers, a negative
coefficient for population growth is expected, but a positive one is also possible.
3.2. Methodology
Panel data is used due to the fact that it provides certain advantages in comparison to cross-sectional data. It allows “us to exploit the time-series nature of the relationship between
FDI and growth” (Carcovic and Levine 2002, p. 4). Moreover, panel data allows for less
multicollinearity and more degrees of freedom than cross-sectional data. Furthermore, it takes account of inter-individual differences as well as intra-individual dynamics, thus allowing one to control for unobserved or missing variables (Hsiao, 2005).
18 The scale effect expresses the idea that population growth should lead to economic growth due to increased
number of researchers and higher demand.
19
Jones (1995) points at the fact that in the developed world the number of people doing R&D activities has increased significantly while the growth rates have experienced a slight increase or even a decrease in some countries.
20 It is named semi-endogenous since it states that the long-run economic growth does not depend on the policy
changes, which corresponds to the neo-classical theories, but it also shows that the growth is endogenous since it depends on the R&D growth rate, which is in line with the endogenous growth theories.
19 In regard to the econometric models, although Pooled Time Series Cross Section (PTSCS) captures variation through time and space, errors can be serially correlated and/or correlated across entities (Podestà, 2002). Even using a model that controls for these disadvantages, generally PTSCS can suffer from omitted variable bias (Adeleke, 2014). Although SUR model handles different error variances and contemporaneous correlation, it cannot control for endogeneous regressors. In those cases IV should be introduced. (Moon and Perron, 2006). However, suitable instruments are not easily available. Therefore, the choice of regression model is between fixed and random effects estimation techniques because they control for country-specific, time-invariant characteristics and thus are adequate for testing the FDI effect on growth. There is an important difference in one of the assumptions: in the random effects model no correlation between the explanatory variables and the unobserved effects is assumed while in the fixed effects such correlation is allowed.
The methodology is based on the paper of Sukar, Ahmed and Hassan (2007) and the model is further expanded by dividing the countries in two groups by income: low-income and income SSA countries, using a dummy variable, which is 1 if the country is middle-income and 0 otherwise. In addition to the stated control variables in the paper of Sukar, Ahmed and Hassan, political stability and average years of schooling are included. Therefore, the model specification is the following:
𝑦𝑖,𝑡 = 𝜆𝑙𝑛𝑌𝑖𝑡0 + 𝛽 1𝐹𝐷𝐼𝑖,𝑡+ 𝛽2𝐹𝐷𝐼𝑖,𝑡. 𝐷𝑖+ 𝛽𝑘𝑋𝑖,𝑡+ 𝛼𝑖+ 𝜇𝑡+ 𝜀𝑖,𝑡 𝑦𝑖,𝑡 𝑙𝑛𝑌𝑖𝑡0 𝐹𝐷𝐼𝑖,𝑡 𝑋𝑖,𝑡 𝐷𝑖 𝛼𝑖 𝜇𝑡 𝜀𝑖,𝑡 GDP growth per capita Logarithm of the initial value of GDP per capita in the period FDI as a percent of GDP Vector of control variables Dummy variable for the income status of the countries Country-specific effects Time-specific effects Error term
20 Furthermore, due to data availability limitations for the political stability indicator21, data for the period 1995-2014 is used. In order to avoid implications and bias due to business cycle fluctuations, the data is divided into four non-overlapping five-year periods: 1995-1999, 2000-2004, 2005-2009 and 2010-2014. Other authors follow the same approach of using 5-year pooled data, for example, Carcovic and Levine (2002) and Johnson (2006). In the model above, i denotes country while t indicates period.
