The effects of FDI on economic growth in Africa

Abstract

This paper investigates the growth effects of FDI in Africa for the period 1990–2016 using a dynamically common correlated effect approach for an error-correction model. It uses an analytical classification of African economies, with each being fragile, factor-driven or investment-driven. It also accounts for interaction effects and the problem of cross-sectional dependence that previous studies overlooked. While the long-run effect of FDI on growth is significantly positive in investment- and factor-driven economies, its short-run effect is insignificant in the latter type of economies. The effect of FDI on growth is insignificant in the fragile category both in the short-run and long-run, however.

KEYWORDS:

• FDI
• economic growth
• country classification
• institutions
• political stability
• Africa

JEL Classifications:

• C23
• F21
• F43
• O55

Acknowledgments

I would like to thank Prof Scott Hacker, Prof Alemayehu Geda, Prof Hyunjoo Kim Karlsson, Prof Almas Heshmati, and Prof Pär Sjölander for their useful comments and suggestions. Especially, I am truly grateful for all the support I have received from Scott and Alemayehu. I am also grateful to Swedish International Development Cooperation Agency (Sida) and Jönköping University for financially supporting my doctoral study, which this paper is a product of. Comments by participants in the Friday seminars at Jönköping International Business School (JIBS), Jönköping University in Sweden, East Africa Business and Economic Watch conference in Kigali, and Addis Ababa International Conference on Business and Economics in Addis Ababa were very valuable. Any errors, however, are mine.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 This study uses the analytical African countries classification as fragile, factor driven, and investment driven economies outlined in Geda and Yimer (2015) (see Table A1 in Appendix 1 for the list of countries in each group).

2 Available data shows that the top three sectors in terms of attracting FDI flows to Africa are primary (48%), service (31%) and the manufacturing sector (21%) (UNCTAD 2016).

3 It is also informed by the works of De Mello (1997), Borsworth, Collins, and Reinhart (1999), and Ramirez (2000).

4 UNCTAD (2016) definition of FDI stock is used in this study. UNCTAD (2016) calculates FDI stock using the perpetual inventory method. FDI stock is defined as ‘the value of the share of capital and reserves (including retained profits) attributable to the parent enterprise, plus the net indebtedness of affiliates to the parent enterprises. It is approximated by the accumulated value of past FDI flows’ (UNCTAD 2016).

5 The need for adjusting trade share for population size is motivated by the fact that ‘small countries (in terms of their population size) generally need to trade more with the outside world to provide all available goods for the domestic economy. On the other hand, large countries usually trade less with other nations. Thus, higher trade within the domestic economy should not be taken either as an implication for less degree of competitiveness or as an implication for the less efficiency of it than international trade’ (Neuhaus 2006). Thus, to account for this, the population effect from the trade share should be taken out (see Neuhaus 2006).

6 For instance, using the following panel regression will generate the residuals to be used as the variable on trade share adjusted for population size. Run the regression given as $\text{ln}\text{T}{\text{S}}_{it}={\beta }_1\text{ln}\text{PO}{\text{P}}_{it}+{\epsilon }_{it}$, where $\text{T}{\text{S}}_{it}$ is trade share measured as $[(\text{export + import})/\text{GDP}]\ast 100$ and $\text{PO}{\text{P}}_{it}$ is total population for country $i$at time $t$. The residuals, ${\epsilon }_{it}$, provide the data for the trade share adjusted for population size (see Neuhaus 2006). Note that this variable will be used in regressions without its log transformation (see Neuhaus 2006).

7 The relationship between financial sector development and economic growth has been studied extensively, typically resulting in the conclusion that a well-developed financial system promotes productivity growth and economic growth (Alfaro et al. 2004; Hermes and Lensink 2003). But, it can equally be argued that, especially for African countries, a well-developed financial system may hamper growth depending on whether financial development reduces or increases capital flight. For instance, Geda and Yimer (2016) reported a positive effect of financial sector development on capital flight in Ethiopia. On another study, Geda and Yimer (2015) found the significant negative effect of capital flight on economic growth in Ethiopia. In a FDI-growth study, Akinlo (2004) reported a negative effect of financial development on growth by increasing capital flight in Nigeria.