The 5-year average of GDP per capita growth rate in period t is denoted by yi,t and calculated as the sum of GDP growth rate for the years in the period and divided by the number of years in the period.22 lnY0it (lnInitialGDP)represents the value of GDP per capita in the initial year of the period under consideration. For instance, if the period is 2000-2004, the value for year 2000 is taken. FDIi,t is the average value of FDI as a share of GDP in a period, calculated analogically to yi,t.. Xi,t denotes a set of control variables: average years of total schooling (Education) in the initial year of the period23, average inflation (Inflation), average population growth (PopGrowth), average trade as percentage of GDP (measured as the sum of exports and imports as a share of GDP - Trade), average government consumption as percentage of GDP (GovConsump), investment (gross capital formation) as percentage of GDP (Investment) and the average estimate for political stability and absence of violence and terrorism (PolStability) during the period. All control variables averages are calculated similarly to yi,t (except for the education variable). As usual, εi,t is the disturbance term.
Furthermore, when the effect of one of the independent variables depends on the value of another of the independent variables, an interaction effect is present (Jaccard, Turrisi, 2003). Therefore, following the usual practice in most papers, the products of FDI with trade, government consumption and investment are added to the regression equation while the variables are kept individually in it, too. Thus we can see whether FDI affects growth on its own or through the interaction term (Borensztein, et.al., 1996). The dummy variable Di takes value 1 if the country is middle-income and 0 if it is a low-income one. Also the interaction between it and FDI is added, thus the separate effect as well as the effect through the interaction term can be
21 Political Stability and Absence of Violence/Terrorism is one of the Worldwide Governance Indicators. The data is
available from 1996 onwards.
22 If data is missing for a year/years in the period, the average is calculated dividing by the number of years for which
data is available (otherwise it would be as if this years had value of 0, which is not true).
23
21 observed. Which estimation performs better – fixed or random effects, is checked with the Hausman test.
Having pooled cross-section time series data, “the error process may be homoskedastic
within cross-sectional units, but its variance may differ across units: a condition known as groupwise heteroskedasticity” (Baum, 2001, p.101). Therefore, after the model was developed,
the modified Wald statistic for groupwise heteroskedasticity in the residuals of a fixed-effects model was used to test whether heteroskedasticity is present. The null hypothesis that homoskedasticity is present is rejected and heteroskedasticity is confirmed (see Appendix, Table A2). Furthermore, since a panel data is used, the error term εi,t might be autocorrelated. In this
case, the estimators are consistent, but the standard errors are not, thus clustered standard errors are necessary because they allow for correlation within units. Having autocorrelation and heteroskedasticity requires the use of clustered robust standard errors (Stock and Watson, 2015). Therefore, in all models presented in the paper, clustered robust standard errors are used.
In the paper of Sukar, Ahmed and Hassan (2007) not only the logarithm of GDP value per capita in the initial year of the period is included, but also the logarithm of the lagged value of GDP per capita. However, here only the GDP per capita in the initial year is added due to the fact that the two variables are highly correlated – the correlation value turned out equal to 0.9949. High correlation increases parameter variance due to collinearity and thus leads to issues like parameters’ insignificancy although R2
is large, implausible magnitudes of parameters, etc. (O’brien, 2007). In addition, Jyun-Yi and Chih-Chiang (2008), Alfaro, Chanda, Kalemi-Ozcan and Sayek, (2003) and others also include only the initial value of GDP.
3.3. Data
The initial panel data set employed data for the period from 1990 to 2014 for 47 SSA countries. However, due to the importance of the political stability parameter, the period 1995-2014 was considered (the period 1990-1994 is left out). Furthermore, after dropping the missing observations for the dependent variable (GDP growth per capita), the data set included 41 countries. After the main model was developed, dropping the missing observations for the independent variables led to a data set of 32 countries and 123 observations (Appendix, Table A1 lists the countries).
22 The countries status is obtained from the Country Analytical History by the World Bank Analytical Classifications (presented in World Development Indicators). All lower-middle-income and upper-middle-lower-middle-income countries are assigned status middle lower-middle-income and then this information is used in order to assign 1 or 0 as the value for the Dummy variable for each country in every period. The dummy value is 1 if in the given period the predominant status of the country was middle-income and 0 otherwise.24 The list of countries and their economy status in each of the considered four periods can be found in Table A1, Appendix.
The data for GDP level, GDP growth, FDI, inflation, population growth, gross capital formation, government consumption and trade is taken from the World Bank’s World Development Indicators dataset while the data for political stability and absence of violence/terrorism is obtained from the World Bank’s Worldwide Governance Indicators dataset. GDP growth per capita, population growth and inflation are expressed in percentages. The gross capital formation, government consumption and trade are expressed as percentage of GDP. The political stability variable measures the likelihood of political stability and absence of violence and terrorism and is estimated as a score between -2.5 to 2.5. We can see that the maximum value of it is 1.04 while the mean is -0.58 (see Appendix, table A4) which indicates that the countries in the SSA are not as politically stable as other regions in the world and this could be seen as a signal that we should control for this in the regression since companies are not so willing to invest if political risk is present. Some of the reasons are that war, riots, disorder, host government’s attitude, changes in the FDI regulations, etc. can affect negatively the decision to invest in these countries or to make the companies that are already there, disinvest (Moosa, 2002). In the Appendix, Table A3 provides detailed explanation about the variables, table A4 presents the summary statistics of the data and table A5 pictures the correlation matrix.
24 For example, if the country had low income status from 2000 to 2003 and middle-income status for 2004, value of
0 is assigned for the period 2000-2004. Actually, 23 out of the 32 countries in this paper dataset have not changed their status during the period 1995-2014. Only 9 have managed to change it: Cameroon, Congo, Rep., Cote d’Ivoire, Ghana, Kenya, Lesotho, Mauritania, Senegal, Sudan and Zambia. All of them moved from a low to a middle income status.
23
3.4. Results and discussion
The first three regression results are presented in table 3a. Model 1 shows a simple fixed effects model with dependent variable GDP growth per capita and independent variables FDI, inflation and population growth. Adding the logarithm of the initial GDP value gives the second model and although the coefficient is not significant, the R2 improves from 0.21 to 0.24, which indicates that accounting for the past values of GDP is important. Trade, government consumption and investment are added, thus constituting model 3.
Table 3a: Basic models
(1) (2) (3)
VARIABLES Model 1 Model 2 Model 3
FDI 0.0981 0.105* 0.0663 (0.0621) (0.0612) (0.0599) Inflation -0.000689*** -0.000775*** -0.000760*** (4.58e-05) (4.79e-05) (0.000113) PopGrowth 1.971*** 1.785*** 1.506*** (0.456) (0.356) (0.348) Trade 0.0269** (0.0118) GovConsump 0.0188 (0.0948) Investment 0.0340 (0.0462) lnInitialGDP -1.874 -3.337** (1.412) (1.560) Constant -3.392*** 9.201 16.41 (1.199) (9.001) (10.17) Observations 168 165 153 R-squared 0.214 0.239 0.318 Number of countryid 45 43 42
Country FE YES YES YES
Dependent variable: 5-year average GDP per capita growth. Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
The coefficients of FDI, inflation, trade, investment and initial GDP are as expected. However, FDI is insignificant while inflation, trade and initial GDP are significant at the 1%, 5% and 5% level, respectively. Population growth is positive and significant at the 1% level,
24 government consumption is surprisingly positive, but insignificant and the R2 improves to 0.32. Comparing model 3 results to the ones obtained from Sukar, Ahmed and Hassan (2007), what is noticed is that in both papers the coefficients have the same signs, except for population, which may be explained with the different periods used in the two models. However, R2 is not very high, which indicates that other variables are necessary to account for the growth drivers.