8 Details on the underlying data sources, the aggregation method, and the interpretation of the indicators, can be found in the WGI methodology paper of Kaufmann, Kraay, and Mastruzzi (2010).

9 The aggregate political and institutional indicator (POLINS) is computed from normalized values of the three indicators (Political stability, regulatory quality, and rule of law). The normalization is done using the following formula: for any variable $x$, $x_{\text{normalized}}=(x-x_{\text{minimum}}/x_{\text{maximum}}-x_{\text{minimum}})$. This is done on a yearly basis for each country in the models. Then a simple sum (assigning equal weights) of the normalized values for the three variables is taken on a yearly basis for each country.

10 The post estimation CD-test proposed by Pesaran (2004) is based on the averages of all pair-wise correlations of the ordinary least squares (OLS) residuals obtained from the individual regressions in the panel data model. This is given by $\text{C}{\text{D}}_p=\sqrt{\frac{2T}{N(N-1)}}\left(\sum _{i=1}^{N-1}\sum _{j=i+1}^N{\hat{\rho }}_{ij}\right)\Rightarrow N(0,1)$, where${\hat{\rho }}_{ij}$ is the estimate of the pair-wise correlation of the estimated residuals given as ${\hat{\rho }}_{ij}=\frac{{\sum }_{t=1}^T{\hat{\epsilon }}_{it}{\hat{\epsilon }}_{jt}}{{\left({\sum }_{t=1}^T{\hat{\epsilon }}_{it}\right)}^{1/2}{\left({\sum }_{t=1}^T{\hat{\epsilon }}_{jt}\right)}^{1/2}}$ (see Pesaran 2006, 2007). It should be noted that equation (13) is applicable for the cases when the panel is a balanced one. For unbalanced panel a different test equation should be used (see Pesaran 2006, 2007).

11 The models are first estimated using the traditional panel cointegration approach. Then residual-based post-estimation Pesaran (2004) CD-test is undertaken. The test indicated the presence of cross-section dependence problem in the residuals. Thus, to account for the cross-section dependence, the models are re-estimated using dynamically common correlated effect approach.

12 Care should be taken in interpreting the overall effect of FDI in Model 2. The overall effect of an increase in FDI (the total effect that comes from the non-interactive term of FDI and the interactive terms) can be calculated by taking the partial derivative of equation (9), which is $\frac{\mathrm{\partial }\text{log}{Y}_{it}}{\mathrm{\partial }\text{log}\text{FD}{\text{I}}_{it}}=d_3+d_5\text{log}{H}_{it}+d_{10}\text{POLIN}{\text{S}}_{it}$. The overall effect of FDI, when the interaction terms are statistically significant, can be calculated by plugging the average value for $\text{log}{H}_{it}$ and $\text{POLIN}{\text{S}}_{it}$ in $\frac{\mathrm{\partial }\text{log}{Y}_{it}}{\mathrm{\partial }\text{log}\text{FD}{\text{I}}_{it}}$. In cases where $d_5$ and $d_{10}$ are statistically insignificant, the overall marginal effect equals $d_3$ (which is also the marginal effect due to the non-interactive term).

13 The large body of literature on growth has mostly found some measure of human capital as a significant determinant of growth (see e.g. Barro 1991; Mankiw, Romer, and Weil 1992). However, some other authors have argued that factor accumulation is not the key to growth in African economies (see e.g. Bils and Klenow 2000; Easterly and Levine 1997).

14 Additional sensitivity analysis was performed on the original specifications of Model 1 and Model 2 by substituting the population size adjusted trade openness variable by trade openness without adjustment for population size. The results of such exercise showed that the results of Table 8 are robust for this modification too (not reported). In addition, the robustness of the results reported in Table 8 were also checked by dropping the commodity price variable as that variable is susceptible for structural breaks (not reported). The general finding is that the coefficients on the FDI and its interaction variables are found to be consistently estimated and the sizes continue to be largely the same. All the other control variables have their expected sign in this alternative specification too.

15 In equation (22), the interaction terms are left out of the test equation as they contain FDI in each of the interaction terms.