Following the approach of Sukar, Ahmed and Hassan (2007), model 3 was obtained. Adding to it a political stability variable not only improves the R2, but also enters the regression with a positive coefficient, significant at the 10% level (table 3b, model 4). It points at its important role to provide an environment suitable for investment, which leads to growth. The dummy for income status of the countries and its interaction with FDI as well as an interaction term of FDI with government consumption are added in model 5. FDI has a positive, significant coefficient and its interaction term with the government coefficient is significant and negative. However, government consumption separately has a positive sign. The other variables keep their signs the same. Nevertheless, a variable that accounts for human capital is missing so far, thus the average years of schooling and its interaction with FDI are included in the regression equation. Thus, model 6 shows positive, significant coefficients for this interaction and the dummy. Model 7 adds time (period) fixed effects to model 6 and then a test is performed to check if they are necessary (table 3c). The Prob > F is lower than 0.05, indicating that the null hypothesis that the period coefficients are jointly equal to 0 is rejected, therefore time effects should be accounted for. Moreover, after the development of model 7, it was conducted with two estimation techniques – fixed and random effects and then the Hausman test was performed in order to check which model is more efficient, confirming that the fixed effects estimation should be the preferred one (see Appendix, Table A6 and Table A7). Therefore, model 7 is the main model of this paper and it presents the results that will be discussed further.
The coefficients of all variables preserve the same signs as in model 6 and model 7 explains 57% of the change in GDP growth. FDI exerts a positive, significant at the 5% level effect on economic growth with a coefficient of 0.271. Trade and inflation have the expected signs and their coefficients are significant at the 5% and 1% level, respectively. Considering inflation as an indicator for the macro stability, it shows that a sound macro environment is necessary in order to provide the conditions needed for investment and growth.
25
Table 3b: FDI effects on GDP per capita growth. Low-income and middle-income countries
(1) (2) (3) (4)
VARIABLES Model 4 Model 5 Model 6 Model 7
FDI 0.0501 0.274*** 0.254** 0.271** (0.0553) (0.0624) (0.0991) (0.0993) Dummy 1.300 2.125* 1.051 (0.876) (1.110) (1.176) Dummy x FDI 0.0145 -0.205* -0.141 (0.0860) (0.114) (0.126) GovConsump x FDI -0.0120*** -0.0213*** -0.0206*** (0.00218) (0.00520) (0.00577) Education x FDI 0.0635** 0.0508* (0.0275) (0.0285) Inflation -0.000778*** -0.000771*** -0.000690*** -0.000685*** (0.000108) (0.000106) (0.000136) (0.000136) PopGrowth 1.349*** 1.276*** 1.251*** 1.085** (0.437) (0.405) (0.384) (0.403) Trade 0.0253** 0.0245*** 0.0276** 0.0224** (0.0122) (0.00893) (0.0105) (0.0108) GovConsump -0.0110 0.0152 0.0879 0.105 (0.0866) (0.0829) (0.0919) (0.0868) Investment 0.0451 0.0590 0.0532 0.0381 (0.0456) (0.0371) (0.0403) (0.0403) PolStability 0.936* 0.868* 0.640 0.718 (0.490) (0.452) (0.437) (0.426) Education -0.126 -0.693* (0.328) (0.361) lnInitialGDP -3.537** -4.479*** -5.217*** -6.113*** (1.574) (1.364) (1.293) (1.296) 2.Period -0.809* (0.465) 3.Period 0.195 (0.545) 4.Period 1.244 (1.021) Constant 19.04* 24.13*** 27.86*** 37.40*** (10.41) (8.861) (8.462) (9.715) Observations 153 153 123 123 R-squared 0.343 0.438 0.504 0.566 Number of countryid 42 42 32 32
Country FE YES YES YES YES
Period FE YES
Dependent variable: 5-year average GDP per capita growth Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
26
Table 3c: Test for time fixed effects
This corresponds to the result about inflation found in Makki and Somwaru (2004), Lumbila (2005). In addition, trade indicates the economy openness, supporting the idea that trade liberalization increases GDP growth. Thus, this result can be considered in line with the paper of Balasubramanyam, Salisu and Sapsford (1996). Moreover, investment has a positive insignificant coefficient while FDI has a significant one, which confirms the other finding in the paper of Balasubramanyam et.al. (1996): FDI is a more important growth driver than domestic investment in the developing countries because FDI is associated with positive spill-over effects in regards to human capital and technology.
The initial value of GDP has the expected negative coefficient and its significance is at the 1% level. This is the convergence rate which confirms the neoclassical suggestion that countries with lower initial income have a higher growth rate and thus countries with the same economy fundamentals converge to the same steady state in the long run.
Government consumption has an insignificant positive coefficient; however, the sign of it is surprising. Nonetheless, the interaction term of it with FDI is significant at 1% level and with a negative coefficient, which indicates that when government expenses increase, FDI’s positive effect on the economy is lowered. The reason is that a sound government policy is important for the economy, but the government consumption can also have a negative effect if it is crowding out other types of investment. Political stability is insignificant, but still positive. Moreover, the interaction terms of FDI with trade, investment and political stability turned out insignificant, therefore they are not included in table 3b.
Population growth has a positive coefficient, significant at the 5% level. However, normally, the expected sign is a negative one, confirmed by Alfaro, Chanda, Kalemi-Ozcan and
Prob > F = 0.0472 F( 3, 31) = 2.96 ( 3) 4.Period = 0 ( 2) 3.Period = 0 ( 1) 2.Period = 0 . testparm i.Period
27 Sayek (2003), but here the positive sign might be considered not that surprising since the population in SSA is currently undergoing a rapid transition, more and more young people are entering the job market while the fertility rates are decreasing (Westoff, Bietsch, Koffmann, et.al., 2013), which leads to a lower dependency ratio, thus raising the economic growth.
Moreover, the average years of schooling has a surprising negative coefficient, significant at the 10% level, while the coefficient of the product of FDI and the education variable is positive and significant at the 1% level. The negative coefficient might be explained with the still low level of education and education quality in SSA as well as with the fact that in the regression, the level of average years of schooling in the first year in the period are included.25 Therefore, this resembles the initial value of GDP and the lower the starting level of education, the higher the growth rate since the effect of the FDI externalities in regards to human capital will be bigger. The positive sign of the interaction term points that FDI is growth-enhancing not only separately, but also when education is present. This can be interpreted in the following way: FDI is associated with technology advancement and spillover effects and if education is increasing, the FDI effect will be higher since people will be able to better cope with the new technologies and enhance growth, in turn. The constant term is significant at the 1% level and since it represents the average effect of the fixed effects in this regression, leads to the conclusion that the country differences that we cannot measure matter for economic growth.
The dummy variable keeps its positive sign, meaning that the middle income countries have a higher intercept, keeping the other variables constant. The interaction term with the dummy and FDI is with a negative sign, meaning that the slope for middle income countries is 0.271-0.141=0.13, which shows a steeper slope for the low-income countries. This is in line with the conditional convergence idea and the coefficient of the initial value of GDP. However, once the time effects were added, these differences turned insignificant. Therefore, we can conclude that income differences in SSA cannot explain differences in the FDI effects on growth and other factors, such as trade and education are crucial for the economic development.
Moreover, the time effect of the second period (2000-2004) is significant and affects negatively the economic growth. This is explained by the specific events that happened in SSA
25
The model was also performed by including the difference of average years of schooling between every two periods. However, the results were similar to the ones obtained when with the initial level of education in the period.
28 during this period. A food crisis was already present and further exacerbated by the serious droughts. In 2003, 25 countries in the region were faced with food emergencies, 10 of which were also undergoing a social unrest/civil war and 4 were just emerging from conflicts. Furthermore, the HIV pandemic was worsening the situation further (Clover, 2003).
4. Robustness check
In order to check the robustness of the results obtained and to deal with endogeneity, an IV estimation is performed. Following Makki and Somwaru (2004), lagged values of FDI, lagged values of trade and logarithm of GDP were used as instruments. In this case, some variables change their sign (average years of schooling, the interaction between average years of schooling and FDI, the interaction between FDI and the dummy variable and trade) and the only significant one is inflation.26 Testing for endogenous regressors did not confirm the presence of endogeneity. Moreover, the test for instruments shows that they are weak. Furthermore, other variables were also used as instruments (Political stability, Average years of schooling, lagged values of GDP). In all cases, the results yielded are similar to the ones presented in table A8, Appendix (in some cases, only trade changed its sign and investment became significant). All regressions turned out to be misspecified due to either failure to reject the null hypothesis of exogenous variables27 or due to invalid/weak instruments. Therefore, we are inclined to believe that in this case the fixed effects model should be preferred.
Another check was done using the data for FDI from UNCTAD STAT, keeping all other variables data unchanged. Performing the main model with this FDI data, yields Appendix table A9. The variables have the same signs and they are also significant, except for FDI and Education, which became insignificant while the interaction between the dummy and FDI became significant at the 10% level. Although FDI is insignificant, it is still positive. The insignificancy may be explained with the difference of the way FDI as a share of GDP is calculated by World Bank and UNCTAD. The variable by UNCTAD calculates inward and outward flows and stock as share of GDP while the one from World Bank includes only the net inflows of FDI as a share of GDP. Thus, the insignificancy might be explained by the fact that
26 lnInitial GDP is removed from the regression in order to keep lnGDP as instrument. Otherwise the test for
overidentifying restrictions shows that the model is not valid.
29 outflows are included. Therefore, the World Bank measure should be preferred. However, the other variables’ effect on economic growth is confirmed.
To further check for robustness, another fixed effects model is conducted, but using 10-year periods instead of 10-year ones. The calculations are made analogously to the ones about 5-year periods, as explained in section 3.3. Thus, the two periods are 1995-2004 and 2005-2014 and table 4a shows the results from the regression.
Following the main model in this paper, Model 8 has period fixed effects, but then the performed test cannot reject the null hypothesis that all time effects coefficients are jointly equal to 0 and indicates that in this case time effects are not necessary. Therefore, model 9 excludes period fixed effects and will be the one compared to the model with the 5-year average data. Comparing both model results, one can notice that all variables keep the same sign except for inflation, trade and education. However, all variables are insignificant, apart from FDI and lnInitialGDP, significant at the 5% level. A possible explanation for the parameters that became insignificant might be the fact that only two periods are used due to data limitations. The 10-year periods may have smoothed out business cycle effects, but the presence of heterogeneity indicates the differences between the countries and it is possible that different coefficients might offset one another.
Lastly, an OLS regression was performed using the average data for the whole period of twenty years (in this way the panel data is transformed into a cross-sectional one and thus fixed effects estimation is not feasible). FDI, initial GDP and inflation are significant with the expected signs of the coefficients (table 4b). Surprisingly, the interaction between FDI and education is negative, significant at the 10% level. However, the education in the regression is the value of average years of schooling in the first year of the period, thus pointing out at the fact that the education in 1995 in SSA was at a low level, unable to fully exploit the benefits of FDI.
Nonetheless, having all of this in mind, the results from the main model cannot be considered robust, except for the importance of FDI and the initial income level.
30
Table 4a: 10-year average data
(1) (2)
VARIABLES Model 8 Model 9
FDI 0.340 0.447** (0.212) (0.197) Dummy 0.659 1.238 (1.067) (0.939) Dummy x FDI -0.0605 -0.123 (0.208) (0.192) GovConsump x FDI -0.0116 -0.0203 (0.0149) (0.0122) Education x FDI -0.00194 0.0196 (0.0519) (0.0449) Inflation -0.000108 0.000275 (0.000762) (0.000548) PopGrowth 0.377 0.427 (0.818) (0.790) Trade -0.0246 -0.0197 (0.0174) (0.0211) GovConsump 0.163 0.177 (0.174) (0.163) Investment 0.0117 0.0522 (0.0640) (0.0539) PolStability 0.505 0.334 (0.738) (0.661) Education 0.146 0.270 (0.477) (0.623) lnInitialGDP -7.053** -5.641** (2.638) (2.115) 2.Period 0.984 (0.810) Constant 44.09** 32.85** (18.23) (12.77) Observations 63 63 R-squared 0.695 0.666 Number of countryid 32 32
Country FE YES YES
Period FE YES NO
Dependent variable: 10-year average GDP per capita growth Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
31
Table 4b: OLS with 20-year average data
(1) VARIABLES Model 10 FDI 0.687** (0.243) Dummy 0.951 (1.504) Dummy x FDI 0.0302 (0.139) GovConsump x FDI -0.0177 (0.0224) Education x FDI -0.0876* (0.0472) Inflation -0.00225*** (0.000725) PopGrowth -0.884 (0.939) Trade -0.00570 (0.0125) GovConsump 0.0214 (0.150) Investment 0.148 (0.0899) PolStability 0.173 (0.583) Education 0.496 (0.307) lnInitialGDP -1.409* (0.812) Constant 7.744 (4.902) Observations 32 R-squared 0.626
Dependent variable: 20-year average GDP per capita growth. Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
Furthermore, there are certain caveats that need to be mentioned. First, the fixed effects estimation is a tool with which the omitted variable bias can be avoided. The log value of initial GDP of the period accounts to some extent for the past condition, but still the problem of possible
32 simultaneity in the model is not solved. Thus, the endogeneity issue is not resolved.28 Second, due to data limitations the period is shorten to twenty years, but a longer period might give further information. Third, as Hanushek (2013) states, quality of schooling might be more relevant than the quantity, but the data for education quality is scarce, thus it is not included in this paper. Therefore, average years of schooling might not be a good enough proxy for human capital. Fourth, again due to data limitations, no variable accounting for the financial markets is added. Future studies might address these issues.
5. Conclusions
There are a lot of studies about the link between FDI and economic growth, but the literature about Sub-Saharan Africa is not that abundant. Moreover, the focus usually is on governance differences, geographic location or resources endowments. Therefore, the aim of this paper is to analyze the effect of FDI on GDP growth in SSA considering income differences in the countries in the region. The data sample constitutes of 32 countries for the period from 1995 to 2014 and fixed effects estimation is used. The results indicate that FDI exerts a positive effect on economic growth significantly. Furthermore, the interactions of FDI with government consumption and average years of schooling are significant and with the expected negative and positive coefficients, respectively. Also, a confirmation about the initial level of income, the importance of macro stability and openness of the economy as growth drivers is found. Moreover, the hypothesis that low-income and middle-income countries may differ significantly in the effect of income status on growth is not proven. However, when 10-year and 20-year averages are used, the results confirm the importance of FDI in the region and the significance of the convergence coefficient, but impose some doubts about the conclusions from the 5-year average regression about the other variables.
The policy implications are that FDI should be considered growth-enhancing. Moreover, economy stability and openness should be promoted. Also, the education should be improved since it facilitates the positive effects of FDI as well as a sound fiscal policy is necessary. These
28 The main model was also performed with one-period lagged value of GDP instead of lnInitialGDP. The
coefficients have the same signs, R2 is lower, and the second period is dropped. The model was done also with one-period lagged values of FDI instead of FDI. In this case, the only significant variables are initial GDP, political stability and inflation rate. However, using lagged values of GDP or FDI is not a complete solution for endogeneity since partially controlling for reverse causality cannot solve a simultaneity bias.
33 conclusions are in line with what Seetanah and Khadoroo (2007), Ndambendia and Njoupouognigni (2010), Lumbila (2005) have found.
The caveats that should be emphasized are that the possible endogeneity issue is not resolved and due to data limitations, three issues appear: the data period is shortened to twenty years, the proxy for education is average years of schooling despite the fact that the education quality measure is a better proxy for human capital and no variable in the model accounts for financial markets. Future studies might solve these weaknesses and gain further insights about the economic growth and FDI in Sub-Saharan Africa.
34
References
Adams, S. (2009). Foreign direct investment, domestic investment, and economic growth in Sub-Saharan Africa. Journal of Policy Modeling, 31(6), 939-949.
Adeleke, A. I. (2014). Fdi-Growth Nexus in Africa: Does Governance Matter?.Journal of
Economic Development, 39(1), 111.
Aghion, P., & Howitt, P. W. (2008). The economics of growth. MIT press.
Aoki, Y., & Todo, Y. (2008). FDI and economic growth in less developed countries. OECD Journal: General Papers, 2008(1), 1-31.
Balasubramanyam, V. N., Salisu, M., & Sapsford, D. (1996). Foreign direct investment and growth in EP and IS countries. The economic journal, 92-105.
Barro, R. J. (2004). Sala-i Martin X (2004) Economic Growth.
Baum, C. F. (2001). Residual diagnostics for cross-section time series regression models. The Stata Journal, 1(1), 101-104.
Borensztein, E., De Gregorio, J., & Lee, J. W. (1998). How does foreign direct investment affect economic growth?. Journal of international Economics, 45(1), 115-135.
Carcovic, M., & Levine, R. (2002). Does foreign direct investment accelerate economic
growth?. University if Minnesota. Department of finance Working Paper. June.
Choe, J. I. (2003). Do foreign direct investment and gross domestic investment promote economic growth?. Review of Development Economics, 7(1), 44-57.
Clover, J. (2003). Food security in sub-Saharan Africa. African Security Studies, 12(1), 5-15.
Esso, L. J. (2010). Long-run relationship and causality between foreign direct investment and growth: Evidence from ten African countries. International Journal of Economics and
35
Hsiao, C. (2005). Why panel data?. The Singapore Economic Review, 50(02), 143-154. Jaccard, J., & Turrisi, R. (2003). Interaction effects in multiple regression (No. 72). Sage. Johnson, A. (2006). The effects of FDI inflows on host country economic growth. The
Royal Institute of technology. Centre of Excellence for studies in Science and Innovation http://www. infra. kth. se/cesis/research/publications/working papers.
Jones, C. I. (1995). R & D-based models of economic growth. Journal of political
Economy, 759-784.
Jyun-Yi, W., & Chih-Chiang, H. (2008). Does foreign direct investment promote economic growth? Evidence from a threshold regression analysis. Economics Bulletin, 15(12), 1-10.
Lumbila, K. N. (2005). What Makes FDI Work?: A Panel Analysis of the Growth Effect of
FDI in Africa. World bank.
Makki, S. S., & Somwaru, A. (2004). Impact of foreign direct investment and trade on economic growth: evidence from developing countries. American Journal of Agricultural
Economics, 86(3), 795-801.
Michałowski, T. (2012). Foreign direct investment in Sub-Saharan Africa and its effects on economic growth of the region. Prace i Materiały Instytutu Handlu Zagranicznego
Uniwersytetu Gdańskiego, (31,[1]), 687-701.
Njoupouognigni, M. (2010). Foreign aid, foreign direct investment and economic growth in Sub-Saharan Africa: Evidence from pooled mean group estimator (PMG). International
Journal of Economics and Finance, 2(3), 39.
O’brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality & Quantity, 41(5), 673-690.
Prettner, K., & Prskawetz, A. (2010). Demographic change in models of endogenous economic growth. A survey. Central European Journal of Operations Research, 18(4), 593-608.
36
Sayek, S., Alfaro, L., Chanda, A., & Kalemli-Ozcan, S. (2003). FDI spillovers, financial
markets and economic development (No. 3-186). International Monetary Fund.
Seetanah, B., & Khadaroo, A. J. (2007). foreign direct investment and growth: New evidences from Sub-Saharan African countries. University of Mauritius, 27.
Sukar, A., Ahmed, S., & Hassan, S. (2007). The Effects of Foreign Direct Investment on Economic Growth: The Case of Sub-Sahara Africa.Southwestern Economic Review, 34(1).
Westoff, C. F., Bietsch, K., Koffman, D., Agadjanian, V., Asamoah, B. O., Agardh, A., ... & Sonko, S. (2013). Indicators of trends in fertility in sub-Saharan Africa. Studies In Family
